• by Phil Leggiere , March 26, 2008

Behavioral profiling has made dramatic leaps forward in ever more granularly segmenting consumers by interest, psychographic and other criteria. Re-messaging has ever more successfully brought behavioral targeting into real time. Yet aligning the historical understanding of a consumer's interests with understanding of their immediate intent remains a key challenge for online marketing, as Toffer Winslow, executive vice president of sales & marketing at ChoiceStream, explains below.


Behavioral Insider: What has ChoiceStream been up to in advancing your approach to personal relevance targeting over the past year?
 
Toffer Winslow: We've been working to expand the domains where personal relevance targeting can be applied. Up until recently our main focus has been on movies, music and TV. These of course remain important -- but we've gotten much more involved in the retail vertical as well. That makes sense because online retailers face the same kind of challenges as any large catalog vendor. How do you recommend the right content or items to the right customer at the right time?
 
The area of automated recommendation technology has attracted very smart people obsessed with creating the perfect algorithm. What's been missing is ease of implementation and use. In particular, "data feeds," the standard format of most solutions on the market, have been very complex to set up.

BI: How has your understanding of behavioral data evolved?
 
Winslow: Behavior is a complicated, multi-faceted phenomenon. What we're focused on is continually improving our understanding and enhancing discovery for consumers. What individual users do on a site within a single session and across multiple sessions indicates both their interests and their intent. That's a very complex intersection and one where behavioral targeting has remained fairly undeveloped.
 
Interests basically stay quite stable over time. Behavioral data can identify broad interests pretty readily and accurately. But inferring intent is more difficult. The learning curve of personal recommendations is to learn how to figure out how a personal interest profile relates to intent in a given situation. The challenge is to develop easier ways of taking interest profiles and predict what intent is RIGHT NOW. The cost of wrong recommendations is serious.

BI: Could you illustrate what you mean a bit?

Winslow: I'll give myself as an example. I spend a lot of time shopping for books and I'm an avid reader of historical fiction. But then say I start shopping for baby books because I'm planning to buy them for someone who has little children. It's an anomalous ‘off-profile' type of behavior for me which doesn't fit my defined interest. Yet my intent now is defined by looking for baby books. How should a recommendation system respond? If the system keeps serving me recommendations just based on the interest profile, it's missing the opportunity to sell to me based on my immediate intent.
 
Yet if I go back to the bookstore and start shopping again for historical fiction, I'm not going to want a slew of recommendations for more baby books. So an effective system has to differentiate between general interests and specific intent in context, and decide which recommendations make the most sense to the individual user at a given session.
 
BI: Where can you see this extended beyond the domains you mentioned?

Winslow: Whenever one has a broad catalog, the method of segmenting by personal relevance can become applicable. One area of interest in this regard is publishing. When you think of major content publishers with huge libraries, the challenge is to better connect users to items which will be of the most interest. We also see possibilities in the travel vertical as well. If you're interested in certain kinds of travel experiences you'll likely be interested in recommendations of related experiences. There's potentially an interesting play in advertising as well. If you as a Web site owner know what the interests and intents of visitors to your site are, you can not only recommend specific things to them on your Web site, but also serve highly relevant ads on 3rd party Web sites.

BI: What kinds of applications are on the horizon for personal relevance and other media?

Winslow: For most consumers the TV experience is very similar to what it was a decade ago. There's a long-term architectural upgrade underway. However, once you're looking at truly digital set-top boxes, personal recommendation engines can -- and we think, will -- become an integral part of the marketing mix.
 
The use of behaviorally based personal relevance targeting in storefront kiosks is something that will become much more common this year. There's great interest on the part of some major brick and mortar retailers. The way this would work is, if you have a store card and go into an in-store kiosk the recommendation engine will offer you items with a high likelihood of being of interest to you based on your in-store and online purchase history.  It will also assist your in-store experience by telling you what relevant products are in stock and what's in the store.

  ;