• by Kate Kaye , June 4, 2004

In a move that expands the applications of its ad targeting and optimization technology, Poindexter Systems, Inc. has released the latest version of its Progressive Optimization Engine (POE) software.

POE 3.0, a Web-based Application Service Provider (ASP) product, extends the audience clustering and predictive modeling technology for use with search, customer relationship management functions, and email. The new version also broadens POE's publisher-side applications--something that underscores Poindexter's shift in 2002 to serving both the buying and selling sides of the Internet media market.

Poindexter's POE technology segments audiences into clusters based on demographics, geography, time of day, and user patterns to optimize online advertising campaigns, and to determine which audience segments warrant higher media rates.

POE 3.0 features an auction-pricing model for publishers that builds on the company's PassBack technology. The PassBack system predicts which ad impressions are unlikely to result in satisfactory audience response by assessing user demographics, geography, and other criteria--and in turn, enables publishers to offer those impressions to another advertiser. This, of course, also allows publishers to boost the price of particular ad impressions.

Because the goal of Poindexter's advertiser clients is to increase return on investment by--among other things--lowering media costs, some argue that the increased commitment to helping media sellers raise rates could alienate some clients. "You can't play both sides for very long; you either work for buyers or you work for sellers," contends Dave Morgan, CEO of Tacoda Systems, Inc., an optimization technology firm that services Web publishers like Advance Internet and iVillage. "The market doesn't support Switzerlands."

Advertisers and publishers "are going to have to take the word of the company that's deciding the inventory is more valuable," comments Gary Stein, senior analyst at JupiterResearch. Serving both media buyers and sellers "is not necessarily a conflict of interest," he opines, "but there is a potential for abuse." In the long run, suggests Stein, increasing the value of preexisting technology is good for companies like Poindexter.

In addition to bolstering POE's publisher-aimed functions, Poindexter has added search-related applications to the software to assist advertisers in their search marketing efforts. Essentially, the system exports a set of instructions for targeting certain search keywords to audience segments, and helps marketers determine the value of specific terms. Other ad targeting and optimization technologies such as the one offered by Tacoda Systems integrate search by enabling publishers to target ads according to previous user searches.

POE 3.0 also allows advertisers to apply data to campaign components other than standard image ads, such as email and customer relationship management communications. "We're giving control of a piece of inventory to the marketing group instead of just the media department," explains Joseph Zawadzki, Poindexter's founder and CTO. The company serves advertisers in the financial services, automotive, and travel sectors, in addition to media clients like America Online and The New York Times Co.

The latest iteration of POE gives clients the ability to attribute a level of importance to certain data points collected over time. For example, an online florist may need to attach more or less weight to information gathered during holidays such as Mother's Day or Valentine's Day, depending on particular campaign goals. In addition, advertisers can customize campaign variables such as lifetime customer value, user behavior, or interests to better target offers. "It allows advertisers to define things that are important to their campaigns," Zawadzki continues.

The POE technology incorporates a variety of standard predictive methodologies including Bayesian Belief Networks, CART, and nonlinear programming--pitting them against one another to determine which works best for each specific campaign use. As information on user response is gathered, it is filtered through each model and tested against another set of data to isolate the most appropriate method. "Offline marketers have been using these models for years," says Zawadzki, admitting: "We don't care which [methodology] wins. What we really care about is what generates the highest lift."