The Category Exploration System:
Assessment
This chapter has described a computer system that performs useful work -- producing a review of a product category -- and does so in such a way that it is not easy to predict the report content ahead of time. Even though this system is really a prototype, it possesses many aspects of a real system. It is, therefore, instructive to evaluate the system (1) as a prototype system and (2) as a real system.
As a prototype, the Category Exploration system shows the viability of applying intelligent search techniques to the problem of evaluating a product category. Viability arises partially from the fact that the system uses actual marketing data and partially from its model of intelligent search. The system also shows a way in which report generation can be separated from report printing (or customization and printing) through an intermediate form, the word processor document. In both cases, as discussed in the Extensions section following, more work may be needed, but the current implementation provides a more than adequate solution.
A real system is capable enough to be deployed and used; the Category Exploration system is capable of (1) handling actual data, (2) outputing its results in the form of a word processor document and (3) using enough intelligence in the search to be able to finish a review. In short, the system is capable of getting work done for a user, where getting work done means producing a category review report that includes insightful material. Yet, as the following questions indicate, some features need improvement.
The initial CEx prototype was designed to produce the output shown in this document. After much discussion, it was determined that the form of the output could be improved to better match the need of someone who was trying to understand the performance of a category. Most of this improvement involves placing more analysis on each page. For instance, a page written about the category as a whole can present
Figure 21
The system does not have an active connection to a relational database, but its internal structure should permit one without much effort. It should be possible to set up for another database by reading meta-data about items, brands, sizes, etc., but since this system used only one database, this option is untested. Verification that the system will scale up from 1 market/5 chains to N markets/M chains needs to be provided.
Is the system capable of delivering an acceptable review?
The prototype system succeeds in producing "acceptable" contents of a review report, contents of the individual pages and analysis for the production of the written pages. Should any firm require adjustments to the system, the flexible design allows for a completely acceptable system to be produced by working closely with the client firm, its development staff, and the end-users within the firm.
Can there be more explicit control over the search process?
The prototype system has not made available anything, either in the form of parameters or rules, for influencing the search process. In other words, the search procedure is hard-coded. Perhaps an acceptable system would require more user control over the search process. For example, since the prototype system does not have any sense of when a situation has been adequately explored, perhaps a user should be able to request further searches and deeper drilling into the database.
How maintainable is the sytem?
A real system allows customization and interchangeability. The Category Exploration system is a prototype in which these features are supported to a very high degree, by virtue of the object-oriented programming environment (Smalltalk/V) in which the system is implemented. Customizing along many dimensions is possible and quite easy, compared to most conventional systems. Special case treatment by category, manufacturer, or retailer are possible, as are specialization of analysis methods, search methods, and the contents of individual pages, or the report as whole.
A working illustration of the concept of interchangeability is the manner in which output is generated. All examples of output shown in this chapter have been taken from a WinWord document, where the WinWord document was produced by the Smalltalk part of the system. A WinWord document is actually a special case of the more general notion of document that the system supports. This means that the Category Exploration system can just as easily produce a text file or a PostScript file with the same text, graphs, and charts.
Generalizations
One can generalize what we have done with the Category Exploration system in a number of ways. One way discusses a philosophy of "us versus them" that is encoded in the system. Another discusses the generalization of the entire system, by shifting from the consumer packaged goods industry to other industries.
"Us versus Them"
As introduced in the Conceptual Foundation section, the factor that most influences the evaluation of a product category is the number of noteworthy situations that can be found. The notion of "being noteworthy" refers to the definitions of "compared to the market" and "fair share." These two definitions reflect a philosphy of us versus them, where "us" is the retail chain being reviewed and "them" is the rest of the market -- the competitors.
There are many other ways in which "us versus them" could have been interpreted, some of which are meaningful in the current implementation. The table below lists some of these possibilities. The current system supports all the variations in which geographic locations appear; it is just a question of loading in data in an appropriate form.
Us Them
a chain all other chains
a market all other markets
a region all other regions
a store all other stores
a brand all other brands
a size all other sizes
a time period a previous time
period
Changing the system to reflect a new "us versus them" philosophy opens up some new possibilities for finding interesting situations buried in marketing data. In fact, it is possible to imagine a system that has two search levels --- the first, like the current implementation, and the second, which changes the nature of the first search by choosing the "us versus them" form that yields the most interesting results.
Application to Areas other than Consumer Packaged Goods
Most of the description of the Category Exploration system shows that the system is not tied very strongly to the consumer packaged goods industry. The general business problem can be stated as follows: numeric data has been collected about a number of objects (persons, places and things) in one or more locations (a geographic location, a situation or a perspective). Objects are described in terms of attributes such as shape, size, color, flavor, cost, desirability, etc. In the presence of a large number of attributes and values for those attributes, it is a fairly complex task to summarize the set of items in terms of the numeric data collected, and furthermore, it is a difficult task to decide which locations are most unusual, which subsets of items are most unusual, and how situations are changing over time.
This business problem describes a variety of situations. As an example, consider the problem of banks who provide automatic teller machines. They might want to be able to analyze deposit and withdrawal information as part of the transactions their customers make.
