Challenging Behavior: Predictive Vs. Reactive

I had the rare opportunity to moderate a behavioral targeting panel discussion last week, for the Boston Interactive Media Association. The panelists included the usual heavy-hitting suspects, such as Tacoda, Revenue Science, and 24/7 Real Media. Yet, it was insightful and refreshing to see a few emergent players as well, such as Poindexter (which made headlines recently with additional venture capital funding and the hiring’s former chief scientist to further develop behavior-based solutions); Claria (with a recent BehaviorLink ad-network announcement); and wild-card entrant Blue Sky Factory (which promises site-side behavioral targeting with dynamic content delivery).

As the moderator, I not only had to challenge the panelists but also foster a vendor-specific learning environment for the audience. One primary discussion point was the difference between predictive and reactive systems of behavioral targeting. The philosophical distinction between the two, and the industry relevance and implications, morphed into this week’s column.

This point not only differentiates current tracking tools and methodologies between vendors, it really made me ask myself: is behavioral targeting about predictive intelligence of consumer behavior or dynamic interpretation of consumer actions?

Is the Past Really a Prologue to the Future?

The predictive model assumes your click-stream history and site visits represent your online behaviors. This translates into a “you are what you do” mindset. Consequently, on your fifth visit to, you’d likely be served a Chevy truck ad.

Wait a minute — are we forgetting something here?

The Internet has empowered consumers to conduct commercial transactions online. It grants access to an infinite world of knowledge. But it’s also spawned a generation with MADD — media-attention-deficit disorder. With a click of the mouse, consumers can go from Hotmail to Travelocity to iTunes to to eBay and back to Hotmail as they please.

What’s the behavioral pattern? I’m not sure. But I’m quite certain this abundance of online activities has somehow made profiling and segmentation more complex.

Marketers already know consumer interests are fickle. An iPod giveaway or other attractive value proposition can prompt them to take action. Seasonality, economic and political stability, and a gamut of factors can easily influence consumers’ online behaviors and activities. If we rely solely on past data, it will not only hinder our perception and understanding of the present, it will also limit interpretation of the future.

So the question is this: assuming the system can effectively monitor a user’s online click-stream and “behaviors,” can it detect changes in behavior and dynamically react accordingly?

The Quest for Artificial Intelligence

We’ve all heard the classic marketing adage, “Solutions are only as good as the assumptions.” For many behavioral targeting systems, the assumptions are:

  • Exhibited online behaviors will continue linearly, as will the consequent projection of future activities.

  • The duration of the tracking period (determined by the advertiser/vendor) is reflective of the lifetime value or product purchase cycle.
  • Modern-day consumer behavior is as predictable as peoples’ unending fascination with Macy’s Christmas window displays.

I don’t know about you, but the only assumption I’d bet on is the last one.

As consumer behaviors become more dynamic and complex, marketers will have an even a tougher time staying on top of trends. There’s a desperate need for targeting systems that can adaptively respond to these changes in the market. I’m specifically referring to creative tailoring and decision-making via artificial intelligence, with human input and rule setting.

Creative tailoring is the essence of advertising relevance. To collectively move the online industry to a more sophisticated level, smarter systems that can deliver relevant creative (which behavioral targeting already promises) must be in place so marketers can focus more on developing media strategies rather than on the tactical executions.

What Does This Mean for Online Media?

The fundamental difference between a predictive and a reactive model is the synthesis and interpretation of data. While the former bases future decisions on historical patterns and assumption, the latter represents smart adaptation of the evolving (or fickle) consumer behaviors and the marketer’s quest for the blissful state of marketing artificial intelligence. Undoubtedly, we need to migrate current behavioral targeting capabilities to a more predictive and reactive model for future growth.

A recent study found the average CPC (define) of keywords increased as much as 25 percent from Q1 to Q4 in 2004. A natural inflation of media CPMs (define) is also expected, due to the increased number of advertisers who are finally realizing online’s value to brands and making upfront buys similar to their offline counterparts. All these factors have already made advertising more competitive than ever in 2005. Advertisers are consequently forced to find other solutions to improve performance.

Behavioral targeting promises to minimize wasteful impressions and increase media efficiency by delivering more relevant advertising. It all boils down to real-time adjustment and creative optimization. I’m not sure whether I’m simply circumambulating a semantic differentiation between predictive and reactive, but I’m very certain it’s an important functionality distinction marketers must be aware of and proactively address. At the end of the day, if I could just initiate an insightful discussion between colleagues, I’ll have done my duty as a columnist.

Related reading