But marrying dynamically targeted creative with that behavioral data continues to be a work in progress. Almost every piece of research I have seen over the years points to the effectiveness of leveraging more custom, targeted creative to behaviorally targeted consumers. But the cost of multiple creatives (or the crappy look of some dynamically created ads) still turns clients away. Anyway, just retargeting visitors to e-commerce sites by broad content categories is famously effective. But one provider, Permuto, has an interesting model for aggregating data from e-commerce sites into a more automated system of targeting people who are in-market for specific products with equally specific ads.

"We create a buyer's channel for display advertising," says Shaukat Shamim, CEO and co-founder. Like a number of other companies lately, Permuto claims to be bringing to display advertising the accountable performance metrics of search, and giving performance campaigns the reach of display. In this case, Permuto focuses on the e-commerce segment. "The e-commerce guys generally are not spending into display, even though it has reach," he says. Their focus on performance tends to lead them to search, "but search doesn't do targeting -- it does response."

Like traditional BT, Permuto is partnering with e-commerce to identify people in-market and track their behaviors to reach them elsewhere on the Web. But Shamim says the key difference is in the granularity of the data collected and the algorithms the company uses to determine consumers' in-market intent. "The difference between us and behavioral targeting is the difference between just knowing someone walked into Best Buy and knowing what product he is picking up and why he is spending his money, whether he is looking at reviews, etc.," he says.

In order to determine what products a user really is interested in and the level of intent, the system weights different actions variously. Having specific items in an abandoned shopping cart will have a heavy weight but only for a set period of time. Reading reviews of a product will have lighter weight -- but if married to comparison shopping behaviors and abandoned shipping carts, it will trigger high intent. "The predictive model we have built has data points from lots of shopping-related activities. We create scores for 65 million people," says Shamim.

Permuto puts those scored shoppers into an automated marketplace where advertisers buy traffic much as they do with search. In this case, though, they are buying against specific types of products, and the engine matches users' profiles with the actual product catalog from an e-commerce site. "We give [advertisers] access to people who have a high intent for particular kinds of products," says Shamim. "The e-commerce guy feeds those products into the system. They will give us their budget on a CPC basis, similar to how they do it in search. They use their product feed and we figure out where to put the display ads to convert for them."

The ads themselves are dynamically created with assets from the e-commerce catalog feed to target the interested user with relevant items and offers. "[Advertisers] generally don't get into what the creative will be," he adds. "They will give us different options to tweak." The system determines over time which ads are generating the highest yield. There are a number of variables that can come into generating the creative itself. The ad can combine one item like a camera with its corresponding lenses if the user has an intent for both. It can display multiple items in a kind of comparative shopping format.

According to Shamim, his approach brings more performance-based pricing into behavioral targeting. "In our world, return on advertising spend is the only effectiveness that counts.  Our customers look at the channel seriously as it gives them search-like return on ad spend," says Shamim.  "Where a typical BT solution leaves the ultimate effectiveness and ROAS to the customer, our platform puts performance first.  We are paid when somebody actually clicks on an ad on CPC."

Watch what shoe or vitamin you put in that online shopping cart before you abandon it. The same item may follow you in ads you encounter elsewhere on the Web. Of course, this could ultimately evolve into a good thing for consumers. One can imagine someday a perfectly transparent and ubiquitous behavioral re-targeting system in which the consumer can game the machine. If I knew I was being tracked by a system like Permuto's, I might intentionally fill and abandon a shopping cart in the hopes that the network would record my intentions and target an even better, highly personalized offer to me in my later travels.

Consumers can be pretty wily too if they know the rules of the game.

This commentary is insightful. I recommend it to others.