Geographic significance of delivery zones in an e-commerce grocery delivery strategy


Material Information

Geographic significance of delivery zones in an e-commerce grocery delivery strategy
Physical Description:
viii, 215 leaves : ill. ; 29 cm.
Herrel, Keith
Publication Date:


Subjects / Keywords:
Geography thesis, Ph. D   ( lcsh )
Dissertations, Academic -- Geography -- UF   ( lcsh )
bibliography   ( marcgt )
theses   ( marcgt )
non-fiction   ( marcgt )


Thesis (Ph. D.)--University of Florida, 2006.
Includes bibliographical references.
Statement of Responsibility:
by Keith Herrel.
General Note:
General Note:

Record Information

Source Institution:
University of Florida
Rights Management:
All applicable rights reserved by the source institution and holding location.
Resource Identifier:
aleph - 003594259
System ID:

Full Text








The completion of this dissertation would not have been possible if it were not for

the loving devotion of my wife, Tsuneko Herrel (also known as "Jaato") and the

enduring love and support of my mother Jean Herrel, who passed away before she

could see me graduate with my Ph.D. I know, however, that she will see me graduate

as she looks down from heaven. My three darling children, Schawn (also known as

"Chub"), Lina Lee (also know as "Screaming Meemie" or "Angel Princess"), and

Aaren (also known as "BoupToYouBoy"), have given me the will, perseverance,

fortitude, and just plain heartwarming desire to achieve in life, both academically and




ACKNOWLEDGMENTS............................................................ ........1..

LIST OF FIGURES........................ ....... ........................................v............. v

ABSTRACT .................................................. viii

CONCEPTS AND METHODOLOGY.......................................................

Purpose................................................. ...... ..................
C concepts ................... ........... ................... ...................3
Methodology................................ ...... .... ................4

LITERATURE REVIEW.....................................................................8

Online Grocers' Failures, Difficulties, and Possible Solutions...................9
Spatial Decision Support Systems................................................. 18
Online Retailer Geographic Issues ..................................................25
Grocer Web Site Design Issues......................................................29
A artificial Intelligence........... ........................................ ................ 30
Online Grocers' Store Locations.....................................................45
Geographic Information Systems for Traffic Flow Simulation...................48
Geographic Information Systems for Vehicle Routing and Logistics.............51
Networks ............................................ ........ .............56
Data Visualization and Data Representation........................................59
Geographic Data and Surface Representation...................................66
Geographic Information Systems in Business......................................86
Enabling Geographic Information Systems Technologies.........................92
Other Supporting Literature........................................................ 107

SIMULATION ELUCIDATION........................................................... 122

AVERAGE, AND MEDIUM PROFIT...................................................... 178

Exceptional Profit.................................................................... 178
Low Profit ................................ ... ...................................... 179
Medium Profit .................................................................... 179

C O N C LU SIO N ............................................................................. .. 186

SOURCES CITED...........................................................................199

BIOGRAPHICAL SKETCH................................................................ .. 215



1 Hexagonal layout......................... ...........................................

2 Duration of delivery in days.................................................... 146

3 Profit options settings............................................................. 146

4 Minimum dollar delivery.......................................................147

5 Mandatory profit.................................................................. 147

6 Delivery charge per concentric region........................................ 148

7 Delivery charge per region interface........................................... 148

8 Delivery charge per region pop up form.......................................149

9 Simulation options form......................................................... 149

10 Number of orders error catching.............................................. 150

11 Number of deliveries drop-down list........................................150

12 Duration of deliveries error catching ..........................................151

13 Duration of deliveries drop-down list........................................ 151

14 Region and legend color coordination..................................... 152

15 Report dashboard for the simulation run on April 13......................152

16 Simulation main map result with profit settings..............................153

17 Pause function.................................................................. 153

18 Profit options form........................................................... 154

19 Allow dispatch alerts radio button............................................. 154


20 Truck dispatch notification.................................................. 155

21 Simulation main map result showing truck usage............................155

22 Report dashboard for the simulation run on April 15......................156

23 Population report ............................................................ 156

24 Profitability chart................................................................ 157

25 Total mandatory profit per simulation report.................................158

26 Residence report.......... .......................................... ........ 159

27 Settings report.................................................................... 160

28 D delivery report .................................................................. 161

29 Simulation options form.......................... ....... ................... 161

30 Simulation main map result showing three simulations.................... 162

31 Simulation charts report....................................................... 163

32 Simulation charts report showing second and third simulation ...........164

33 Settings report showing three simulations......................................165

34 Total mandatory profit per simulation report showing three simulations.. 165

35 Residence report showing orders from the second and third simulations.. 166

36 Mandatory profit per concentric region on profit options form........... 166

37 Mandatory profit per concentric region on settings report.................. 167

38 Submit Delivery Charge button.................................... ........... 168

39 Delivery charge per region interface........................................ 168

40 Delivery charge per region form............................................169

41 Region A3 showing the delivery charge....................................... 169


42 Profitability chart showing profit for region A3............................170

43 Mandatory profit per simulation report showing profit for region A3.... 170

44 Settings used with Intra-Regional Drive Time function...................171

45 Intra-regional drive time interface.......................................... 171

46 Intra-regional drive time pop-up form........................................172

47 Settings report showing 15 minute intra-regional drive time............ 172

48 Delivery report showing increased intra-regional drive times.............173

49 Inter-regional drive time interface................... ........ ......... 173

50 Settings report showing increased inter-regional drive time..............174

51 Delivery report showing increased inter-regional drive times........... 174

52 Overall average population per region.....................................175

53 Population per region interface................................................ 175

54 Main map showing various amounts of orders placed....................... 176

55 Regions showing higher and zero populations............................176

56 Report dashboard showing various populations...........................177

57 Population report showing various populations........................... 177

58 Profit and simulation options for high-profit scenario....................... 180

59 Profitability quotient and profit gauge for high-profit scenario .........181

60 Profit and simulation options for low-profit scenario......................182

61 Profitability quotient and profit gauge for low-profit scenario............. 183

62 Profit and simulation options for medium-profit scenario................ 184

63 Profitability quotient and profit gauge for medium-profit scenario........ 185

Abstract of Dissertation Presented to the Graduate School
of the University of Florida in Partial Fulfillment of the
Requirements for the Degree of Doctor of Philosophy



Keith Carl Herrel

May 2006

Chair: Grant Ian Thrall
Major Department: Geography

This work contains a geographic and logistic tool that will allow grocers who

are contemplating to deliver groceries via e-commerce to better decide whether they

should embark on that initiative. As web-based orders become more commonplace,

this computer application will have the potential to assist both large and small grocers

in the critical decision of starting a delivery service. The causes of failures of grocers

are traced. An operational computer program is demonstrated, with embedded

algorithms that make for a realistic simulation. The computer simulation together with

the user-friendly interface "front end" provide a practical and also scholarly solution to

a complex geographic problem.



The potentially lucrative grocery home-delivery market largely remains untapped

because of the financial risk and operational difficulty of running such a service. Also

recent history is fraught with unsuccessful attempts at delivering groceries. The rapid

bankruptcy of the high-profile Webvan company is possibly the worst debacle of all the

grocery delivery attempts.

This dissertation contains a logistical simulation application to solve one of the most

plaguing problems that some grocers have, which is the problem of deciding to initiate a

home delivery service or not. The primary purpose of the simulation is to reduce the

financial risk involved with rolling out an e-grocery initiative.

This means that grocers can take a quantum leap forward in satisfying the annual

multi-billion dollar United States grocery market by using my application. This is also

true for other industrialized countries. How is this possible? First, the great number of

grocer home delivery failures is no secret. This could be resulting in many grocers

hesitating to start a delivery initiative. Second, there could be a notable benefit to the

people in much of the industrialized world if grocers can successfully deliver to time-

strapped customers. Traffic jams and fuel consumption could decrease if people do not

use their cars to shop after work. Third, a grocer can decrease the number of employees,

such as cash register attendants and stock persons needed in the brick-and-mortar store

because customers who would otherwise go to the store will not do so if the groceries are

home delivered. Fourth, the infirm and elderly will benefit because the competition to

satisfy people who accept home deliveries could very well drive down delivery prices

that otherwise would be out of the economic means of these people. This list of benefits

is not exclusive, but should give justification to the claim that grocery home delivery is a

benefit to not only the grocer who could make a greater profit or acquire more customers,

but also to society in general.

No computer simulation application as presented here has ever been presented in a

public forum, including academic journal articles or news media. Grocers who are

contemplating deliveries will find this profit estimating simulation application easy to use,

powerful in its functionality, and versatile to allow its use in the actual delivery

operational operations Actual customer addresses can be input into the application to

generate simulated deliveries to loyal and prospective customers. Actual times that are

required to deliver to each region can be input before the simulation begins. The output of

the delivery data is shown is easy-to-read automatically generated reports. Round trip

truck times that include the average time required to deliver within each region can be

input separately into the application to give an accurate report of how long it takes a truck

to leave the store, deliver to the households, and return to the store. Also, the number of

trucks required to deliver to the population of all the regions can either be specified or


Because rarely is it possible to give a clear-cut yes or no answer about whether a

business venture should be undertaken or not, the conditions where a online e-grocer (or

a grocer who is taking orders by telephone) will reach a predetermined amount of profit

can be determined by running this application. In other words, questions like, "by what

pathways can profitability be achieved?" and "what values do you need in combination to

reach a predetermined profit?" must be answered before a grocer should begin delivering


Because profitability is naturally the goal of any grocer, this application gives an

indication to the grocer about what clusters or sets of attributes that will work to achieve

the required level of profit. Each simulation that is run can portray a different scenario of

what is needed to achieve profitability.


In terms of this logistic application the term "Mandatory Profit" is used. The

Mandatory Profit is the profit that must be attained on each run to a region. A region,

from a grocer's standpoint, is an area that has customers with a certain profile, or takes a

certain amount of time of travel. A grocer must determine if deliveries to a region will be

profitable or not. If the grocer decides that each dispatch of a delivery truck (or "run") to

a region must result that revenues must exceed operating costs by $X, then all the grocer

needs to do is input the $X profit figure into the Mandatory Profit dialogue box of this

simulation. The dialogue box is a data input box on the logistic computerized interface

that I have developed. Additionally, the Mandatory Profit can be specified according to

different regions, allowing myriad possibilities for different profitability attainment


The expected profit margin can be input into the application. This percentage is

calculated in conjunction with the Mandatory Profit and the Delivery Charge to ascertain

whether the initiative can be profitable. The Profit Margin can be varied from 0.1% to 3%

of the value of the groceries delivered.

The Delivery Charge is another variable that can be input. The grocer can specify

any amount of delivery charge, from zero to as high as he or she would like. This

delivery charge is added to the calculation of each order generated and is used to

calculate the Mandatory Profit.

Another variable that can be added to each simulation scenario is the Minimum

Dollar Amount. If a grocer feels that it is not worth his while to deliver anything less than

$20 of groceries, the merchant could input that option into the simulation.

Succinctly stated, an objective of the simulation is the determination of the

conditions where an online e-grocer (or a grocer who is taking orders by a telephone) will

reach a predetermined amount of profitability.

When a grocer uses the logistical application, he or she will be in a significantly

better position to decide whether to initiate a home delivery service, or at least make a

more calculated decision about whether more costly analyses are justified or not.

Although the logistical simulation application contains sophisticated algorithms, and

performs computational tasks, a main purpose of mine is to create an application that is

intuitive and easy to use. This I have done.


This presents a spatial decision support system for the delivery of groceries from

fixed "brick and mortar" locations to customers' homes. The spatial decision support

system (SDSS) will be demonstrated with a simulation. The simulation will perform the

following functions.

* Determine the mix of variables that are necessary for the online grocer to generate a
profit while delivering groceries.

* Tell the total profit that can be generated through deliveries, given a certain profit

* Determine the amount of trucks necessary to deliver to the customers.

* Show what regions normally reach a profitability margin, and at what times they do

* Provide very good, medium, and poor profitability scenarios using the simulation
application I developed.

The simulation includes identification of the algorithms required to implement the

SDSS. When combined with reasonable data inputs, the information flow and algorithms

will allow for the identification of which geographic settings the SDSS can be

successfully deployed, and which are unlikely to be met with success. Even if no

geographic settings are indicated as being likely candidates for an actual SDSS of this

type, the simulation will be a success as it will indicate to investors that profit margins in

the contemporary grocery industry are insufficient to take advantage of the technology

and online ordering and store distribution of groceries to customers.

It is assumed that the grocery store has surveyed its current customer demand base to

determine the amount of groceries the customers will order from home each week. In

other words, the grocery store management is assumed to have reasonable expectations of

market penetration by neighborhood under various pricing schemes.

The geography of the delivery market will be show as follows.

* An urban region will be depicted on a map.

* A pattern that consists of honeycomb regions will be overlain on the map (Figure 1).
* The home store will be located at the center of the honeycomb area. In Figure 1
diagram, the store is located in the center blue hexagon.

* The hexagonal regions will be identified by capital letters. The home store in the
center region will be in region "H" meaning "home".


* Each customer will be associated with a hexagonal region.

* A random number generator will select customers for use in the simulation.

* The simulation will compress time.

* The contiguous hexagonal pattern will continue outward from the store, in a radial
manner (Figure 1).


Figure 1. Hexagonal layout


Although the main purpose of this dissertation is to emphasize the geographic

significance of delivery zones in an e-commerce enabled grocery delivery strategy, the

work necessarily encompasses more than geography and GIS (geographic information

systems), and this multi-disciplinary approach is reflected in the literature review. An

important part of this section is "Grocer web Site Design Issues." This is because,

although the logistic application that I developed is run through Microsoft ExcelTM, the

application can be used as a web service, which basically means it can be used by

computers at remote sites for a fee. The section about artificial intelligence is included to

better acquaint the reader with potentially important grocer decision-making tools such as

intelligent agents that could traverse networks and read mobile computers in delivery

trucks. The Literature Review would be incomplete without the incorporation of the two

sections that include GIS for traffic flow simulation, vehicle routing, and logistics. These

concepts are the important technologies for grocers who want to optimize their trucks'

routes, which can result in more satisfied customers because of timely deliveries, and less

fuel and manpower expenditures. The manner which polygons, points, and lines are

represented on a GIS along with the way data is depicted near these geometric figures

should be important to grocers who deliver through a GIS interface so as not to include

superfluous data, or omit important data to delivery, operational, and managerial

personnel. A review of the literature about these concepts is also included in this section.

Online Grocer's Failures, Difficulties and Possible Solutions

There is a lot of supporting and relevant literature that, when combined, can provide

the building blocks for the utilization of delivery zones in geographic enterprise

optimization systems (GEOS) for online grocers. After reviewing the below-described

literature, the solution proposed in this dissertation for online grocer deliveries is not only

necessary, but also feasible.

In 2000, at the minimum seven pure online grocers were operating in the United

States. By the end of that year the only two that remained were Webvan and Peapod. In

an ominous statement, Ring and Tigert (2001) said, "no company has ever made a profit

in this (pure online grocery) business."

Perhaps the most visible of all grocery delivery companies' demises is that of

Webvan, the multi-billion dollar venture that started optimistically, but ended soon after

it began. Much literature exists concerning the quick bankruptcy of this and other online

grocery delivery companies (Partch, 1999; Saccomano, 1999b; Tweney, 1999; Briody,

2000; Bubny, 2000; Sanborn, 2000; Evans, 2001; Heun, 2001; Partch, 2001; Ring, 2001).

The founder of Borders Books, Louis Borders, created Webvan in 1999 to deliver

groceries and other household items to homes within 30 minutes of receiving the order

(Tweney, 1999). George Shaheen, a managing partner of Anderson Consulting, whose

father was in the grocery business, was the CEO in charge when Webvan went bankrupt

(Partch, 2001). There was much hope and skepticism when $11 billion was raised in

Webvan's 1999 initial public offering (Evans, 2001). The hope was that Webvan,

optimistically called the spot where "the internet meets your doorstep," could capture a

portion of the annual $650 billion United States online grocery, prepared food, and

drugstore market (Sanborn, 2000). Of this amount, $450 billion of food is purchased

annually in the United States (Ring, 2001).

It was a grand scheme that depended on a network of large and expensive automated

warehouse distribution centers as springboards for the delivery vans. Partch (1999) wrote

in "Home Delivery, Another Problem Solved," that a single distribution center would

have the capacity to dispatch as many groceries as 20 supermarkets. The distribution

centers were approximately 330,000 square feet each. The first distribution center was

located in Oakland, California; and the company planned to build another 25 similar

centers in other cities (Ring, 2001).

A portion of Webvan's plan did have a customer allure to it. The fledgling company

offered 24-hour free delivery (with orders more than $50) with prices up to 10% cheaper

than brick-and-mortar grocers (Partch, 1999).

Besides the money raised from the IPO, deep-pocketed investors like Knight-Ridder,

CBS, and Softbank sank money into the company (Saccomano, 1999b). But, contrary to

stockholder and other investor optimism, literature shows that many people were also

pessimistic about Webvan's viability. Dan Rabinowitz of Peapod Inc., another online

grocer, called Webvan's business model "a lot of smoke and mirrors" (Saccomano,

1999b). In the same article, grocery consultant Cristopher Hoyt shows his doubt by

saying that Webvan would not be able to solve the perishable food distribution problem,

and that even if they can deliver the dry foods efficiently, the grocery market for dry

foods is being supplanted by cheaper super centers.

Webvan's investors and managers would have benefited from Michael Brown's

commentary on automated grocery distribution centers. Brown said in 1993, "the best

strategy at present may be to postpone investment in a centralized warehouse with high

levels of automation, correspondingly high capital costs, and a lengthy payback period"

(Browne, 1993).

Miles Cook, vice president of consulting firm Bain and Company, said, "online

grocery is a murderously difficult market," (Heun, 2001). Evidently Cook, Rabinowitz,

Hoyt, and other skeptics were right because only 20 months after its inception, Webvan

fired 2,000 employees and filed for bankruptcy (Evans, 2001).

It seems that the less-than-optimal business model employed by Webvan is not the

only factor to blame for that company's demise. Paul Bubny (2000) explains that

customer sentiment regarding online purchases is an inhibiting factor. His

PricewaterhouseCoopers' study revealed that 21% of internet users would never purchase

groceries online. This says nothing of the percentage of current non-internet users that

will not buy groceries online in the future when they begin to use the internet. Bubny

(2000) calls grocery shopping a "care-giving function," and says consumers are not ready

to allow a "detached process such as online shopping" to supply their food. Further, a

survey conducted by NPD group revealed that a mere 2% of consumers would buy

groceries exclusively from an online site.

The same article, however, states that, with proper incentives, customers can be

convinced to buy groceries online. Respondents to the PricewaterhouseCoopers' survey

stated that price is their main concern when buying online groceries, which includes free

delivery. This implies that if other factors such as freshness and rapid delivery are

fulfilled, consumers may be persuaded to buy groceries online if the price is right.

Mainstream grocers are not turning a blind eye to online grocers. In fact, the threat

posed to large grocers' markets by smaller online grocers is real. Alan Mitchell says that

even a 5 to 10% rise in home grocery shopping could "severely dent" a superstore's

underlying profitability (Mitchell, 1999). This statement can be especially portentous to

brick-and-mortar grocers if they consider that overall online purchases are estimated to

keep on increasing (Killgren, 1999). Evidently with this in mind, large United States

grocers like Albertsons and Safeway are initiating brick and click grocery business

models (Briody, 2000). Safeway and Royal Ahold, a Dutch conglomerate, both formed

joint alliances with pure online grocers. By joining with foundering Peabody, Royal

Ahold saved Peabody from probable bankruptcy. Safeway invested $30 million in Not long after built two fulfillment centers in the

south-central United States, the company closed them down. Safeway subsequently

decided that filling online orders from its store shelves was a better model and began

doing so (Heun, 2001).

Albertsons is moving forward with a purely internal brick-and-click delivery

initiative. Possibly as a result of Webvan's closure, Albertsons' online sales in the Seattle

area increased 300% after Webvan stopped operating. Like Safeway, Albertsons fills

orders from its stores' shelves. Safeway is also implementing a fleet management system

to better route their trucks and more closely monitor driver and vehicle performance

(Barnes, 2002).

Ring and Tigert (Tigert, 2001) enumerate six groups of dissuasive factors

contributing to customer skepticism toward online grocery purchases. According to them,

place, the first factor, is important because customers like to be in a place where they can

touch and smell the merchandise. The next group, product, says that online grocers may

not have enough stock keeping units (SKU) that the customer desires. An SKU is

essentially an individual type of product. Third, value, is "more bang for the buck,"

which is often sacrificed by relatively high delivery fees. Service comes next. When a

customer visits a store, there are hopefully more knowledgeable and courteous employees

present than a single delivery truck driver. Communication, the fifth factor, comes in the

form of advertisement or notification to the customer. Ring and Tigert (2001) do not

expound on how this group is an inhibiting factor for online grocers. The sixth group,

technical problems, appears in the form of wrong orders or late deliveries, which can ruin

online grocers.

Another problem, as told by McKinnon and Tallam (2003), is the security of

delivering to homes, especially when the home is unoccupied. Groceries can be delivered

four ways to unoccupied residences. 1. The driver can enter the house. 2. A drop box is

available. 3. Another customer-designated location is available. 4. The goods are left at a

separate local agency, which delivers the groceries when the customer is home.

The article does say that overall, unattended delivery of groceries is relatively

secure when compared to the unattended delivery of more valuable items. Of course, the

worth of the goods matter little if the cause of the security breach is the delivery driver

having access to the interior of the home. The article does say that although the

unattended groceries may not be stolen, the fact that they are left visible implies that the

homeowner is not present which can invite burglary into the premises.

McKinnon and Tallams' (2003) security implications are relevant because a

significant amount of the prospective home grocery delivery market may not by online

because of the security implications. In the same article, the authors list different ways

groceries could be left at an unattended home. Those ways are a fixed interior box with

three compartments of various temperatures with a keypad-lockable door; a fixed

external box connected to an external wall; a mobile box delivered to the customer and

connected to a cable to the house; and communal collection boxes. The communal boxes

can be of various types. Primarily they are in a central place like a popular parking lot or

near a train station. Communal boxes can have numerous compartments, each with its

own lock and changeable code to open the lock (McKinnon and Tallam, 2003). When

deciding on a grocery delivery model, consideration should be given to each of these

drop-off strategies to decide which would probably work best given the catchment

neighborhood, type of customer lifestyle segment profiles (LSP) catered to, and logistical

strategy used by the online grocer. Of the LSP databases used for this type of analysis,

MOSAIC, ACORN, and NDL's Lifestyle database are three of the most highly regarded

(O'Malley, 1997).

The type of drop-off method is not unrelated to the "skill" the drivers have in

delivering the groceries. Skill here relates to the ability of the drivers to learn routes on

the fly. If an online grocer delivers to individual residences, especially when guaranteeing

delivery within an hour, it becomes highly unlikely that any two routes taken within, say,

one or two days, will be exactly the same. Michael Haughton expounds upon the

relationship of uncertain daily deliveries with the drivers' ability to learn new routes on

the fly (Haughton, 2002).

Haughton (2002) states that fluctuating demand can result in inefficient delivery and

dissatisfied customers. He refers to a "learning burden," and proposes a correlation metric

to assess the familiarity of a driver with certain rectilinear delivery routes. Because my

research is focused on solving the problem of individual residence delivery, which

presupposes daily fluctuating demands, Haughton's (2002) literature is relevant to this


Cross and Neil (2000) call online grocery shopping a "booming global phenomenon."

Obviously this statement does not seem to include the plights of the failed United States

online grocers. Evidently, these authors are referring to online grocers in Great Britain,

Scandinavia, and Iceland that are not doing as badly as those in the United States.

Optimistically, however, Cross and Neil tell how convenience is a driving factor toward

the success of online grocers. They mention how technologically savvy customers are the

most profitable ones because they are more inclined to shop online. The article also says

how employers can use online grocers as a fringe benefit to their employees. The

employer can pay for the delivery service to the office where the employees pick up their

groceries when they leave work.

Another factor that could provide inertia toward successful online grocer businesses

is UCCNet. This is a subsidiary of the Uniform Code Council, which is the organization

that developed the universal product code. UCCNet allows e-commerce companies to

register their supply-chain related data. UCCNet is an internet trading network that could

benefit buyers and sellers of food, beverages, and other items (Violono, 1999). Because

the network aids collaborative planning, forecasting, and replenishment, smaller online

grocers who use UCCNet should be better able to procure needed stock in times of

unexpected spikes in online purchases. More literature pertaining to UCCNet is reviewed


Catering to the growing United States Hispanic market is an online grocer called

LatinGrocer. Saccomano (1999a) states that by 2009, 40% of the United States

population will be Hispanic, creating a very large grocery market. LatinGrocer's unique

point is that its services are also offered in Puerto Rico.

An effort to identify the success factors in e-grocery home delivery was made by

Punakivi and Saranen (2001) using simulation. The article supports the intent of this

dissertation by explaining that the lack of a proper logistical home delivery infrastructure

is the main deterrent to a successful online grocer.

Using Finland as an example, Punakivi and Saranen (2001) concluded that home

delivery can be as much as 43% cheaper for customers when compared to the customer

driving to the supermarket. Punakivi and Saranen (2001) write that the cost to deliver

groceries will be less when the delivery time window is larger and customers live in

densely populated areas. These are intuitive factors. What might not be so intuitive,

however, is that the number of needed delivery vehicles increases by 250% when a

manned delivery model (when the customer must be home) is used instead of an

unmanned model which uses drop boxes. Tanskanen and Punakivi (2002) write that the

only factor more costly than the deliveries in a manned home delivery model is the actual

picking and packing of the groceries One reason for the higher manned delivery costs is

that up to 60% of the packages might be undeliverable because of people not being home.

In Great Britain, some main grocers are using unattended delivery methods. In Finland,

grocers are using shared multi-temperature drop boxes for unattended delivery. A factor

that affects the costs and profitability of the shared reception box model is the number of

separate compartments in each box (Punakivi and Tanskanen, 2002).

It is important to note that online grocers might have to operate somewhat contrarily

to flexible production techniques. This is because although flexible production uses

strategies such as just in time deliveries (Browne, 1993), online grocers who precisely

forecast a significant proportion of their deliveries should consider pre-packing those

groceries to save time during busier order packing times. Shorter delivery time windows

notwithstanding, the ability for customers to choose what time the groceries are delivered

is another factor that increases delivery costs. In other words, given a 40-minute delivery

time window, the model that allows customers to choose what delivery time window is

the more expensive delivery method (as opposed to the grocer deciding what delivery

time window to dispatch the truck) (Punakivi and Tanskanen, 2002).

Speaking of delivery costs, online grocers must consider the tradeoff of shipping

from fewer sites, which increase delivery costs (Browne, 1993), or dispatching the

groceries from more outlets, which could increase the packing costs. Browne writes that

products with relatively low value densities should be distributed by local delivery. This

would apply to most groceries.

Gorr, Johnson, and Roehrig (2001) talk about the difficulty of home delivery

services and how GIS can help with delivery solutions. They describe the constraints

imposed upon delivering hot meals to residents home. Their research is applicable to this

dissertation because online grocers may offer hot delicatessen foods delivered to homes.

Indeed, it might be this hot (and cold) food service that helps the online grocer to gain

market share. But, as this particular article states, the time window that exists to deliver

these foods is greatly decreased because of the necessity to maintain very hot or very cold

temperatures. Gorr, Johnson, and Roehrig (2001) write that 45 minutes is the maximum

allowable delivery time for these deliveries.

Gorr, Johnson, and Roehrig (2001) also discuss the feasibility of using a spatial

decision support system (SDSS) for delivering hot meals. The combination of GIS,

algorithms, and managerial judgment are parts of SDSS. A particular problem the authors

address is the situation created when customers change addresses. The SDSS algorithms

should be robust enough to handle this type of change (Gorr, Johnson, and Roehrig,


One approach that online grocers might pursue when considering their overall

strategic plans to deliver groceries is to look at the composite customer equity that the

delivery regions encompass. Customer equity, which is often used as an augmentation to

brand equity, is the sum of the lifetime values of all of the customers a business has

(Niraj, 2001). If, when contemplating a delivery plan, the customer equity calculation

does not look as if it could be increased or even sustainable, it might be advisable to

rethink the grocery delivery plan. Some aspects to look at when determining the

sustainability of customers are complexity factors and efficiency factors of those

customers (Niraj, 2001). Succinctly stated, these two factors are antithetical forces.

Whereas complexity factors, such as difficulty fulfilling a grocery order, decrease a

customer's lifetime value, efficiency factors, which make a delivery to a particular

address easier, can help increase the lifetime value of the customer.

Spatial Decision Support Systems

Research about using GIS as an integral part of SDSS began years before the above-

noted article by Gorr, Johnson, and Roehrig (2001). Crossland et al. (2001) studied the

effectiveness of using GIS in SDSS to see if the time used to solve problems could be

significantly decreased. The article reinforces the important point that most business data

have "one or more spatial components." Three of these components, as stated in the

article, are relevant to online grocer business models customer addresses, delivery

vehicle locations, and site selection. Calling GIS a "highly evolved technical toolbox,"

the article, says that GIS should not only be used to find solutions to problems, but also to

guide the users to problems that might have been unforeseen before the adoption of a

GIS-enabled SDSS (Crossland, 1995). The worth of decision support systems (DSS) in

general is stated by (Vlachopoulou et al., 2001). They say that DSS "couples the

intellectual resources of individuals with the capabilities of the computer to improve the

quality of decisions and to support managerial judgment" (Vlachopoulou et al., 2001).

The value of spatial aids to managerial judgment should not be underestimated. Thrall

(1995) endorses the value of using GIS to optimize judgment in managerial decision

making. Thrall writes that, "...using GIS to improve judgment is a crucial stage in

reasoning with GIS because it links GIS to the market economy." Thrall expounds on his

GIS-facilitated judgment maxim by describing the value of business geography. Thrall

said, "Business geography integrates geographic analysis, reasoning, and technology for

the improvement of the business judgmental decision. Without the demonstrated ability

to improve the business decision, there is no business geography. This differentiates

business geography from the traditional descriptive or explanatory objective of economic

and urban geography" (Thrall, 2002).

Gorr et al. (1995) tell how a GIS-enabled SDSS can allow decision makers, or their

assistants, to use databases in real time. This is of course important for online grocers that

employ metamorphic delivery zones because, as this dissertation will show, it is when a

delivery zone reaches a certain profitability threshold that the truck should be dispatched

to that zone. A decision maker, or his or her assistant, must be able to make the real-time

decision to send the truck. Gorr et al. (1995) show that spatially related decisions can be

made faster and with less error when SDSS are used. Supporting the assertion that

deliveries should be made only after a certain profitability level has been reached is Niraj

et al.'s (2001) research that says a company should "eliminate transactions that do not

add value." If value to a company (considered here as profitability) is not created by a

delivery to a zone, then perhaps delivery to that zone should be deferred to a later time

when the value of that zone increases. Care should be taken that this strategy does not

result in dropping customers who might be unprofitable in the short run, but might

become valuable customers in the foreseeable future (Niraj et al., 2001).

A SDSS incorporates GIS with a database management system (DBMS) and

peripheral software tools that allow interaction with the analysts and other users

(Tarantilis et al., 2002). Tarantilis et al. call SDSS a "new scientific area of information

systems applications." Possibly, the relative newness of SDSS is in part the reason that

little if any literature exists about the utilization of this technology to successfully solve

the problem of delivering groceries that are purchased online. I will demonstrate in this

dissertation that spatial decision support systems can be the difference between the

success and failure for any online grocer of significant size.

The importance of routing information systems is underscored by Browne as he says

that the end of the era of "cheap freight" transport is here, where reliability used to be

attained without modern information systems (Browne, 1993).

Tarantilis and Kiranoudis (2002) use SDSS to solve a vehicle routing problem (VRP)

using the backtracking adaptive threshold accepting (BATA) model. The constraints

posed in these authors' research mirror those that would be inherent in a typical grocery

delivery situation. They are minimize the distance traveled, do not exceed the capacity of

a vehicle, and, one vehicle delivers each customer's order. Although the article does not

specifically mention web services, it does say that reusable software components are

important parts of a SDSS. Because, more recently, the importance and viability of web

services (for example, services created by the .Net platform) have gained mainstream

exposure, the plausibility of creating a component-based SDSS is greater than it is

without using web services.

Tarantilis and Kiranoudis (2002) give a detailed and graphical explanation of all the

components of a SDSS. ArcView is the GIS used. BATA, in conjunction with the

Dijkstra algorithm, determines which vehicle should deliver to which customers and the

order that it should do so. BATA is a "metaheuristic" method, meaning that it relies on

more than one heuristic to arrive at a solution. The article states that the SDSS is used by

Athens, Greece taxi companies, newspaper companies, and for meal delivery. Benefits

attained by using the SDSS include the ability to make semi-structured or unstructured

vehicle routing decisions, transport cost reduction, and better control of distribution

(Tarantilis and Kiranoudis, 2002). Also, when routing decisions are made using a SDSS,

the decisions made might be more intuitive if some sort of multimedia capabilities exist

within the system. Cartwright and Hunter explain how decision support systems can be

enhanced by the incorporation of multimedia. These authors say, "users of geographic

information need far more than the printed map, and they are likely to need rapidly-

produced and well designed spatial displays combining a number of different types of

data" (Cartwright, 2001).

As pertains to this dissertation, the key words expressed by Cartwright and Hunter

are "rapidly-produced.. spatial displays." Another notion proposed by these authors is

that although the incorporation of multimedia into GIS is very valuable, care must be

taken to not "overwhelm" the users. Ogao and Kraak (2002) agree with this by saying

that animation can distract the users if it is overused. These cautions should be adhered to

by online grocers who incorporate delivery zones into their geographic enterprise

optimization systems.

Ralston, Tharakan, and Liu (1994) espoused the virtues of SDSS before GIS was

considered a mainstream technology. The authors said that GIS must be merged with

spatial analysis tools for GIS to be used to its fullest potential. Agreeably, GIS alone,

without being augmented with tools such as simulation or artificial intelligence, is a

powerful technology but remains a mere shell of what it could be when used with these

tools. In Ralston et al.'s article it is said that transportation networks are comprised of

physical and logical links. Physical links are the actual roads, railways, or inland water

routes. The Logical links are points along the physical links for deliveries, pick ups, and

inter-modal transfers. Physical links of the same type can be divided into classes

depending on factors such as cost, capacity, and speed. Paths between a source and

destination always have both physical and logical links (Ralston, 1994).

Nasirin and Birks expound upon the types of SDSS used by retailers in Great Britain.

Knowing customer proximity to the store and those customers' purchasing behaviors are

two common functions performed when SDSS is used by retailers. The literature states

that one reason retailers are adopting GIS-enabled decision support systems is because of

the increasing availability and portability of spatial data which often exists on compact

disks. Despite this growing adoption of GIS by retailers, there has been little research

about the methodologies that retailers employ when using GIS to make decisions, even

though successful usage of GIS by retailers involves "enormous numbers" of factors

(Nasirin, 2003). An important point made in this article is that GIS should be used by

retailers to determine who the competition is and where they are located.

Keenan (1998) says that GIS is a technology that can "take advantage of other

technologies," when used to make routing decisions. Keenan also states that straight-line

distance is insufficient for routing calculations, and that time traveled is a better indicator

of the route. Using a decision support system (DSS) is important when ascertaining the

best route because sometimes the best route may actually look circuitous (Keenan, 1998).

According to Keenan (1998), this is where the "soft" constraints, such as decision-

makers' judgment, apply. The importance of the decision-makers' input is why solutions

to routing problems require the use of DSS more and expert systems less (Keenan, 1998).

Many times data and information applicable to the routing problem may already be in the

corporate management information system (MIS). Often information such as customer

demand and vehicle size is business-specific, thereby requiring a customized link to the

data from the GIS (Keenan, 1998). Because spatial data is getting less inexpensive to

purchase, the availability of comprehensive highway databases will become more readily

available Also, according to Keenan, highway data within a GIS when used in with

sophisticated management science routing algorithms can result in useful SDSS.

Genetic algorithms, although not strictly a spatial decision support system

component, are intriguing problem solvers that can help with difficult and time-

consuming geographic problems. Clarke (1998) believes that Genetic algorithms (GA)

utilization will soon become more prominent in GIS. Genetic algorithms use a

"Darwinian" method to find feasible solutions for problems by forcing solutions to

"evolve" by allowing only the "strongest" genre of variables to "survive" repeated

iterations through the problem. Chromosomes, consisting of genes, which themselves

contain alleles, "evolves the population of solutions by repeated selection and mating"

(Van Dijk et al., 2002). Van Dijk et al. demonstrate the use of GA to solve a map labeling

problem. Some of the dynamics of the GA problem solving method are Selection, which

means better parents produce fit children. By using tournament selection iteratively

choosing parents until n sets of two parents are chosen a set of parents is only chosen if

their children are fit. Reproduction, which means each set of parents produce two

children. Crossover refers to a certain amount of genes from each parent which are

imbued into the children. Replacement means some less fit children are exterminated to

make room for other stronger offspring. Finally Termination occurs when, at some point,

the population "converges," or become as close to optimal as possible.

Obviously, this is a very abstract explanation of GA. As a concrete example, Van

Dijk et al. show how map labels are optimally placed using GA while specifying the

direction of the label in respect to the city, degree of congestion of labels, and how large

the labels are in proportion to city population. Online grocers should investigate the

utilization of GA in their routing and customer selection optimization strategies. This is

because if delivery time is considered a scarce resource, the profitability of customers

becomes a paramount issue when deciding routes. Customers can be chosen using GA by

"natural selection" using profitability as one of the gene traits, as described above.

Online Retailer Geographic Issues

"Microscale" competitor analysis is possible with GIS (Sadahiro, 2001). Sadahiro

(2001) uses this approach as he looks at retail location in Japan and discovers that stores

are usually clustered around railroad stations.

Clustering, although often used to denote a sort of condensation of entities within a

region, is also a group of techniques use to analyze complex geographic data to ascertain

if spatial relations exist or not. Pattern spotting and data mining are two ways to

determine the presence or absence of clustering (Murray, 1999). While looking at two

types of clustering problems, the median clustering problem and the central point

clustering problem, Murray discovered that the median clustering problem requires less

computational power to solve than the central point clustering problem.

An aspect of retail clustering that pure online grocers (those without a brick-and-

mortar store) should consider is that they might not be able to capitalize on the important

aspect of being within customers' comparison shopping region. Sadahiro (2001)

recognizes this important factor and develops a probability density function, which

quantifiably tells the degree of agglomeration of retail stores of the same classification.

One reason stores in the same classification locate near each other is to provide customers

with the opportunity to shop comparatively (Sadahiro, 2001). Although this is not the

only force that results in retail agglomeration, it should be recognized as a force that

might take customers away from online grocery sites. Guy describes comparison

shopping as one factor to consider when investigating the grouping of retail stores. He

says that comparison shopping has an "enjoyable" aspect to it (Guy, 1998). Can

comparing prices of competing online grocers online be "enjoyable?" This question is

worth considering by online grocers.

Baker addresses another problem that is encountered by online retailers. He says that

online retailers do not adequately consider central places when selling online. This is

because the fast speed of most internet communications lends itself to the

underestimation of the importance of spatiality in cyberspace. He calls the failure of

retailers to comprehend distance minimization strategies such as the gravity model "a

major barrier to successful marketing and profitability for internet retailing" (Baker,


Recently, consumer activity patterns have been given greater scrutiny by retailers.

Consumer activity patterns could be relevant to online grocers for various reasons. For

instance, activity patterns could help online grocers decide whether to deliver to the home

or use communal drop boxes. Decision tree induction, which is used by artificial

intelligence and statistics, has recently been said to be useful to help model activity-based

models (Arentze, 2003). Albatross models model travel demand using a rule-based

decision tree. Arentze (2003) says that decision tree induction can be useful for modeling

spatiotemporal behavior, such as the behavior exhibited in activity patterns. Activity-

based travel behavior (ABTB) is the term coined for this type of activity monitoring

(Frihida et al., 2002). The concept of "space-time paths" is covered by Frihida et al.

(2002) in their article about ABTB. According to these authors, space-time paths can help

explain, predict, and plan for past and future activity patterns of commuters.

The reason for the attention given to activity patterns is that previously, retailers

looked at discrete trips when calculating the amount purchased by consumers. The

current consumer trend, however, is to engage in multi-stop "tours." The itinerary of

these tours is influenced by retailer policies, travel costs, and other services and

entertainment consumed on the tour. These "hybrid" multi-stop tours are blurring the

distinction between retail shopping trips and other types of trips (Roy et al., 2001).

Online grocers might be interested in the efforts to classify hybrid trips. Some

classifications enumerated by Roy et al. are tours that a) start and end at home, b) start

and end at home with a work stop, c) start and end at work, and d) tele-orders, which are

orders placed over the telephone.

Of course online grocers would be interested in the fourth hybrid trip category tele-

orders. But there should be ample reason for online grocers to concern themselves with

the other three categories as well. This is because, according to Roy et al. (2001),

consumers consider aggregate properties of tours. This means that if groceries ordered

online can become part of the tour (by picking up the groceries at communal drop boxes),

or might not become part of the tour (thereby saving the consumer time by having the

groceries delivered at home).

The duration and sequence of activities undertaken during tours is important; so

much so that various models of activity duration during trips have been developed. Two

of the models are the unconditional and conditional risk models. Some of the many

determinations made by Popkowski et al. (2002) when using these models are shopping

after work decreases with age; single people shop often after work; poorer people are

more likely to indulge in leisure after work; after leisure people are likely to conduct

personal business, then shop. Online grocers could look at these determinations when

deciding web page content, delivery routes, and more.

Task allocation is similar to activity patterns. Task allocation is the allocation of

subsets of tasks by household members while considering context variables such as

constraints and imperatives relevant to those household members such as car availability,

and urgency (Borgers et al., 2001). Other issues to consider when attempting to predict if

a household member will perform a certain task are whether the task is mandatory or

discretionary; or is shared or performed alone. Borgers et al. (2001) assume that task

allocation can be correctly simulated. Although this assumption is debatable, it may be

worthwhile for online grocers to consider situations when household members must

perform a mandatory task, such as shop for groceries, while at the same time be

constrained by only having one vehicle because a spouse or child is using the sole car

when the shopping trip must be performed.

The logistical requirements of online grocers are different than that of brick-and-

mortar supermarkets. Although there is an overabundance of literature about retail

supermarket logistics, there is little that applies solely to the logistic requirements of

online grocers.

Third party logistic firms (3PL) can supply all or a portion of a companies'

warehousing and distribution needs (Balakrishnan, 2000). This could be a novel idea for

online grocers who do not want to pack and ship their own groceries. It might

also be an innovative service provided by a new kind of 3PL that caters exclusively to

grocery delivery. A difficulty in using 3PLs for grocery delivery could be in the

compensation scheme. There are many variables, such at type of goods (frozen or hot),

weight, changing demand, and expedited foods, to name just a few. According to

Balakrishnan (2000), achieving fairness in compensating multiple distributors can be

challenging. Fee tables usually are made to calculate delivery compensation. Two

primary ways to determine fee values are tariff and cost-based approaches. The former

gauges the delivery compensation by geographic regions, while the latter estimates the

delivery costs using a function of distance and weight of the goods (Balakrishnan, 2000).

Microsoft's Solver contains linear programming capabilities that can be useful to

even smaller online grocers that are developing fee schedules or in-house delivery driver


Grocer Web Site Design Issues

Wang and Gerchak (2001) support the importance of stock keeping units (SKU) in

the decision of a customer to shop at the store. Of course, customers prefer stores with

more SKUs. Therefore, a tradeoff has to be reached when allowing greater "virtual" shelf

space for more profitable items or showing more SKUs per web page.

Wang and Gerchak's (2001) research discusses the relationship between demand and

the allocation of retail shelf space. Because, according to the article, demand increases

when more shelf space is allocated to a product, an online grocer that more prominently

displays the higher profit items on its web site might have lower delivery costs relative to

the cost of the groceries delivered. This issue deserves more research. Interestingly,

Wang and Gerchak (2001) say that for any two identical retailers, the total profit made

between them depends only on their total inventory level, not on how the total inventory

is allotted between them. This could be significant when two online grocers, or an online

grocer and a brick and mortar grocer compete in the same market area. The word

"identical" can be ambiguous. But, if identical is defined as the same amount of SKU's,

then, if this proposition is correct, an online grocer that has the same amount of SKUs as

a brick and mortar grocer should have a reasonably good chance of selling as much as the

brick and mortar store. In "Designing an Effective Cyber Store Interface," Kim and Eom

(2002) talk about the importance of clearly showing on a web site that risk-free and on

time delivery is offered. They propose four elements of customer satisfaction. The

elements are product and/or service, support, bad experience recovery, and extraordinary

service. The factors of risk-free and on time delivery could be classified under any or all

of Kim and Eom's (2002) elements of customer satisfaction. Throughout their article it is

said that an e-commerce site should allow customers to comparison shop. Taken in

conjunction with showing the greatest possible amount of SKUs on a web site, if a site

allows, or even encourages customers to compare the prices of the host grocer's SKUs

with the prices of other grocers products (online or not), customer satisfaction and sales

might increase.

Artificial Intelligence

The study of neural networks is a branch of artificial intelligence that pertains to

roughly mimicking the brain to make calculated decisions. Literature exists that shows

the attempts to include the decision-making power of artificial neural networks (ANN) in

GIS. Although I have not located any literature that specifically uses ANN to determine

delivery routes for online grocers, ANN has been used with GIS for other types of

applications, which shows the feasibility of using these technologies together.

The predictive power of the combination of GIS and ANN to forecast land use

changes in Michigan is documented by Pijanowski et al. (2002). The article says the use

of GIS and ANN can "aid in the complex process of land use change." Also, "ANNs are

powerful tools that use a machine learning approach to quantify and model complex

behavior and patterns." This is precisely the type of functionality online grocers need in

order to predict demand and create tentative delivery zones to better prepare for that

demand. Pijanowski et al. state that because of the spatial nature of the variables, the

incorporation of ANN with GIS is "essential." The same should hold true for an online

grocer's logistical dynamic routing application.

Black (1995) writes that using ANN with the gravity model. He uses a gravity

artificial neural network (GANN) to show commodity flows between nine United States

census regions. Using as input values regional flow of commodities, regional flow

attraction, and interregional distance, he discovered that the accuracy of the gravity

model improved when he moved from the unconstrained (conventional) gravity model to

the GANN. Black concludes that flow modeling can be "revolutionized" by using the

GANN model. What Black's research contributes to the purpose of this dissertation is the

clarification that ANNs increase the flexibility and usefulness of not only GIS, but also

spatial modeling in general.

Black (1993) states that an industrial firm must make three decisions after it has

decided to operate. 1. Where to locate, 2. What technology to use, and 3. What the scale

of production will be. Although Black was mainly referring to manufacturing firms, his

three points can also apply to an online grocery initiative. The technology inferred by

Black can include GIS, computer simulation, SDSS, intelligent traffic systems and more.

Scale could refer to the scope of the business model as referring to delivery methods and

delivery zones, while firm location is of course where the grocery store and its

distribution centers) are located in comparison to the customer base. These three factors

should be considered within the entire online grocery delivery model.

It should be noted that a concern of both Pijanowski and Black in the above-

mentioned research about ANN and spatial technology is the problem of spatial

autocorrelation. Research exists addressing this problem. One example is from Duckham

(2000) who discusses "error sensitive GIS." Relevantly, this literature also touches on the

realm of artificial intelligence because it uses induction, an established artificial

intelligence technique. According to this article, "an inductive learning algorithm should

be able to automatically deduce rules that embody the patterns in that data..." According

to Gahegan (2000), inductive machine learning is gaining popularity in the geographic


Interestingly, even though, according to Duckham et al. (2000), the inductive

algorithm deduces; it does not induce, this literature is noteworthy because it endorses an

artificial intelligence (AI) solution to spatial problems such as autocorrelation that could

be otherwise intensified by the incorporation of another AI solution, namely artificial

neural networks (ANN). Additionally, because of the low profitability and labor

intensiveness of ascertaining the amount of error in geographic data, the article says that

there probably will not be many companies strictly concerned with quality issues in

geographic data. Duckham et al.'s (2000) research therefore recommends the inductive

learning algorithm because it is a relatively low cost way of ascertaining spatial data


Neural spatial interaction models are somewhat related to gravity models. Three

gravity model input variables relate to measures of "origin propulsiveness, destination

attractiveness and spatial separation" (Fischer, 2002). Neural spatial interaction models

relate to gravity models not only because they expedite mathematical modeling, but

because they are adaptive enough to "deal with incomplete, inaccurate, distorted, missing,

noisy, and confusing data" (Fischer, 2002).

Although the field of artificial intelligence includes technology such as expert

systems, neural networks, and fuzzy logic, intelligent agents possess a particularly

promising niche in the field of online grocery supply chain optimization and visualization.

Because of their ability to "automatically" scan intranets, extranets, and the internet,

intelligent software agents (or just "intelligent agents"), or bots, can be programmed to

periodically monitor databases, data marts, and directories. The gleaned data could then

be made available in a more informative format, such as in GEOS-enabled online grocery

logistic solutions.

Although using bots to collect data can be difficult (Fontana, 2002), companies, such

as SAP, are aggressively pursuing the utilization of bots in ERP (enterprise resource

management) and SCM (supply chain management) solutions. Steve Ranger tells how

SAP is using bots to gather supply chain partner data and make the resulting information

available through portals (Ranger, 2001).

One of the forerunners in intelligent agent research is the Carnegie Mellon

University. A look at their intelligent software agent website gives an inkling of the

branches of this field of artificial intelligence (Carnegie Mellon, 2001).

The Carnegie Mellon site shows a few bot technologies that could be of particular

value to online grocery logistic plans. Local area discovery and multi-agent learning are

two such technologies. In multi-agent learning the bots learn in "dynamic environments"

such as that which exist between different networks of companies interior supply chains.

Business rules markup language (BRML) is an intelligent-agent facilitated

technology that could help heterogeneous e-commerce-related applications to exchange

"rules." It is referred to as an "interlingua" technology that was developed by IBM while

the company was working on the Business Rules for E-Commerce project. What this has

to do with intelligent agents is that bots have been relying on the knowledge interchange

format (KIF) to transfer knowledge. BRML is supposed to add "prioritized conflict

handling" capability to KIF, which is important for maintaining business rules and when

transferring information between systems such as those between different internal supply

chain partners.

West and Hess (2002) say that software agents (intelligent agents) should be

employed to help users with interactivity and difficult spatially relevant jobs. This also

pertains to the user friendly, or procedural aspect of GIS knowledge management, even

when the procedures to use the specific GIS package are contained within the instruction

manual that is geared for GIS programmers and analysts (West and Hess, 2002).

Intelligent agents can facilitate comparison shopping. Kim and Eom (2002) suggest

using intelligent agents to not only allow customers to compare product prices, but also to

compare products on multi attributes. For online grocers the various attributes could be

product freshness, nutritional information, average delivery time per product, and more.

The article states that just as important as allowing your customers to do comparison

shopping through agents, is the need to not block competitors' intelligent agents from

entering your site to perform comparisons (Kim and Eom, 2002). This is because if a

store does not appear on an online shopper's online comparison list because that store

blocked the incoming agents, that shopper might think that particular store does not carry

the item that the shopper is looking for.

Intelligent agents are discussed by Tsou and Buttenfield (2002) in their literature

about GIServices (GISServices are explained below). Agents can be used on a distributed

GIS-enabled computer architecture to find and bind spatial data objects across networks.

According to these authors, three intelligent agents perform the distributed GIS system

find and bind functionality. They are filer agents, interpreter agents, and decision agents.

Collectively, these agents find the requested information, filter out unnecessary data,

bridge heterogeneous data environments, and autonomously make decisions such as what

server to use to send the requested information to the GIS user. Sometimes spatially

enabled artificial agents communicate between themselves using the knowledge query

and manipulation language (KQML) (Tsou and Buttenfield, 2002). Online grocers with

more than one distribution center should consider using this type of technology when

adopting GIS-enabled delivery solutions.

Although not strictly necessary for the development of delivery zones, swarm

intelligence and complexity are scientific fields that have recently come to the fore in

some business strategies. A primary reason that this literature is reviewed is because of

the intuitive value that these two concepts can have for the construction of delivery

regions in a geographic enterprise optimization system (GEOS).

Literature exists that explains the combination of GIS with "swarm intelligence." A

project undertaken by the Department of Geography at Southwest Texas State University

incorporates swarm intelligence with GIS to "study and simulate" various entities in

supply chain situations (Zhang, undated).

Swarm intelligence can loosely be defined as a neural network of interacting

intelligent agents. Whereas a neural network is commonly a closed network, meaning that

inputs and outputs are controlled, swarm intelligence can denote the utilization of

intelligent agents moving over disparate systems to detect unforeseen patterns (Payman,


Eric Bonebau is one of the pioneers of applying the concept of swarm intelligence to

managing business. He is a coauthor of "Swarm Intelligence: From Natural to Artificial

Systems, 1999. According to Fredman, ants set up supply chains to accomplish

sophisticated tasks, and the emulation of these processes can help companies deal with

complex environments. Her literature explains how logistics is a "natural application of

these ant-foraging algorithms."

Conventional supply chains use more or less centralized methods to forecast demand

and fulfillment. According to Roy (1998) centralized planning methods can be disrupted

by fluctuations in customer demand. More suitable than centralized planning is

conceptualizing the supply chain as a supply web, and incorporating swarm intelligence

to traverse the web to better calculate and compensate for demand fluctuations (Roy,

1998). One particular statement by Roy (1998) summarizes a main point of my emphasis

on profitability thresholds as driving forces behind the shape of delivery zones. This is

"any changes in customer demand...can easily affect how feasible or profitable a given

plan may be. These unplanned events will always happen, so it is desirable to create a

management system that adjusts to them more gracefully" (Roy, 1998). He uses a vehicle

analogy is used to demonstrate this point. Simply stated, instead of having a "centralized"

vehicle and adding parts to it, it can be better to conceptualize having thousands of parts

that "flock together" to become a vehicle (Roy, 1998). This concept may be very useful

when parceling groceries together for individual orders or routes.

According to Roy (1998), if swarming intelligent agents learn about region-specific

pricing dynamics they could help "develop new strategies to take advantage of changing

market conditions. Roy (1998) says that each swarming agent can individually generate

"internal make" suborders. It is not difficult to visualize how this could facilitate

packaging and shipping of groceries purchased online. After some time the agents can

attain the ability to "create new forecasting techniques and learn which of the techniques

are most accurate" (Roy, 1998). When it comes down to it, accurate forecasting is the

linchpin that online grocers depend on. At an extreme, if forecasting could be 100%

accurate, no wasted efforts in packing or shipping would be incurred, which would of

course result in significant competitive advantage.

Enterprise Resource Planning (ERP) is an established technological methodology

that is used to facilitate and assimilate information flows between and within departments

in an enterprise. ERP uses centralized databases, common user interfaces, and

sophisticated middleware to optimize the information flow within a company, or between

supply chain members, thereby increasing the accuracy and timeliness of inter-corporate

and intra-corporate decision making. ERP vendors are increasingly incorporating

forecasting functionality into integrated applications. J.D. Edwards sells the Demand

Planning 4.0 collaborative application module to enable partners to better make forecasts

throughout the supply chain (Ferguson, 2002a). According to Ferguson, the software uses

"market intelligence" and demand information to predict product demand. Although the

article does not state from where the market intelligence originates, it would be desirable

if the information comes from current industry-specific marketing research data. Demand

Planning 4.0 also accounts for holidays and promotions when forecasting demand. A key

word in the description of this supply-chain centered product is "collaboration." This

effort by J.D. Edwards shows that there is importance in having forecasting information

readily available when collaborating throughout the supply chain. More collaboration-

related literature is reviewed below.

Southwest Airlines is saving $10,000,000 annually by using swarm intelligence to

help route cargo (Fredman, 2003). Problems were arising from ground crews using their

judgment about what flights they should forward luggage and other cargo to. Naturally,

the cargo handlers would wait for the next flight that had space in the cargo hold that was

going in the same direction as the luggage. This procedure, however, was not optimal, as

discovered after using swarm intelligence. This technology revealed that many times it is

better to leave the luggage on a plane than to use ground crew manpower to change the

luggage in mid route. Southwest's freight transfer rates have reduced by up to 80%, the

necessity for storage facilities has been decreased, and payroll has been reduced because

of the decrease in necessary manpower (Fredman, 2003). Other noteworthy companies

that are using swarm intelligence-related solutions effectively are Unilever and Capital

One. As stated by Fredman (2003), the use of swarm intelligence in complex

organizations helps these organizations harness their tacit knowledge. Tacit knowledge is

knowledge that is difficult to codify in databases, but has high value because of the

exclusivity it has to the company that is using it. In other words, tacit knowledge can give

a company its competitive edge. This is precisely what is needed for businesses to

succeed in the low profitability margin online grocery industry.

As expected, "swarming" intelligent agents perform tasks that mimic life forms such

as ants, bees, or birds which exhibit swarming behavior (Resnick, 1998). Foraging for

data (like ants foraging for food) is the primary task of swarming agents. It is not hard to

imagine how swarming intelligence could help create delivery zones. Swarming agents

could collect data from tens of thousands of entities in delivery regions such as PDAs,

trucks, databases, and directories, and present this information in real time to strategic

decision support systems.

Deborah Gordon (2002) published an interesting article about task allocation within

an ant colony. Amazingly, although essential tasks such as caring for the young ants, nest

construction and maintenance, and foraging are accomplished in a very efficient manner,

nothing is in charge of "managing" these operations (Gordon, 2002). According to the

literature, it seems like some of the task allocation within a nest depends upon the number

of ants that are already involved in performing that particular task. For example, if an ant

leaves the nest to forage for food and does not return, no other ants leave the nest that day.

But if it does return, successively increasing amounts of ants will leave to find food and

perform other tasks like carrying any dead ants back to the nest.

Communication between ants is performed when the insects touch their antenna.

Also, ants emit an odor that is specific to the task that they are performing. The ants do

not tell the other ants what to do; instead it is the interaction pattern that determines the

probability that an ant will perform a job (Gordon, 2002). Another relevant fact is that

because ants live only a year, there is no handing down of knowledge from one

generation to another. The work that gets done is due to indigenous knowledge that the

nest collectively possesses.

Gordon's (2002) research alludes to ongoing work that uses intelligent agents to

simulate the collective jobs that are performed by social insects such as ants and wasps. It

should be relatively easy to imagine the benefit that the incorporation of these types of

agents into an information system that creates delivery zones. The potential decision-

making accuracy, resilience, forecasting ability, and longevity of such a system are some

of the benefits that make the utilization of swarm intelligence a technology that is

worthwhile of study by online grocers.

A relevant project undertaken at the department of Zoology at Michigan State

University is the Multi-Agent Based Economic Landscape (MABEL). The project

simulates the group dynamics of individuals in institutional and organizational

environments by using swarm intelligence with GIS (Mabel, 2002). Although the

incorporation of intelligent agents into GIS-augmented online grocery delivery routing

applications has yet to develop, many of the technological tools already exist to make this

combination possible.

Ants and bees have been studied to determine how these insects efficiently work

together to accomplish sophisticated tasks such as procuring nutrition or building

tectonically durable nests and hives. These types of studies are not only the

underpinnings of swarm intelligence theories and practices, but also integral aspects of a

field called complexity science.

Van Uden et al. (2001) call complexity science a body of knowledge that is "not

trivial" for business, and the "contender for the top spot in the next era of management

science." Carol Kennedy (2000) says that complexity science has "exciting possibilities"

for various fields including logistics. The Southwest Airlines example above was an

instance of a company using artificial intelligence to solve a "chaotic" situation.

Complexity science recognizes that systems, whether those systems are biological,

technical, or organizational, exist in a chaotic world where forces are constantly

impinging upon the system. Regions, such as delivery regions serviced by online grocers,

can clearly be considered examples of systems existing in a chaotic and complex

environment. Whenever online orders are coming into the grocer's system, the regions

that need delivery should change shape or composition. Various complexity issues are

inherent when considering the delivery, not least of which is the profitability threshold

that encompasses myriad variables such as packing time, crime rate along the route, fuel

costs, driver familiarity with the area, customers) lifetime value, competitive pricing,

traffic en route, position of other delivery drivers, customer payment history and payment

method, customer profile, and much more. The boundaries of the region are a constantly

changing factor because the shape and geographic area of the region's boundaries can

allow, or disallow the region to meet a profitability threshold.

Many properties make a system complex, such as "incompressibility" which is the

inability to explain a system in a level that is less complex than the complexity level of

the system itself, without losing some of the explanatory aspects of the system. Another

notion that is perhaps more relevant to online grocery delivery methods is that a complex

system is both deterministically chaotic and anti-chaotic. Deterministically chaotic means

being "incredibly sensitive to small disturbances," and anti-chaotic means being

"incredibly insensitive to large disturbances" (Van Uden et al., 2001). It is probably best

for a delivery zone delineation system to be resilient enough to withstand large,

unprecedented spikes in orders, while being selectively reactionary to small patterns or

states that could make a region profitable or not.

Van Uden et al. (2001) sum up complexity science by saying that it is the body of

knowledge that claims that "everything is connected to everything else." This author's

literature also tells how unwise it is to concentrate on select parts of a system while

ignoring the interactions between these parts. If we consider delivery zones as parts of a

logistics system, we might find that interactions between these zones are important in

determining if the zones should be combined into one deliver zone to meet the

profitability threshold, or remain separated.

Complexity science tells us that permanent boundaries never exist (Van Uden et al.,

2001). Here again is an argument in favor of changing the boundaries of delivery zones in

real time, depending on the overall logistical, operational, and profit "habitat" at hand.

"In complexity science, all boundaries are emergent and temporary" (Van Uden et al.,


Another complexity science maxim that should be followed by online grocer

delivery zones is the property of self organization, which is synonymous with anti-chaos.

As was explained in the section about swarming insects, complex systems have the trait

that they can self organize. Concisely stated, when there are a large amount of elements

in a large state space, when the conditions are random, these elements "tend to converge

into small areas of this space" (Van Uden et al., 2001). The delivery zones should do

exactly this. Instead of considering each and every customer as an individual profit

creating entity, the system should converge the customers into profitability zones,

regardless of the geographic proximity of the customers with each other and with the

store (up to a certain range).

Interestingly enough, this literature says that a trait of self-organization is that if the

starting conditions are similar, "a quantitatively similar pattern will always emerge" (Van

Uden et al., 2001). This can be somewhat reassuring for online grocers because there will

of course be times when a certain proportion of the deliveries will be similar to previous

deliveries. This phenomenon should significantly aid in forecasting and pre-packing the


Stephan Toffler's book "Adaptive Corporation" is based on many complexity

science maxims. He advises the usage of "complex adaptive systems" to cope with the

"unpredictability of the modern marketplace" (Lloyd, 2000). Similar to the self-

organizing nature of complex systems, Toffler's "Sense-and-Respond" system allows the

realization of benefits such as mass customization and the ability to capitalize upon wild

marketplace swings. This book shows the growing awareness of the viability of running

organizations as complex systems, instead of operating the organizations hierarchically.

The ability to mass customize can be very important for online grocers, especially if we

consider packaging as an operational entity that also can be mass customized which

indeed it is because the contents of each package is different than each other package.

Therefore, the concepts offered in Toffler's book can apply to online grocers. Both the

meaning of "complexity science" and "complex" can vary depending on who is defining

the terms) (Van Uden et al., 2001). Christoph Adamnil (2002) said, "Nobody knows

precisely what is meant by the word 'complexity.'" John Casti (2001) attempts to clarify

what the meaning of complex is. By using the examples of infinitely continuing repeating

decimals as opposed to finite decimals, he says that of course the infinite non-repeating

decimals are more complex. He goes on to say that randomness is also measured by

degrees, while Adamil (2002) asserts that randomness is the opposite of periodical

occurrences. The relevance of this literature is that an organization might want to

determine the degree of complexity or randomness of occurrences in an environment

before deciding on what type of system to implement. If one agrees with Adamil's (2002)

assertion that "randomness does not give rise to organisms," then before a company

establishes a system to cope with complexity, the randomness of the environment should

be determined in order to predict if the system will be viable that is, if any type of

analogy between an information system and an organism exists. Casti gives a general

outline about how to determine complexity and randomness in various environments.

Adamil (2002) views complexity as the amount of information that an organism stores

about its environment.

Global companies are beginning to become aware of the value of complexity science

in their business operations. General Motors, Capital One, and Ford are three such

companies (Wujciak, 2003). Perhaps the example most germane to this dissertation is that

of Ford. Ford allowed a large array of automobile configurations to be ordered by

distributors, which created difficult to manage manufacturing schedules and inaccurate

demand forecasts. Ford solved the configuration problem by using methods to reduce


One possibly useful way of thinking about "populating" a routing information

system that creates delivery zones on the fly is a concept offered by Adamil (2002). He

says that entropy is "potential knowledge" and that "sequence entropy" is a length of a

tape, and the marks on the tape is the information. Measurement, which in the case of

online grocers would be the geographic measurements of the delivery zones, populates

the tape. In other words, the geographic measurements of the delivery zones turn the

online grocer's entropy into information. Most relevantly, Adamil (2002) states, "the

information-filled tape allows you to make predictions about the state of the system that

the sequence is information about." In this case the state would be the state of online

orders at any particular time, and the predictions would be forecasts of those orders.

Intuitively, Adamil (2002) says that a "complexity catastrophe" is a rapidly changing

environment. We could view all of the above-stated online grocer failures as resulting

from complexity catastrophes. That is if we consider "rapidly changing" to be the state

caused by the reception and response to online grocery orders. "If the changes are fast

and extreme, not only will the organism be maladapted to this new environment, but also

its measurable physical complexity will have decreased commensurately" (Adamil, 2002).

This is exactly the situation Webvan found itself in, which of course led to its demise.

Therefore, any online grocer that begins a delivery initiative must keep its physical

complexity at the utmost minimum that is necessary to fulfill the orders. In other words,

multimillion dollar distribution centers, such as those created by Webvan, may not be

able to cope with the dynamic environment created by online grocery sales.

Online Grocers' Store Locations

If an online grocer has not yet decided on a site to open the brick-and-click store, the

site selection process becomes highly relevant to the businesses' overall strategy. Thrall

(2000) asserts that investment in retail location is one of the most important investments a

business can make. It can also be the most costly.

Moreover, when one considers that the central store is the epicenter of grocery

dispatches, which in turn can contribute to the success or failure of the company because

of the high delivery expense and the low profit margin, the site selection process takes on

a high magnitude of importance. There is much literature that extols the value of GIS for

the retail industry (Grimshaw, 1989; Brown, 2000; Murayama, 2001a; Thrall, 2002a;

Hakala, 2003). Although I have found no literature that deals with the use of GIS to

create metamorphic delivery regions for online grocers, an abundance of literature exists

telling how important GIS is for retail location, delivery, and retail marketing.

Online grocers should use heuristics and techniques that are different from brick and

mortar grocers when deciding on the location of their store or distribution center.

Although the term "online grocer" is used throughout this dissertation, I suggest that any

store selling online should be of the brick-and-click kind. This paper will refer to

exclusively online grocers as "pure online grocers."

Vlachopoulou et al. (2001) enumerate transportation type(s), customer type(s),

competitor locationss, and sales levels as some of the factors to consider when choosing

a site. The authors recommend using GIS to facilitate the site selection process because

of the data and spatial modeling complexity involved. Site selection tools fall in various

categories. These categories include analog models, rules of thumb, linear programming,

simulation, checklists, gravity models, and linear programming. The "parasitic" approach

is mostly used by smaller stores that copy the location decisions of larger retailers (Clarke,


A broad rule of thumb for online grocers to consider is that goods that are consumed

frequently should be distributed from a dense network of locations (Borchert, 1998).

Groceries certainly fit this category, so an online grocer will have to reach a tradeoff in

the site selection decision as to have more distribution centers, or stores, and less delivery

trucks, or vice versa.

Hernandez et al. (1998) say that because retailers have notoriously been "cavalier" in

their approach to store location. Techniques for deciding on the location of a store range

from using checklists to sophisticated artificial neural network (ANN) systems. Moreover,

some location decisions are in response to location decisions made by the competition.

To better strategize retail location, some companies have made semi-autonomous

property divisions within the company to manage store location (Hernandez et al., 1998).

This means that smaller scale online grocers might have to have some type of "equalizer"

because these stores will probably not have the resources to create an entire department

devoted to store location or even marketing. Therefore, online grocers could use

Hernandez et al.'s (1998) tactical, or local marketing, strata to help compete. Hernandez

et al. divides retail location techniques into three strata strategic, monadic (location mix,

such as relocation, re-fascia, and re-merchandising), and tactical (local marketing). One

tactical technique that online grocers should be able to efficiently capitalize upon is

changing online food prices in response to competitors' prices. Although brick-and-

mortar stores can do this by using the UPC (universal product code) and point-of-sale

technology, these stores must rely on periodic advertisement to inform customers of

prices. Online grocers who have developed a significant customer base can inform these

customers of price changes and coupons by e-mail or as soon as the customers log on to

the website.

The maxim that those who do not know history are doomed to repeat it can be

applicable to online grocers' site selection strategies. Thrall gives a detailed account of

the family "club," showing how the amount of people per dwelling has decreased though

time in the United States (Thrall, 2002a). Borchert (1998) supports this fact, while also

talking about decreasing birth rates of families. The ramification of these and other

demographic trends should be looked at closely by online grocers when deciding where

to locate.

Borchert (1998) writes about how lower level stores that were located in city centers

are losing their place in established retail hierarchies. It might be possible, however, for

online grocers to stem the tide of location displacement by virtue of their delivery

strategy. This conjecture remains to be seen, however. Nevertheless, the reinvigoration of

many downtown areas, plus the governmental dissuasion of commuters to use cars in

downtown areas (primarily in Europe) (Borchert, 1998), can also have an influence on

online grocers location strategy.

Clarke (1998) takes the utilization of GIS for retail location planning and divides it

into three eras, before approximately 1985, when the technology was not used at all for

retail location; from that time to the late 1990's, when GIS gained a foothold in many

retail organizations; and the present time when data mining and optimization techniques

are used with GIS for retail site selection. He supports his threefold division by showing

that retail location has become much more complicated recently, necessitating the usage

of artificial intelligence and optimization techniques.

Geographic Information Systems for Traffic-Flow Simulation

Recently efforts have been made to use GIS for simulating traffic flows for various

reasons. Considering the great amount of money to be made or lost with an online

grocery, simulating the time to delivery should be an integral aspect of any strategic plan.

Above, it is described how Punakivi and Saranen (2001) used simulation to

determine the delivery times to houses using various criteria such as attended and

unattended deliveries. Simulation is also used to determine traffic flows within cities'

main commerce areas and between different city centers (Medda, 2003). Traffic

simulation can be divided into macroscopic, mesoscopic and microscopic traffic flow

simulation. Mesoscopic simulation shows traffic flows at a scale between macroscopic

and microscopic. Nokel et al. (2002) have published research about mesoscopic traffic

simulation, while Wahle, Chrobok, Pottmeier, and Schreckenberg researched microscopic

traffic flow simulation. The research about mesoscopic traffic simulation states that there

are many times when the area simulated must be wide enough to be considered

macroscopic, but the finer level of roadway detail necessitates a granularity that the

authors classify as mesoscopic (Nokel et al., 2002). On the other hand, Wahle et al.

(2002) contend that because of the increasing processing power of computers,

microscopic traffic simulation is becoming more feasible. The article also says that

microscopic traffic simulation is valuable as a transportation planning tool which helps

vehicles navigate the roadway easier.

If a comprehensive grocery delivery plan is to be implemented, simulation using all

three scopes of simulation (macro-, meso-, and microscopic) should be considered.

Wahle et al. (2002) say that the output, such as travel times, from traffic simulation

systems can be used as input for other intelligent systems. This is precisely what is

necessary for a dynamic (real time) grocery delivery schedule information system. The

system tested by Wahle et al. (2002) supposedly functions in real time.

Simulation can be used to overcome the current analytical and dynamic modeling

inadequacies of GIS applications (Wu, 1999). He recognizes the importance of modeling

space-time spatiotemporall) processes through simulation with GIS, and is supported by

literature by Bernard and Kruger (2000). Integration of GIS with spatiotemporal models

will allow the "forecasting of space-time patterns in a reasonable, repeatable and

consistent way," precisely what would be required of an application that shows delivery

trucks' performance based on profitability thresholds. While noting that GIS-based

simulation is quite complex, Wu (1999) says that three contributions can be made to

decision makers by this type of simulation description, prediction, and prescription.

Relating to the macro, meso, and microscopic simulation scopes stated by the authors

mentioned above, Wu says that GIS-enabled simulation may reveal relationships between

micro and macroscopic behaviors. As applied to grocery delivery, the profitability

ramifications related to small deviations in order times might not be discernable in a

microscopic simulation, but might be noticeable if the simulation was run on a

macroscopic scale. As stated by Wu (1999), the "purpose of simulation is to see how a

global structure is evolving from uncoordinated individual behaviors."

Another benefit of GIS-enabled simulation noted by Wu is that the simulation results

can be stored in GIS "scenario libraries" for use in later simulations (Wu, 1999). Bernard

et al. discuss the integration of GIS with simulation in some detail. The difference

between coupling and integration of GIS applications is explained. Coupling is

transferring data between two GIS, while integration is a "monolithic" model where GIS

and simulation tools are implemented on top of a common data and method base

(Bernard, 2000). The literature goes on to explain common ground between initiatives

undertaken by both the OpenGIS Consortium (OGC) and the Simulation Interoperability

Standards Organization (SISO). The OGC's mission is to facilitate the worldwide

proliferation of "geoinformation services." This is to be accomplished in part by the

utilization of two models and a specific architecture, called the Open Geodata Model, the

Information Community Model, and the Service Architecture. Much of this initiative is

fashioned after the web services paradigm. Of similar importance is the High Level

Architecture which was developed by the US Department of Defense, and is supposed to

be a standard for interoperable simulation components (Bernard, 2000). According to this

literature, these initiatives are necessary because there is a "gap between the offered

interfaces and services in the GIS area and the requirements of the simulation area."

The integration of simulation and GIS can be highly relevant for online grocers who

deliver. This integration should allow the use of "what if" analyses to ascertain if delivery

regions will meet their profitability threshold or not, which, in turn, will allow managers

to better decide when to pre-pack the groceries that have the most probability of being

purchased by customers in those regions.

Geographic Information Systems for Vehicle Routing and Logistics

Pain Stubing of Ernst and Young said, "Execution is the most important thing for

any retailer" (Reardon, 2000). This can not be overstated when considering online retailer

delivery dispatches. GIS, and the role the technology plays in routing, can help online

grocers dispatch delivery trucks in the most cost-effective manner.

Closely related to traffic simulations are routing scenarios. There are significant

amounts of literature that pertains to the role GIS plays in vehicle routing that can apply

to online grocery delivery. Sometimes the GIS can be a component in a spatial decision

support system that helps determine the best routes.

Campbell et al. (2001) tell how GIS has helped create new logistical solutions. Their

paper talks about a hybrid distance approximation solution to assign routes to snow

removal trucks in Montreal. Their hybrid model strives to "combine accuracy of shortest

paths ... with the simplicity of approximations." One aspect tying this research to my

dissertation is that the shortest path algorithm alone is perhaps insufficient to determine

the best route for delivering groceries. There are other factors that might make a longer-

distance path preferable. Essentially, these authors' hybrid method speeds up the routing

process by creating a reduced road network that eliminates superfluous roads. According

to the authors, the calculation of routes in very large networks can be cost and time

prohibitive (Campbell et al., 2001). A problem with many routing tools is that they view

the network and traffic costs as "static" entities, or entities that have attained an

"equilibrium" state (Wu, 2001). This is clearly unacceptable for fast changing online

grocery delivery schedules. The article reinforces this idea by saying "... a static model

of congestion is an oxymoron." The need for forecasting traffic flows and congestion is

addressed in the article. The acronym GIS dynamic traffic assignment (GIS-DTA) is

coined by the authors. Aided by a GIS, analysts can change delivery routes according to

changes in predicted traffic flows. This simulation was performed using Arc/Info GIS.

The importance of coordinating delivery schedules with traffic congestion is underscored

by Browne as he says that delivery reliability decreases as congestion increases (Browne,


A notable amount of recent literature exists about using simulation in conjunction

with spatial decision support systems (SDSS) to solve the vehicle routing problem.

While explaining their Hierarchical Path View Routing (HPVR) model, Huang

breaks down vehicle routing systems into two broad types. One is where individual

vehicles perform their own routing calculations using CD ROM based maps and on-board

computers. The other method uses a centralized path (route) discovery model. The

centralized path discovery model was demonstrated by Huang (1997) to be less expensive

to implement

The HPVR model is relevant to this dissertation. The model uses only a subset of all

the roads possible to suggest a route for a vehicle. They call this method "fragmentation"

of arteries. The fragments are classified by road type (highway, side road, etc). This

method makes routing computational costs cheaper because not all the roads are used to

calculate any particular path from origin to destination. Because of constantly changing

road conditions, the status of routes must be recalculated frequently. If HPVR is used,

this frequent recalculation of routes can be performed more quickly because fewer roads

are used within each hierarchy.

It is worth mentioning here that the hierarchical set that includes highways would be

used to calculate paths between regions, which may not be necessary for local grocery

deliveries. However, this model has merit because if it is employed by local grocers,

unnecessary or superfluous roads may be omitted from local route calculations, thereby

speeding up the route-optimization process. Alternately, if the online grocer wishes to

branch out (for example, on weekends) to other customer regions, it could incorporate the

main artery hierarchy into the analysis. Also, although I don't propose using a

hierarchical method per say to form delivery routes, the hierarchical segregation of roads

for logistics and transportation planning has intriguing potential to be used for further

grocery delivery route optimization efforts.

GIS can also assist with determining accessibility measures for roads depicted with

ITS (intelligent transportation systems). Miller et al. have developed a method of ranking

the accessibility of arteries using ArcInfo (Miller, 2000b). Their space-time accessibility

model (STAM) calculates the accessibility of nodes in a network based upon drivers'

activity schedules. Activity schedules are essentially the times that a specific set of users

will use the roads. Accessibility to roads will be greater or less depending upon the time

of day. Based upon "travel diaries," drivers' activity schedules are divided into

mandatory and discretionary travel times. A similar concept might be used to group

online grocery customers into mandatory or discretionary delivery time windows.

Another consideration online grocers might have when developing a logistics

strategy is the type of traffic jam that occurs during a certain time of day. Mahnke and

Kaupuzs (2001) classify traffic flow into three "regimes." The first is a free flow of small

densities of vehicles, which is obviously the preferable time to embark on a delivery

route. Next is the "coexisting phase," where traffic jam clusters coexist within free

flowing traffic. And the "viscous overcrowded situation," where a high density of cars

move at low velocities." These broad classifications could be assigned as variables to the

delivery zones to better ascertain if the groceries can be delivered on time. Quite possibly,

for example, easternmost delivery zones could be experiencing the first category, while

northwest sections of town may be undergoing the viscous overcrowding condition. With

close inspection, it could be discovered that this pattern repeats itself daily. This

information should help to enable the scheduling of grocery deliveries with close regard

to factors such as congestion patterns.

An inseparable aspect of online grocers' delivery solutions should be the

incorporation of GIS into their logistic plans. The entire concept of delivery zones is

merely a hollow conception if not executed with a GIS.

A company called IVU Traffic Technologies uses GIS as an integral part of its e-

logistics solution (IVU, 2000). This company shows foresight by creating a solution that

allows visualization of logistical spatial information to allow informed management

decisions. The company asserts that GIS functionality is important for e-logistics

solutions because of the importance of location visualization for management decisions.

IVU claims, as this paper also asserts, that data visualization, as facilitated by GIS, is

especially important for e-businesses that utilize e-logistics solutions. An example of

advances made in the GIS visualization field is 3D Analyst. This software application is

an Environmental Science Research Institute (ESRI) product that allows realistic

animations such being able to "fly through" a region (Thrall, 2002b).

A Korean governmental organization that incorporates GIS into an e-logistics

solution is the Electronics and Telecommunications Research Institute (ETRI) which

develops GIS / e-logistics postal solutions in their Postal Technology Research Center

(Electronics and Telecommunications Research Institute, 2003).

Kipling Holding AB, a German mobile internet technology company also develops

e-logistics solutions using GIS. Kipling's niche is the utilization of global systems for

mobile communication (GSM) in their e-logistic solutions. GSM is a prominent

technology used for second and 2.5 generation wireless telecommunication services

(GEO Community, 2001).

Global systems for mobile communications is showing itself to be a valuable

technology in the advancement of e-logistics solutions as evinced by an Indian e-logistics

solution provider called, most appropriately, e-Logistics Ltd. According to the owner of

e-Logistics Ltd., Mr. V. Sanjeevi, the usage of GPS to monitor trucks is more expensive

than using GSM in the e-Logistics solutions (GIS Development, 2001). Sanjeevi also

asserts that en-route delay of delivery trucks can be reduced significantly if e-logistics

solutions are utilized.

According to IVU Traffic Technologies, "about 80% of all data in a company is

location driven" (IVU Traffic Technologies, 2001). The company's FilialWeb technology

enables the selective presentation of spatial information. The word selective is important

here because even if companies are able to garner and store real-time spatial information,

it is another matter to be able to present it selectively to the operational employees that

must make a decision about whether a grocery delivery route is profitable or not.


LeHeron et al. (2001) explain how supply chains should be networks of tacit and

codified knowledge; and that within the last decade there has been a large-scale change in

the networks of food consumption. If localized food consumption, especially the

consumption that is fulfilled through online grocers, is thought about as being fulfilled

through a network of supply chain knowledge, then more innovative models of delivery

may be contrived. Using this mode of thought, online grocers might be able to devise

ways of having their suppliers coordinate with the delivery drivers to possibly deliver

more bulky items to the drivers as they are en route to or from the customer premises.

According to Balakrishnan et al. (2000), there has been "increasing emphasis in

recent years on customer orientation providing the right product at the right price, time,

and place." And this emphasis has "propagated up the supply chain."

Industrial networks have been defined as the "spatial pattern of sales linkages"

(Benenson, 1998). According to this literature, dense and complicated networks can exist

even when small sets of linkages (such as those established by small to medium sized

online grocers with their suppliers and customers) exist. These types of networks are not

unlike matrices, where the dimensions are equal to the number of interplaying objects.

Benenson et al.'s (1998) comparison of networks to multidimensional matrices lends

itself to the thought that some problems encountered by online grocers can possibly be

solved by viewing these problems with the assistance of multidimensional databases and

multidimensional visualization methods such as virtual reality. These thoughts should be

discussed in further research about online grocery delivery solutions. Indeed, albeit in an

abstract sense, Benenson et al. (1998) talk about discovering "visually meaningful

patterns" in multidimensional information about business networks. They give an

example of sales patterns that, they say, could be too complicated to interpret by visual

means if depicted on a one or two dimensional map. Moreover, this research cautions to

be aware of sub-networks that can exist within networks.

More in line with the intelligent transportation definition of networks, Zhou et al.

(2000) talks about modeling networks with objects. Under this model, objects that

embody both spatial and thematic information are used to show spatial entities on the GIS.

This literature says that most GIS transportation (GIS-T) applications use a feature-based

model to show roads and other pertinent entities. Features are like classes in object-

oriented programming and modeling. Therefore, GIS-T applications should go beyond

strictly representing features, and use the objects that result from the features. This is

similar to the widely accepted object oriented modeling and programming paradigm. By

using this object oriented approach, transportation networks can be more holisticallyy"

depicted because each entity such as roads, points, intersections, etc, can be objects that

contain the data and functions that work upon the data (Zhou et al., 2000).

Whereas Zhou et al. (2000) recommend the use of "virtual networks" to better depict

multi-modal traffic network analyses, virtual multi-mode networks exist by sharing nodes

and sharing points in a network that does not necessarily reflect the exact shape of the

real multi-mode transportation network. The virtual network abstractly shows the

topological relations of the multi-mode network. Evidently, this is a more efficient way

of representing multi-mode networks, as the representation of multi-mode networks, but

according to Southworth et al. (2000), this is very difficult. Actually, Southworth et al.

(2000), while attempting to model intermodal (similar to multi-modal) freight networks

discovered that it was easier to model these networks without using GIS! Online grocers

may find that a multi-modal delivery system consisting of suppliers trucks, company

trucks (or cars), and even possibly bicycles or employees that walk the groceries to the

nearest customers. A delivery model where these "walkers" and/or bicycle delivery

employees might even meet delivery trucks at predetermined locations to more rapidly

deliver to the "last mile" of customers may be depicted using multi-modal classifications.

Although I do not necessarily suggest this type of break-of-bulk delivery strategy in this

dissertation, the possibility of modeling multi-mode logistics might make it more feasible

to use simulation to better ascertain the cost effectiveness of such last mile break-of-bulk


Southworth et al. (2000) claim that because most GIS packages depict networks

geometrically, the use of these packages to show multi-modal logistics models is

practically useless. This literature says logical intermodal route representations are more


Data Visualization and Data Representation

A scientific discipline exists called "visualization in scientific computing" that

contains collections of methods that are used to perform high-definition simulation

(Bernard, 1998). The existence of this discipline demonstrates the current emphasis on

the importance of data visualization. Gahegan (2000) restates the definition of

visualization as "the process of creating and viewing graphical images of data...with the

aim of increasing human understanding." He also lists visualization as an important

means to achieve collaboration by experts (Gahegan, 2000). Spatial data visualization,

especially when adopted by companies that use integrally spatial strategies, such as

routing and delivering groceries, should put significant emphasis on adopting

visualization technologies to make operational, managerial, and strategic decisions.

The value of clear inter- and intra-corporate visibility is articulated well by John

Fontana (2002), He said, "Companies must learn to read their supply chain's a gauge of the health of their business." He goes on to tell how a

reflexive phenomenon occurs when something goes wrong in the supply chain (Fontana,

2002). This occurrence, often called the bullwhip effect (Turban, 2002) should be

detected as soon as possible to preempt undesirable ripple effects throughout an online

grocer's interior supply chain. Some undesirable outcomes from the bullwhip effect could

be overstocking or stock-outs in peak periods. The bullwhip effect can cause operations

to stock unnecessary inventory to hedge against unforeseen online spikes in demand.

Robert Malone (2002), executive editor of Inbound Logistics magazine, said,

"companies need greater visibility along as much of the supply chain as possible." Clyde

Witt (2002) supports the notion of the value of supply chain visibility. His article about

Ford's six-sigma quality control program tells how visibility along the supply chain is

important in achieving six sigma logistical operations. If a supplier fails to deliver on

time, or throughput on the factory floor drops below a certain amount, the proper people

can be notified, creating a clear picture of the events that occur throughout the supply


Jan Bowland, managing director of KPMG Consulting, agrees with the notion of

overall supply chain visibility as an important aspect of logistics. She asserts that

shipment visibility is rapidly gaining recognition as an important priority for shippers,

particularly inter-modal carriers (Kuhel, 2002). Bowland adds that the value of supply

chain visibility goes further than just the awareness of what product is at what location at

what time. Full spatiotemporal knowledge should help to alleviate the necessity for safety

stocks which are kept in case of emergencies. This is because emergencies are less likely

to occur if full visibility into the supply chain is attained.

Despite the relative awareness of the importance of visibility into supply chain

operations, the amount of executives that are actually using sophisticated spatiotemporal

monitoring models or spatiotemporal monitoring technologies is surprisingly low.

According to John Zipperer (2002), less than 10% of senior executives surveyed properly

track supply chain performance, less than 7% collect correct or relevant information, and

less than one third of the executives track supply chain performance beyond the home


According to Zipperer (2002), telephones and fax machines are still the preferred

method for many buyers to send orders to manufacturers, which creates problems such as

unnecessary errors. The article goes on to mention some outstanding problems with

supply chain monitoring, while noting that the problem of lack of visibility throughout

the chain is the most important one. Because the visualization of the internal supply chain

of online grocers is of primary importance to pare unnecessary efforts, the lack of

awareness of visualization solutions must be addressed when incorporating a grocery

delivery plan.

Cartographic animation is valuable when it results in more intuitive judgments.

Using cartographic animation, users can see "geospatial transitions as they happen in

time, as opposed to simply viewing the end states" (Ogao, 2002). According to these

authors, cartographic animation can allow spatial phenomenon to be viewed holistically

instead of disparate "instances of time." The increase of bandwidth because of optical

fiber networks can facilitate the rendition of three-dimensional rendering technologies

such as virtual reality markup language across networks (Shiode, 2001).

Goodchild (2002) states that transportation research requires many models but

comparatively few types of data. The data that is used, however, is chiefly geographic.

Defining the combination of GIS and transportation science as "GIS-T," Goodchild

(2002) elucidates the three viewpoints of GIS-T, the map view, navigation view, and

behavioral view. Bergougnoux (2000) supports this standpoint and adds that these three

viewpoints can help researchers gain a perspective on the development of GIS-

Transportation (GIS-T) solutions. He also refers to GIS-T as "a significant area for

research and development." Goodchild (2002) uses nodes and links to describe the map

view. This is similar to Keenan's physical links and logical links. But Goodchild (2002)

notes some flaws with the link/node view such as features that overshoot or undershoot

the endpoints. Also, because streets are one-dimensional lines, information pertaining to

the respective sides of the streets is lost. Interoperability between spatial databases is

another problem. These undesirables are characteristics of the map view.

The navigation view "requires a massive extension of attributes provided in the map

view" (Goodchild, 2000). These attributes include factors that show hindrances to the

flow of traffic such as turning restrictions (for example, illegal U turns) and one-way

streets. An inhibitor to successful representation of streets in the navigation view is that

collecting data about the flow of traffic and representing that data in lane views is more

expensive than just showing single line, one dimensional streets.

The behavioral view adds the time dimension and takes into consideration how

discrete transport objects such as vehicles, boats, trains or people move on the network.

Goodchild (2002) says that all the different methods of representation necessary in the

behavioral view are not yet included in any one GIS package. Whether the logistics

model used by an online grocer needs the navigation view or the behavioral view can be a

topic for further research.

Research performed by Winter (2002) underscores the point made by Goodchild that

the representation of turning restrictions are an important aspect of the navigation view.

According to Winter (2002), every turn has a cost to it such as time used to decelerate

and then accelerate, along with waiting time; and these costs should be incorporated in

any representation of a transportation network used for logistical purposes. However,

incorporating turn costs in a network is expensive because they greatly increase the data

necessary to represent the graph, and makes it necessary to create more sophisticated

routing algorithms (Winter, 2002). One particular point made by Winter (2002) could

especially apply to online grocers delivery plans. He states that a circular tour is when an

entity leaves a point, makes a trip, and then returns to that point. This is of course what

the grocery delivery truck will do, so estimating the cost, amount, and location of turns

that will facilitate, and not hinder, the delivery truck making a circular tour back to the

main distribution center should be important factors in the overall logistical plan for

grocery delivery.

Data modeling features of GIS should be improved to allow better user interfaces

such as "exploratory" data analyses that allow users to progress through non-

predetermined paths (Spaccapietra, 2001). This functionality would obviously be useful

for grocery delivery.

West and Hess (2002) explain that metadata pertains to both the technical and

business aspects of a business, and that the business metadata is especially useful to the

end users. Much of what West and Hess (2002) say about the importance of intelligent

agents and metadata for end users is reflected by Tsou and Buttenfield (2002) as they

discuss both of these aspects in the context of distributed geographic information services.

According to Tsou and Buttenfield (2002) two type of metadata are required to

implement geographic information services (literature pertaining to GIServices is below)

- operation metadata and connectivity metadata. Operation metadata facilitates the

representation of cartographic information, such as coordinate projection and spatial

footprints, across heterogeneous networks. Data connectivity metadata specifies data

connection protocols such as java database connectivity (JDBC) and open database

connectivity (ODBC) that are used to access spatial data across the networkss.

According to this literature, the use of both operation and data connectivity metadata

enables geographic objects to be "more accessible, self-describing, and self-managing.

Dashboards are tools used to quickly and concisely visualize trends in corporate

performance by employees that have responsibility to monitor those trends. Often

referred to as digital dashboards (Ricadela, 1999a), these tools send important operational

indicators to the computers of corporate decision makers. Some of the uses of dashboards,

as stated by Whiting (2002a), are mitigating threats, highlighting business opportunities,

and depicting performance figures. The underlying technology used in dashboards are

data warehouses that amalgamate information from disparate data stores. According to

Whiting, GE is a prominent user of dashboard technology. The company incorporates

uses a corporate-wide dashboard called the "cockpit" (Whiting, 2002a).

Dashboards are intra-network or internet-enabled business intelligence tools. A

difference between dashboards and portals is that dashboards seem to be more task-

exclusive, whereas corporate portals can include links to a diverse amount of information

sources. Nevertheless, dashboards are becoming popular and are worthwhile

investigating as an integral part of GEOS-enabled online grocer logistic plans. Also, the

term "dashboard" is appealing because it connotes control mechanisms in a functional

layout. The incorporation of spatiotemporal "instruments" into dashboards such as the

depiction of delivery zones could be useful for online grocer logistics strategies and profit


PeopleSoft and Microsoft are two companies that have begun developing dashboard

solutions for businesses (Ricadela, 1999a; Ewalt, 2002). These applications, often

classified as business intelligence solutions, show potential to be incorporated into online

grocer logistic plans. According to Ewalt (2002), dashboards can give front-line

employees better decision making power. An integral part of a GEOS should be some

type of dashboard that shows the color-coded regions to be delivered to changing

dynamically. At times, operations managers might have to incorporate their judgment

about what route a dispatch should go on, especially if the dashboard-enabled GEOS

shows that two routes exist, but only one driver is available at the moment. An instance

of managers using their GIS-assisted judgment to dispatch delivery trucks is a reflection

of Thrall's assertion that, "GIS is one part of a larger information technology that may be

drawn upon to improve our judgment" (Thrall, 1995b).

Some similarities can exist between dashboards that contain GIS and geographic

information portals. Indeed, some literature refers to portals and dashboards

synonymously (Ricadela, 1999b). A differentiating factor between the two technologies

could be that geographic information portals include smart maps. Many aspects of smart

maps are similar to the functions that I said should be contained in GEOS in my

geography master's thesis entitled "The Conceptual Integration of Geographic

Information Systems into Enterprise Resource Planning." Those functions essentially

facilitate the retrieval of information pertinent to an area selected on a map. Called

"center pieces of graphical information portals," smart maps can show information not

usually included in a GIS such as real time information and interactive regions. As an

example, when a user places a cursor over a region of interest on a smart map, an

associated "subject tree" or pop up menu appears, providing the user with choices

germane to the region (Peinel and Rose, 2001). If an information source contained in the

tree is selected, a file will be downloaded from a web browser. According to Peinel and

Rose (2001), geographic information portals can help non-GIS professionals use relevant

spatial information because of the intuitive, visual nature of the information retrieval

aspects of the information trees. Their literature explains that logistics is an application

area where geographic information portals can be utilized. The real-time information

depiction qualities of smart maps might be suitable for the representation of online grocer

delivery zones.

Geographic Data and Surface Representation

Pienarr and Van Brakel (1999) summarize the value of data contained within GIS by

saying, "the value of the GIS is dependent on the quality of the data contained within the

system, undoubtedly making it the most important component of a GIS."

It could be financially prohibitive for online grocers to generate their own data about

roads and customer profiles at least during the first few years of operations. This is why

online grocers should be aware of the various types of geographic data available to them.

Pienarr (1999) divide online geographic data sources into five categories. They are,

educational, commercial, information providers, interactive mapping, and online data

searching. Educational resources include universities and other schools; businesses and

organizations sometimes provide GIS data, often on a subscription basis; information

providers include online GIS journals. Generator services, map browser services, and

real-time maps comprise the interactive mapping geographic data source of data. Online

data searching providers can have search capabilities using metadata, and permit transfer

of data by FTP (Pienarr and Van Brakel, 1999).

Declarative knowledge exists within metadata. When end users have to choose

between hundreds of map themes, declarative knowledge can help with their choices

(West, 2002). This could be especially important if the end users are not GIS

professionals, which is often the case when the average manager or end user is not

trained in cartographic principals, which are necessary to create professional GIS

coverages. Literature by Miller underscores this point as he discusses the potential for

improved geographic representation in spatial analysis through the use of GIScience

(Miller, 2000a). Spaccapietra (2001) says the type of end GIS user is changing. Most end

users are not GIS specialists, and GIS has not become interactive enough for

inexperienced GIS users to use quickly (Spaccapietra, 2001).

Swink and Speier (1999) also highlight the fact that the effectiveness of SDSS is

dependent upon the cartographical representation of the data. Swink and Speier (1999) go

on to tell that because of the increasing complexity of business data that must be

incorporated into maps (such as the data that pertains to logistics and grocery delivery)

the level of cartographic detail increases greatly. Many times it is impossible to show the

complete level of detail that is available, so GIS must be used judiciously to show data

patterns that help users make better decisions. Essentially, as the number of data points to

be considered increases, the complexity of the decision at hand increases Swink and

Speier, 1999). Despite this, Swink and Speier (1999) have discovered that better

transportation-related decisions are made when more customer zones are included in the

analysis instead of less. Results vary, however, if the decision makers have varying

degrees of "spatial orientation skills."

Also worthy of consideration is literature by Bittner and Frank (1999) who argue that

although most of the spatial representations rendered by GIS are currently based upon

analytical geometry, other ways of representing geographic space might be considered.

Most relevant is their proposition that "constraint-based" GIS representation be

considered. They also say that language can be a way to represent formal models on a

GIS. This too is noteworthy because the results of the calculations for delivery zones can

be expressed in words superimposed upon color-coded zones.

Somewhat supporting Bittner and Frank's (1999) research is research by Marble

(2000) who states that GIS inadequately uses the spatial tools available to it. Calling

current GIS-related computational approaches "myopic," he says that the dimension of

time has often been omitted in spatial analyses (Marble, 2000). Granted, since the year

2000, more literature has surfaced that pertains to spatiotemporality; but Bittner predicts

that GIS and spatial analysis are on the "brink of a major revolution." In a sense I agree

with him because much of the utilization of GIS to determine optimal (profit

maximizing) routes for dynamic online businesses like grocers has not yet begun to be

realized. This is an important point that is emphasized in my dissertation. Miller (Miller,

2000a) further expresses the inadequacies of current geographic representation through

GIS by saying that the inception and rise of GIScience might herald in a "comprehensive

re-examination of geographic representation in spatial analysis" (Miller, 2000b). Much of

his assertions are based upon the premise that because computational power is increasing,

so should the power of decision makers who use this spatial information. An example of

the utilization of computerization to help make spatial decisions is the SpaceStat and

DynESDA extensions for ArcView, which perform sophisticated statistical analyses, and

depict those analyses graphically (Anselin, 2000). Anselin's (2000) research also says

that much of the spatial analytical capacity of applications, especially if those

applications can be employed by mainstream GIS packages such as ArcView, MapInfo,

or ArcInfo, will allow non GIS experts to better perform these spatial analyses. If this is

true, decision makers in industries such as online grocery will benefit from the increased

ability to perform complex statistical spatial calculations to ascertain profitable deliveries

and more. Also, possibly as a way to make end users of grocery delivery spatial

applications acclimated to these types of applications, the users and decision makers

could be trained first on more simplistic location-allocation programs such as NEWLAP,

that have intuitive menu-driven interfaces (Lindquist, 2002). Because data (such as the

data in relational databases) is tabular, an important aspect of GIS is that it can combine

tabular data with cartographic data (on maps). When one considers that much customer

transaction data is initially stored in tables, this functionality apparently becomes

valuable. The value of GIS is increased further when combined with lifestyle data about

customers in retailers' catchment areas. In a United Kingdom survey most respondents

preferred the MOSAIC lifestyle database, with ACORN the second-most preferred

(O'Malley, 1997). Further, when extensions to GIS networking software, such as those

that have location-allocation functions, are implemented, catchment areas can be more

accurately determined. Geographic location-allocation software assigns demand areas to

supply centers while simultaneously maximizing supplier coverage and minimizing the

travel time for customers (Figueroa, 2000).

O'Malley et al. (1997) state that the most important data for retailers are the data that

are generated internally. Online grocers can have one important advantage pertaining to

internally generated data, which is the generation of internet-generated data. All data

relating to purchases made online can and should be added to corporate databases to

allow customer analyses. Moreover, many of these analyses can be performed in real

time using procedures like online analytical processing. Whereas brick-and-mortar

grocers use analytics (more on analytics-related literature is below) to generate coupons

on the back of sales receipts to be used on the next purchase; online grocers can generate

incentives such as coupons to be used on the current purchase depending on what is in the

customers virtual online shopping cart at the time. O'Malley et al. (1997) support this

notion, saying that "the future is likely to see greater use of real-time data for


Geographic data is inherently multidimensional, and recent developments in

constraint databases are allowing better representation of multidimensional data.

Succinctly stated, constraint databases attempt to show infinite collections of points in a

finite number of dimensional spaces (Grumbach, 2001). An example would be to show

three-dimensional polygons as conjunctions of linear inequalities. Depending on the

stipulated constraints, query languages can perform operations on the points that

constitute the polygons.

Assessing the exact extent of catchment areas has always been problematic. The

reason for this is that the outer boundaries of geographic "sales cones" have fuzzy

boundaries that can only be estimated (Loffler, 1998). Sales cones are areas that radiate

away from the store, with the vertex of the cone representing the store (Christaller, 1933).

The outer range represents the farthest point at which a customer will travel to a store or a

store can deliver to a customer (King, 1984). This outer range is the boundary of a region,

where the geographic extent of the region is dictated by the distance decay influence on

the customers of a particular retail store that is located within the region. Thrall (2002b)

said, "The rate at which demand declines with distance to the retail center is known as the

distance decay of store patronage."

I claim that the shape of customer sales cones for online grocers will be different

than for strict brick-and-mortar grocers at that same location. Further, the sales cones for

an exclusively brick-and-mortar grocer will change if that grocer begins an online sales

and delivery initiative. Additionally, because of the data gleaned from georeferenced

online sales, the boundaries of the sales cones will become less fuzzy. Also, speaking of

borders of sales cones, it is relevant to note that a "zone of indifference" exists between

market areas where the attraction of neither market area is greater for a particular

customer (Akwawua, 2001).

Also, it is important for online grocers to calculate the decrease of the customers'

friction of distance that occurs because of the zero drive times, and how those non-

existent customer drive times affect the gravity model as pertaining to the "law of retail

trade gravitation," which is, the ratio of shares of turnover of two shopping centers or

central places is proportional to the ratio of their attractiveness and vice versa (Loffler,

1998; Thrall, 2000b).

Customer drive times should not be overlooked when developing an online grocery

delivery strategy. This is because brick-and mortar grocers' first criteria used when

ascertaining a stores prospective customers' trade areas is the customers' drive times to

the store (O'Malley, 1997). Therefore, if, through GIS, an online grocer can discover

areas where the residents' drive times to the nearest brick-and-mortar grocer are

comparatively high, the online grocer could use this information to cater to those

residents because they would benefit most by home-delivered groceries bought online in

comparison to people living closer to the store.

Loffler (1998) expects the relative cost of shopping trips to increase in the future

because of increased drive times to stores combined with greater fuel consumption. This

means that the value of purchasing groceries online should increase proportionately with

the increases in fuel prices and fuel consumption. Luoma et al. (1993) supply a somewhat

antithetical theory by saying that because of the increasing wealth of the customers, the

customers will drive even farther, regardless of the size of a shopping center that he or

she is going to (Luoma et al., 1993). Still another perspective is given by Borchert (1998),

who says that despite any improved customer spending ability, the amount spent for retail

goods increases only marginally. Although all of these postulates may be debatable,

online grocers should investigate and test these assumptions in their regions to better

decide upon marketing and delivery strategies.

Buor (2002) demonstrates how distance decay is relevant for medical services in

Ghana. It should also be mentioned that literature has been published that attempts to

refine some aspects of the gravity model. For instance, Hu and Pooler (2002) propose a

competing destinations model that tries to remove some bias from distance decay

parameters in conventional gravity models. The underlying proposition in the competing

destinations model is that an alternative location can "compensate for the bad features of

another alternative". Said another way, "destinations compete for the attention of

decision makers," (Akwawua, 2001). Luoma et al. (1993) support this by saying that

some models do not consider competing destinations that lie within a customers

"personal threshold."

The theory of intervening opportunities takes the competing destinations model a

step further. According to this theory, "the number of persons going a given distance is

directly proportional to the number of opportunities at that distance and inversely

proportional to the number of intervening opportunities" (Akwawua et al., 2001).

Akwawua et al. (2001) go on to say that some people use the internet to prioritize their

search before embarking on a route. Although there are competing destinations and

intervening opportunities within the regions) that a person is willing to travel, a person

will limit the destinations to which he or she will travel to those that are "perceived

clearly." Online grocers should capitalize on this assumption by having their web sites

outstanding in clarity and functionality.

Chen and Jiang (2000), while emphasizing the importance of integrating disparate

spatial data across networked systems, tell how an "event driven" model of

spatiotemporal database changes can be an effective way to utilize spatial decision

support systems (SDSS). The idea of using events as the basis for computer applications

is not new. Visual basic for applications (VBA) is an event driven programming language.

This means that an event, such as a mouse click, pressing of a command button, etc,

triggers the underlying code to run. Chen and Jiangs' (2000) literature, however, is an

attempt to use the event driven model to represent spatiotemporal data. The authors say

that representation of events in databases is a "hot research topic," and "this event-driven

approach ...provides a new way for simulating system workflow." Although their

research pertains to events within a land parceling SDSS, it is not difficult to extrapolate

from this literature how valuable an event based database schema could be for online

grocery delivery businesses. If a sudden or gradual increase or decrease of online

purchases occurs within two groups of geographically distinct customers, it would be

important to determine the causes) of this change in ordering patterns. Possibly, the

increase of online orders is occurring because certain sections of two adjacent regions are

serviced by two or more delivery trucks on certain days, decreasing delivery times and

increasing customer satisfaction. The event here would be the superimposition of parts of

two or more delivery drivers' routes (a union of area). The end "state" of the region

served by two routes simultaneously would be increased orders (possibly including new

customers). Therefore, the connection between the event and the end state, as shown by

the event-driven database schema, could show to management the importance of

periodically doubling up drivers in certain areas (possibly in areas of high-value

customers). Alternately, the correlation of events and end states, in this scenario, might

show management, after a cost / benefit analysis is performed, that the amount of

increase in customer orders is not sufficient enough to pay for the added expense of

doubling up of delivery trucks for that region or time period. For online grocers interested

in possibly initiating an event-driven database schema, Chen and Jiang (2000) list useful

operators (in augmentation to conventional SQL operators such as AND, OR, etc). Three

of these operators are the sequence operator (;) showing the sequence of events, the

periodic operator (P) designating an event that occurs periodically, the FIRST operator

- which designates that an event occurred before another event (Chen and Jiang, 2000).

Roddick et al. (2001) address other schema issues while talking about the

spatiotemporal receptivity (or non-receptivity) of databases. By using "schema

versioning," temporality can be added to a spatial database. This literature makes a

distinction between schema versioning and schema evolution by saying that schema

versioning is a special type of schema evolution. Schema evolution is the modification of

the database schema without any loss of data, while schema versioning is the ability to

query all data within a database "both retrospectively and prospectively." For dynamic

delivery zone related businesses, such as online grocers, knowledge of schema versioning

and evolution might be of some value. However, as Roddick et al. (2001) state, schema

versioning has not been incorporated well into spatiotemporal databases, but also say,

"spatial schema versioning would be a useful adjunct to many systems."

Spatially extended structured query languages (SQL) that use operators such as

overlap, direction, contains, and distance, are better at representing data, rather than

analyzing the data (Huang, 1999). Two more recent SQL efforts attempt to allow more

spatial analytic capabilities. They are SQL3 MultiMedia Specification (SQL/MM), and

Open GIS Simple Features Specification for SQL (OpenGIS SQL). SQL/Temporal is an

international standard for spatiotemporal data modeling (Peuquet, 2001). Many of the

functions contained in SQL/Spatial are similar to those in ArcView. Some operations are


Huang et al., the functions that create new types of spatial features are the most difficult

to define in SQL. SQL/Spatial can be run on a client / server architecture, with an

interactive front end for making queries and a SQL/Spatial server on the back end (Huang,


While considering employing any database schema technique, online grocers should

not forget that not all street databases are directly compatible with each other.

Interpretation of features (ontology) and level of detail are two aspects where

cartographic data can differ (Noronha, 2000). Objects such as turn lanes and traffic

circles might not be represented on all maps. Also, Noronha (2000) says that most

databases store addresses as ranges rather than individual address points. This clearly

could be a problem for online grocers. Another consideration is that if delivery drivers

will be navigating while using mobile cartographic devices (which they should be doing

in order to respond in real time to online grocery order changes), is that maps take much

longer to download than other types of representations. Bertolotto and Egenhofer (2001)

propose a more expeditious way to download maps, which allows initially only a partial

representation of the map that is being downloaded. This literature explains the

difficulties of segmenting portions of a vector map to be downloaded. These difficulties

do not necessarily apply to raster maps. One difficulty is that when maps are overlaid in a

GIS, the process of sending those maps over a network is computationally expensive.

Map generalization, which is decreasing a vector map's detail, is a "complex and time-

consuming process," (Bertolotto and Egenhofer, 2001). Nevertheless, this type of

problem must be solved by online grocers who want to send maps in real time over the

network to delivery drivers.

Surface representation of elevation also may have relevance to routing solutions. The

point at which a delivery zone meets a profitability threshold can be calculated by

incorporating some principles of fluid dynamics that use interpolated depressions and

relief of a map using a GIS. Atkinson says that measures of elevation are relatively

simple to apply to maps, and that this measurement "is of fundamental importance for a

range of applications." Moreover, supporting the assertion that profitability can more

intuitively, and perhaps more accurately, be depicted and calculated over a region is

literature by Chang and Harrington (2000) say that profit can form a "landscape." Their

literature goes on to discuss how different degrees of "fitness" can affect the profitability

landscape. If Chang and Harrington's (2000) concept of "consumer preferences" can be

considered analogous to changing consumer online grocery purchases, we can see how

the spatial representation of profitability might be worthwhile endeavor.

Etzelmuller (2000) discusses different ways of showing changes in surfaces by using

grid-based digital elevation models (DEM). By using DEM, it is possible to quantify

surfaces to compare the surfaces for aspects such as roughness, changes to the surfacess,

and "noisier" or less noisy surfaces (Etzelmuller, 2000). Some of this calculation is

performed by assigning wavelengths to surfaces, and having the higher amplitudes

representing relief, and wavelength valleys representing depressions. The DEM

calculations result in a coefficient that when equal to 1, represents small topographical

changes, and 0 or a negative number means larger topographical changes. This type of

allocation of coefficients to topography could have relevance to the "terrain" transcended

by grocery delivery vehicles. Therefore, this research deserves further consideration

when developing logistical applications for businesses.

Additionally, besides just quantifying surfaces in DEM using Etzelmuller's (2000)

wavelength technique, data pertaining to various features can be overlaid on digital

elevation models. This might be useful in a trough and relief method for calculating

profitability thresholds.

I propose that "conventional" representation of polygons on GIS are not suitable for

many business decisions, such as those decisions that must be made by online grocers

about most profitable routes. Cressie et al. (2000) reinforce this idea by stating,

"polygons are typically counties, health districts, or states, which are politically chosen

entities that often have nothing to do with the etiology of the phenomenon." In many

ways, using predetermined polygons for routing profitability analysis is too constraining,

and cannot lead to accurate determinations of the value of any particular route. Also,

attributes related to polygons change all the time. Stagnant cartographic polygons may

not be able to adequately show these changes. Indeed, Cressie et al. (2000) say that

causation of events can be better determined when data relating to polygons are modeled

according to time.

The literature pertaining to fuzzy points, fuzzy polygons, and spatial uncertainty is

provided because of the possibility of applying these types of uncertainty measurement to

online grocery delivery systems.

Besides the actual representation of polygons, the relationship between the polygons

and the points (customer nodes, delivery trucks, etc) must be considered. When some sort

of non-definitiveness or imprecision exists about the nodes within a polygon we can say

that the points exist with some sort of uncertainty (Leung, 1997; Morris, 2003; Robinson,

2003). In essence, fuzzy sets add a dimension to Boolean algebra that proposes every

proposition is either true or false. Fuzzy sets add a "middle," subjective possibility called

a membership function (Robinson, 2003). These membership functions can be shown on

an x and y graph as triangular, trapezoidal, or Gaussian. Fuzzy sets can provide an

infinite set of values that apply to variables (Morris, 2003). In other words, the degree of

membership of a variable to a class is expressed rather than the probability of

membership of a variable to a class (Peuquet, 2001). Robinson (2003) predicts that as the

sophistication of GIS increases, the need for human input for the definition of the fuzzy

set function should decrease.

We could consider grocery orders from customers within a region that come in to the

online grocer with no determinable pattern (completely stochastic) as random orders.

This could lead to decisions to be made under uncertainty. Some of these decisions would

be to have a buffer amount of inventory, or have extra drivers on standby in case

forecasted orders actually come in. Decisions made under uncertainty relate to fuzzy set

concepts, which have been attempted with GIS (Leung, 1997). A point or polygon

represented on a GIS can be considered fuzzy if either object's shape or location is

imprecisely recognized. Basically stated, a point or polygon can have degrees of

belongingness to sets of properties. The body of literature pertaining to using GIS to

make decisions under uncertainty is growing, however, the literature pertaining to

allotting degrees of fuzziness to points and polygons is limited. Robinson (Robinson,

2003) stated that although the use of fuzzy sets in GIS has grown in the last decade,

commercially available GIS that support fuzzy information processing is rare. The above-

cited authors' research can possibly be applied to grocery deliver zones, especially if the

shape of those zones are considered random or fuzzy. Moreover, because fuzzy sets allow

the meanings of natural sentences (that contain indefinite words like near, where, how,

large, and small) to better be projected on a map, the advancement of fuzzy sets in GIS

seems inevitable, and the online grocers that become early adopters of the technology

may gain valuable first mover advantage in this highly competitive market.

Cheng et al. (1999) say that most natural phenomena are bounded by fuzzy transition

zones. Can the changes in customer catchment region boundaries be considered natural

phenomena? If not, Cheng et al.'s (1999) assertion could also apply to "unnatural"

phenomena. Nevertheless, since online grocers' delivery areas may change, sometimes

hourly, the conception of fuzzy transition zones could apply to these regions. However, it

is difficult to measure the changes in transition zones, especially since fuzzy transition

zones may overlap in contiguous regions (Cheng et al., 1999). According to these authors,

regions change shape during "epochs;" and the changes in the regions are measured by

comparing their size at different epochs, during which time they might have shrunk,

shifted, or expanded. This is called state transition. Peuquet (2001) supports this view by

stating that regional boundaries, along with boundary categories can be fuzzy. This mode

of thought can be of use to online grocers who want to calculate to different degrees

about if a region is profitable or not before dispatching a truck.

Allan and Lowell (2002) propose another way of dealing with spatial uncertainty by

using "abjects." They say that all points, lines, and polygons on a map can belong to

several map classes, which denotes that the prominent map class to which these entities

belong is not always certain. Polygons whose classes are indeterminate are abjects.

Abjects that are surrounded by objects (which definitively belong to a particular class)

can, under certain circumstances be joined or merged with the class, thereby becoming

part of the object itself (Allan and Lowell, 2002).

Funamoto (2000) endorses the notion of multi-level "clumps" of points that exist

within geographic polygons. According to him, geographic clumps develop because of

various phenomenon, such as concentrations of cancer victims within a region. Defined, a

clump is a cluster of points, with each point having a circle of a certain radius emanating

from it. The "clump radius" is the collective radii of the circles. What this research seems

to be is another way of depicting geographic concentration, which could be of value to

logistic models. Funamoto's (2000) research is noted here because of the potential of

incorporating his theories into online grocery delivery models.

Clumps should not be confused with clusters, however. Clusters are the much-

debated macroeconomic theory by Michael Porter about how related businesses "cluster"

around each other (Martin, 2003).

When geographic data is represented in layers, as is commonly done by a GIS, the

resulting spatial entities (points, lines, and polygons) are quantified to arrive at certain

conclusions. Once the data is quantified, it can be studied by order, topology, or algebraic

methods (Yongli, 2000). An online grocer might transform visual cartographic

information into quantitative information by layering a coverage containing the distance

from the nearest distribution center with a coverage containing the amount of purchases

per week to derive a new coverage that can tell if closer customers order more often.

One way to show the quantification of map layers is through map algebra, which is a

way to sample continuous space, and show those samples on a template that contains

rows and columns. Interestingly, algebra, which is most often associated exclusively with

numbers, is actually a way to "facilitate integration and cross-fertilization of abstract

models," (Pullar, 2001). In this case, the abstract model is a map.

MapScript implements map algebra using C++ class libraries which include

functionality for arithmetic, trigonometric, exponential, and relational operations on

templates, where templates refer to arbitrary neighborhoods. Groups of cells within a

template are also neighborhoods. According to Pullar (2001), by using the cell / template

representation of neighborhood values, a "large class of high level statistical and

mathematical operations" can be performed, including time-dependent analyses. This

form of modeling can be used to examine changes across space and time. Pullar (2001)

says, "travel time is related to the cost, or work effort, to move a certain distance." My

addition to this postulate is that since time essentially equals money, if the work effort

results in enough profit, then the cost of the travel will decrease or even become non-

frictional. The relationship of map algebra to the method I am suggesting is that instead

of conceptualizing neighborhoods using two-dimensional templates, such as those used

by map algebra, we can conceptualize the neighborhoods (delivery zones) as three

dimensional entities, incorporating the profitability variable (or other variables) as the

third dimension. This could be called map calculus, but since the online grocer delivery

zones could change, often in real time, fluid dynamics may be the more appropriate way

to conceptualize this regional quantification. As mentioned by Bernard and Kruger,

adaptive geographic grids can be used to simulate the movement of fluid. Also, if

cartographic representations are three dimensional, hydrodynamic movement may be

simulated (Bernard, 2000). Therefore, if innovative thinking is used, modeling methods

such as map algebra and fluid dynamics may be very important steps for quantifying

delivery regions for online grocers. It should be noted, however, that because three-

dimensional data is not readily available on the commercial marketplace (Koninger and

Bartel, 1998) online grocers might have to pay for expensive data collection if this option

is employed. Also, according to Koninger and Bartel (1998), there exists a way to

represent solids called constructive solid geometry, which uses Boolean operations to

make more complex solid objects from simpler three dimensional objects. Pullar (2001)

explains that "mathematical morphology" within map algebra is the theory for analyzing

spatial structures. Two morphological operators, dilation and erosion, change the region

under study by either increasing or decreasing the size of the region. Also, cells within a

neighborhood template can either die, be born, or continue to live, depending on the

properties of the nearest cells. It seems feasible that the dilation and erosion operators, or

some semblance of them, can be used to create delivery regions.

When incorporating map algebra, or virtually any spatial quantification method,

some thought should be given to linear referencing, which can render the necessity of

locating points on the earth's surface unnecessary when performing spatial

quantifications (Scarponcini, 2002). This is done by using anchor points, such as

benchmarks, which are georeferenced points used as initial reference points. Then, from

the anchor points arbitrarily spaced points are placed. The points are one-dimensional

entities used to specify locations. Once points are arranged in a route or network topology

from an anchor point or between anchor points, a linear data topology exists. One trait

inherent in a linear topology is that if an anchor point changes because of, for example,

the change in the location of a traffic intersection, the linear topology does not

necessarily have to change. The origin of the topology merely begins at a different anchor

point. A section bastioned by an anchor point is called an anchor section. Scarponcini

(2002) says that anchor sections are more stable than routes because routes become

unusable as the entire linear datum topology changes. This is not the case in linear

referenced topologies because in linear referencing, measurements relative to the anchor

points) are made when depicting the datum topology, as opposed to absolute

measurements such as measurements made to place mile markers at permanent intervals

from a state or county line, for example. When a route comprised of absolute

measurements changes, each measured point on the route must change. This is not so

with linear referenced topologies (Scarponcini, 2002).

Just as absolutely measured distances can cause problems when route changes occur,

spatial scale poses its own problems when using GIS to decide on best routes, such as

delivery routes used by online grocers. Even the definition of "scale" can be ambiguous,

meaning both "spatial extent" and "amount of detail" (Atkinson, 2000). An aspect of

scale that is important to online grocers is that entities shown at one scale can acquire

considerably different relevance when viewed at another scale. As an example, areas not

serviced might not be visible when viewed on a GIS in small scale, whereas these areas

might become highly noticeable when viewing a subsection of the original area at a larger

scale. According to Atkinson, heterogeneity and regularity of cartographic entities can

change when the scale is adjusted. The same grocery delivery routes that look elliptical at

a small scale might look polygonal at a larger scale. An exception to this phenomenon is

fractals. Fractals are patterns that do not significantly deviate when viewed at different

scales. Fractal patterns exist in both physical and human geography (Atkinson, 2000).

One way to decide on a scale appropriate for the job at hand, such as viewing the

most grocery delivery zones, while seeing the closest points any two delivery trucks are

at the same time, is to scale the map down, then scale it up until "robust" parameters are

discovered. Robust parameters are those parameters that are visible and relevant at all

scales within the study area. The area studied is also known as the "support" (Atkinson,


Adjustments in scale can also affect the way spatial dependence between regions is

viewed. The fact that households of certain LSPs (lifestyle segmentation profiles) live

closely together within a region can be lost if the regions are viewed at too small a scale.

Some of this information loss can be mitigated by kriging, or "smoothing" of information

from neighboring regions to adjacent regions.

Moreover, because scale can be divided into spatial and temporal realms (Maruca,

2002) the problems associated with scale become that much more complex.

Somewhat related to the above-stated micro-, macro- and meso-scopic traffic flow

simulation scales is micro-, macro-, and mesoscale scale representations (Bernard, 1998).

Mesoscale representations extend from a few to several hundred kilometers. Microscale

and macroscale representations are, of course, respectively less than, or greater than this


According to Haining et al. (2000), quantitative measures that are incorporated into

GIS packages are not sufficiently inclusive or sophisticated. Nevertheless, it is important

that online grocers become aware of some of the existing quantitative functionality.

There is no shortage of literature pertaining to this topic, although no writings could be

found that are exclusively focused toward online grocer decision-making using GIS-

enabled statistical processes. Although considerable breadth could be allotted to this

subject, an account of just a few of the more pertinent spatial statistical packages is

included in this literature review.

Spatial analysis in a GIS environment (SAGE) is a statistical package that can be

integrated with Arc/Info. SAGE works with polygon coverages and includes data

management, graphical drawing, querying, and classification tools (Haining et al., 2000).

Polygons input into SAGE can be "dissolved" to create coverages for sets of regions.

Linear regression, a common statistical method used by geographers and other scientists,

is enabled by SAGE. Ordinary linear regression, linear regression with spatially

correlated errors, and linear regression with a spatially lagged response variable are the

three types of linear regression that can be performed by using SAGE. Dynamic brushing,

the operation of quantifying the movement of a shape over a region, is not handled very

well by SAGE, however (Haining, 2000).

a spatial data analysis package that can be integrated with ArcView is called S-PLUS.

Besides being a quantitative system with more than 2000 functions, S-PLUS is also an

object-oriented language. By creating customized ArcView Avenue scripts, users can

increase the analytical capabilities of S-PLUS. Data transfer between ArcView and S-

PLUS is supposed to be relatively seamless (Bao, 2000). S-PLUS graphs include

boxplots, histograms, variograms, and multiple three-dimensional layouts.

Geographic Information Systems in Business

The literature cited leaves little room for doubt that GIS in combination with

decision-making tools and artificial intelligence will continue to be useful as a facilitative

system to allow optimal decisions in many business settings, including the online grocery

business. Business geography and "managerial geography" are gradually being

recognized as a means to a competitive edge (Risto, 1998). Dr. Grant Thrall (2000) has

recently published an important book about business geography, which gives greater

impetus to the recognition of the field of business geography as a bona fide aspect of

geography and an important aspect of business location strategy. Following is a review of

other literature that asserts the usefulness of GIS as a managerial tool.

Grimshaw (2001) says that the primary business benefits attained through scheduling

and routing systems is gained by linking geographic knowledge about customers to these

systems. Although there have been great efforts by small and medium enterprises to

integrate corporate-wide data, such as through ERP systems, most businesses have

ignored this "key dimension" of data. Businesses concentrate on the "what, how, and

why" of strategy but to a much less extent consider the "where" dimension (Heinritz,

1998; Grimshaw, 2001). One reason for this non-integration of geographic data is that

smaller businesses, such as many online grocers, do not have the technical or intellectual

resources to do this. Despite this, O'Malley et al. (1997) have found that 48% of survey

respondents (United Kingdom retailers) develop in-house spatial databases. Yet, these

authors have stated that retailers are "data rich...but... information poor" (Hernandez,

1998). Nevertheless, as Grimshaw (2001) says, because "benefits from improved routing

and scheduling go hand in hand with benefits from improved customer relations," the

geographic aspect of data should not be overlooked, especially by the online grocery

industry, which has low profit margins.

Perhaps one of the most important points Grimshaw (2001) makes is that geographic

knowledge about who the most profitable customers are. In order to better calculate the

profitability of each delivery route, georeferenced customer profitability variables should

be incorporated into the equation. Niraj et al. (2001) bolster the notion of the different

value of customers. Their literature also explains that the lifetime value of a customer

should be calculated. Although their research concentrates on the lifetime value of supply

chain partners, some of their findings can be applicable to the lifetime value of online

grocery customers. One thing they say is that the cost of serving each customer can be

different, thereby affecting the lifetime value of the customer (Niraj et al., 2001). This

becomes clear in an online grocery delivery context when variables like time to walk to

the back door, presence of mean dogs, and high crime neighborhoods are added into the

lifetime value calculation for individual customers.

Of the three categories of GIS stated by Brown (2000), the category of GIS as an

analytical system best fits the needs of online grocery delivery initiatives. As an

analytical system, the use of GIS has various benefits for business decision makers.

Hickman goes as far as to say that by converting large amounts of geographic and tabular

information into a more useful form, GIS "enhances every aspect of the business"

(Hickman, 1999). Most pertinent to this dissertation is the article's mention of the

usefulness of GIS to help track routes in real time.

As "early as" 1989 the value of GIS for business was predicted by some literature.

While reinforcing the fact that much industrial data have geographic dimensions,

Grimshaw (1989) said that retailers can "revolutionize" their marketing plans by

georeferencing customer data. Two of the four factors that he said would influence the

adoption of GIS in business are particularly true for the online grocery industry. Those

factors are better techniques for handling spatial data, and awareness of the spatial data

combined with human skills (Grimshaw, 1989). A main point of this dissertation is that if

online grocers handle their spatial data better, thereby resulting in the recognition of

zones that surpass profitability thresholds, even online grocers, with their notoriously thin

profit margins, could become successful.

In a later article, Grimshaw (1991) expounds upon his earlier literature and states

that GIS should be an integral part of overall corporate information systems and "strongly

linked to the business strategy. Indeed, this is the view that online grocers should take

when considering GIS as part of their logistical model. He also asserts that GIS should be

used as a not only an operational and managerial tool, but also as a strategic tool.

Although there could be debate as to whether he properly classifies the utilization of GIS

to facilitate the "delivery of information-based products or services" as a "strategic"

entity, as opposed to an operational entity, there is little doubt that GIS should be used to

perform the delivery function. Nevertheless, in the same article he asserts that

considerable value can be added to corporate information if the geographical aspect of

that information is included in the analysis. He expands his evaluation of GIS from

mostly a retail-based tool, to a more sweeping analysis of GIS in his more recent

publication (Grimshaw, 2002).

While considering GIS as a part of corporate strategy, Grimshaw (1991) says that the

use of the technology should be considered an opportunity rather than just a cost. Another

way to say this would be that there is an opportunity cost by not using the technology. In

the case of online grocers, if GIS technology is not used it may be infeasible to run a

successful delivery service, which means that by not using the technology the opportunity

to reach the large home grocery market will be lost.

Interestingly literature is still being published that calls GIS "new." For example

David Hakala's (2003) article entitles "Location: The New Killer App" tells how GIS

when used with databases can be used to make "powerful applications." This

phenomenon of calling two decade-old technology "new" means that there is still

significant ignorance within the business community about the potential of GIS as an

operational tool (Hakala, 2003). This was not the case with, say, enterprise resource

planning (ERP). Is the contrast between the adoption of GIS and ERP due to the

possibility that most people are geographically uninformed? Whatever the reason, the

stage is set for the adoption of GIS in industries such as online grocers because of the

need to view geographic information as a profitability enhancer

Despite the relative slow pace of adoption, GIS is gaining momentum in certain

sectors. Murayama (2001b) talks about the adoption of GIS in Japan. Interestingly, he

says that most of the editors of GIS-related academic journals are geographers. This in

itself shows the as yet lack of intellectual osmosis of GIS into other business-related

publications. One might think that integrally spatial fields of business such as logistics or

marketing would begin to publish spatially related journals. Evidently, this is not yet

occurring, which can also help to explain the lack of geographic solutions employed by

businesses in these sectors.

One reason that GIS is gaining popularity in Japan is that data is collected down to

very small regions. Also digital maps have been developed at a "rapid pace" since the

early 1990s (Murayama, 2001b). Also, according to the same article, point of sale (POS)

data is also being accumulated in large quantities in the country. Because of these facts, it

seems promising that this research can have strong implications to online grocery

delivery services in Japan. A countervailing situation exists, however. This is that Japan

has stricter privacy laws about the usage and dissemination of personal spatial data than

the United States does (Murayama, 2001b).

An interesting perspective about the use of GIS in cyberspace has been written by

Masanao Takeyama who says that geographers have just begun to study cyberspace, and

that the study of geographic "telepresence" holds important meaning for determining

"cyberplaces." Cyberplaces are places of "interactivity between cyberspace and physical

places (Takeyama, 2001). Getting a handle on the geography of daily or cyclical

cyberplace patterns may help online grocers decide between using home delivery or

communal drop boxes. Takeyama (2001) suggests that cyberplaces can cause spatial

metamorphism in cities. Online grocers that can detect these new spatial patterns might

be better able to cater to the growing market of mobile and stationary internet users.

O'Malley et al. (1997) tells that although store location is the "most critical decision

a retailer makes," the use of GIS to make such a decision by United Kingdom retailers is

not as pronounced as it could be. A reason for this is that the technology is not integrated

into strategic decision making to any great extent (O'Malley et al., 1997).

Other recent literature exists that supports the premise that GIS and spatially oriented

business practices are being adopted more. Lowe (2003) says, "The use of geospatial data

is moving from the mapping department in the back office to the existing business

practices of the enterprise." He also talks about supporting technologies that can enable

the non-GIS expert to better use spatial business solutions. One such technology is peer-

to-peer (P2P) computing which could allow any GIS application on a computer to

simultaneously be a client and a server. According to Lowe (2003), decoupledd peer-to-

peer spatial environments are removing old barriers that used to exclude non-technical

users." This means that it is becoming more feasible to initiate solutions such as the

delivery route system, which is suggested in this dissertation.

Despite the comparatively slow start GIS has had in taking a foothold in mainstream

business applications, the move toward adoption of GIS by science and industry is called

a "GIS revolution" (Yano, 2001). This literature says that there have been efforts to

increase the value of GIS by making it a geographic knowledge system (GKS). This

transformation has sometimes not been successful because some advanced quantitative

procedures that can be used with a GIS have not been adopted by mainstream businesses.

A factor that might be contributing to this relatively low acceptance of GIS is that United

States businesses have been influenced relatively little by academic geography. This is

because some of the general public do not fully perceive geography as being a bona fide

profession (Yano, 2001).

Enabling Geographic Information Systems Technologies

Because the purpose of this dissertation is to make suggestions that may increase

online grocers' profitability through strategic delivery zones, some review of literature of

the enabling GIS technologies will be of value.

For various reasons, a non-centralized GIS information systems platform can better

serve end users. One reason is that data that is not centrally located can be more readily

accessed by end users who utilize a distributed system. Also, it has been demonstrated

that nearest neighbor queries can be efficiently executed in multi-disk and multi

processor environments (Papadopoulous, 1997).

By the early 1990's GIS users where beginning to realize the redundancy behind

collecting and creating their own data sets, especially when they could relatively easily