INCORPORATING DATA MINING IN A DYNAMIC-PRICING ARCHITECTURE
By Ed Colet
A recent phenomenon in this Internet age is the rising popularity of auctions. In contrast to buying and selling items based on a set price, auctions allow one to buy and sell goods and services at dynamically different price points. The ability for a retailer to dynamically set initial price points for their products can be a challenging task. The technical infrastructure necessary to implement dynamic price setting within an auctioning or trading application requires the interaction of several components. In this column, the architecture is described, and the ways that data mining components can be incorporated are suggested.
Buying and selling items via an auction is certainly not new; for example, auctions have been, for much of history, the exchange model for antique art. But instead of an exotic item being auctioned on site, today even commonplace items (e.g. desktop software, clothing, airline tickets), and services can be had through auctions on web sites. For the rare expensive item, an initial price can be based on expert appraisals and then the competition among buyers (demand) can establish a final selling price. For a wide variety of commonplace items (many of which are not scarce items) available via auctions today, setting an initial price is more difficult because there are multiple considerations. A successful dynamic pricing model needs to be scalable (based on the number of items), set close to real-time, and be flexible enough to accommodate a wide range of products.
The technical architecture to implement a dynamic pricing model so that it can be used for an auction application on the web usually consists of a modular system of components. PCWeek, 11/01/99. page 40 has a description of the type of architecture typically used. A central component is an application server utilizing Enterprise JavaBeans. There is a separate component for business rules (reflecting strategic objectives), a data layer component that stores object data mapped to an external RDBMS (Relational Database Management System), that stores the actual product data. Vendors typically supply templates to allow one to customize the interface(s) between the trading application and the back-end inventory systems. Connecting these components, even with the use of open APIs, is a non-trivial task requiring knowledge of database technology, trading applications, and accounting systems. And it's still possible at the end of the auction interval for a retailer to still be left with unsold inventory because the initial price point was too high, or to have sold all items by the end of the auction period - but at below optimal prices.
By capitalizing on the component architectures already in place, it's possible that data mining can provide a useful benefit to optimize the exchange of items. Data mining is already closely linked with relational databases. Useful patterns about who buys what and when are regularly detected via data mining of customer information and product information all of which are stored within RDBMSs. In the context of auctions, data mining analysis can then be used dynamically to set the price for opening bids and/or dynamically set prices. The other layer in a dynamic pricing model is the business rules engine. Here too, data mining can possibly be used to determine and refine business rules. In fact, the output of many data mining analyses are cast into the form of actionable business rules. Rules affecting how and when to sell items can then be based on data mining analysis. The end result is an exchange of goods and services that is optimal for both buyer and seller.
Ed Colet is the Acting Director of Research at Virtual Gold Inc., responsible for developing analytical methods for data mining and for investigating human factors and usability issues of business intelligence systems. At present, he is in the final stage of completing a doctoral dissertation in the Cognition and Perception program at New York University's Department of Psychology. Ed has also worked for IBM Research at the T.J. Watson Research Center. At IBM, Ed was a member of the group that developed Advanced Scout, the data mining application for NBA teams. His research interests focus on statistical methods and human factors.
For more information, see http://www.virtualgold.com.