What’s the right level of automation in content targeting?

I spoke with two web content management (WCM) vendors in the past week that are investing heavily in online marketing. In the WCM realm, this mostly means user-friendly tools that marketers can use to not only create content for a web site but also to test it and target it to site visitors.

A question I’ve been asking lately is, how automated can / should this targeting be? Interwoven, FatWire, Tridion and others have tools that let marketers segment visitors and then build some rules around how content should be targeted to these segments. This is a fairly manual process – segments and rules have to be manually created and managed. This is workable for sites that have a few broad segments and relatively shallow content / product catalogs. But wouldn’t be manageable with multiple, detailed customer segments and a deep product catalog.

This is one of the reasons Amazon, with arguably the deepest product catalog around, has long applied what we used to call “collaborative filtering” on its site — you know, the “readers who bought X also bought Y.” This approach has its own drawbacks to be sure (on Amazon, I get a strange list of recommendations based on the books I purchase for myself, for my kids or as gifts) but it wouldn’t be feasible for someone at Amazon to manually create cross-sell rules for every item Amazon sells.

A crew of start-ups like Baynote, Aggregate Knowledge and Loomia offer updated approaches to collaborative filtering that use more inputs (like time on page, search terms, clicks, scroll rate etc.) than early collaborative filtering tools. These vendors take different approaches (i.e., behavioral vs. contextual) but they’re similar in making recommendations automatically.

Some WCM vendors note that customers are leery of a “black box” making recommendations with live content on their sites. That isn’t surprising really. They also note that most customers are only beginning to segment customers or to get their feet wet with content testing (like multivariate testing to test layout or content success rates) and aren’t ready for automated recommendations yet. Still, Vignette just signed an OEM agreement with Baynote, so there must be some interest.

So which is the right approach? Ultimately both rules-based and automated targeting are likely to have roles to play. As emerging online marketing suites that include WCM, web analytics, testing and targeting tools come together, they’ll let the marketers choose the right approach for different types of content and/or different customer segments. But we aren’t there yet.