Go, Do, Be.


12/15/2008  — 

Darwin's moth: a survey of behavioral targeting solutions

Edit: These notes are pretty old, but heck, I'll toss them out. Also, a note of fair disclosure: I own no stock in any of these companies.

Upon the discovery and study of Malagasy orchid (Angraecum sesquipedale) a beautiful flower with a preposterously specialized shape, Charles Darwin hypothesized that a moth with a tongue of equally preposterous specialization must exist in order for the plant to pollinate and reproduce. Darwin's hypothesis was confirmed much later when the Hawk Moth (Xanthopan morganii) was filmed doing its business.

It's a close, cause-effect relationship between the players. Imagine what would have happened if Darwin found the moth first: would he have predicted the orchid? The flower needs the moth, does the moth need the flower?

The web site parallels are clear: your site is a beautiful flower, its sweet content is the nectar attracting/enticing visitors. You study your users, learn what they want or need. You add a new feature, determine its success by the number of people that use it. If it's good, the people will come, even if you need to lead people to the good place.

This story is worth keeping in mind when examining the approach taken by behavioral targeting systems currently on the market. Their methodologies seem to fall into one of two categories: one group studies the pages on a site to make assumptions about the visitors to that page. The assumption is that you can generalize something about a visitor if you understand the content on a given page.

The other group studies users clicks and trends to make generalizations about groups clustered within a user population. To me, it seems like this group is studying the elusive, hard-to-capture moth. I recently had the privilege of reviewing some of the BT services. Our needs are pretty typical: we want to extend premium ad inventory by targeting ads based on user behavior. Content-targeting (non-ad) would be more of a nice-to-have. Our requirements are similar to most partner-integrations:

  1. Ease of implementation
  2. Ease of segment (moth) definition
  3. Forecasting segment size
  4. Targeting accuracy
  5. Dart integration
  6. A/B testing for auditing effectiveness
  7. COPPA compliance (we want this to be clean!)

Here's a brief survey of the top behavioral targeting services (email me if you think I should add another).

baynote - study the moth

  • http://www.baynote.com
  • 2 step implementation: first add an observer tag (dead simple), then add the content piece (slightly trickier)
  • groups users based on search terms and click paths
  • doesn't care as much about what's on a page
  • Baynote doesn't seem to have a dart integration, however it seems possible that baynote could serve a dart tag targeted to a user (which won't be too easy to manage in the long run).
  • proven track record with targeting content (think merchandising 'related items' on ecommerce sites)
  • cost: no idear. They have no model for an ad-targeting engagement, but they seem interested in exploring this. We would run a test with them if there's no up-front $$$.
  • Here's something they seem to do well: constant testing to evaluate performance.

Personifi - study the flower (and the moth a little)

  • http://www.personifi.com
  • simple integration: serve their tracking tag through dart. This is a great way of dealing with the integration. Let the ad operations folks manage the whole thing.
  • Personifi spiders the page and maps the content to Personifi's own taxonomy. To me, this is the really cool part: they'll classify your site's content for you.
  • DART integration! passes name/value pairs to dart. Target campaigns to behavioral segments
  • cost: cpm or revshare

LOTAME - study the flower

  • http://www.lotame.com/
  • implementation is 2 steps: add their tracking tag to collect information, then manually identify the clusters of users their system identifies. The manual cluster-identification seems like a lot of work. We need to learn more about this.
  • you can manually upload their classification results to dart for ad targeting, but they're working on automating this piece.
  • cost: cpm or revshare

DART boomerang - study the flower

Edit: This one may have disappeared already. Did it ever really exist?
  • No url. Not a lot of info on boomerang on the dart site. Sounds like they're gearing up for a 2.0 release sometime in '08 which will have a proper admin-interface.
  • Implementation: Drop a pixel or 'boom tag' on a page, associate pages with a 'boom list' or segment name, target ads to 'boom lists'. We need to know what pages (or conversion events) belong to each 'boom list'.
  • Interesting factoid: we could deliver a targeted ad to a user even if the 'boom tag' is not on the page.
  • cost: fee structure based on the number of 'cookies' or 'users' in 'lists'. We'll see what this really means.
  • Interesting factoid: Boomerang will work closely with ad exchange to provide some interesting off-site revenue opportunities.

I noticed that the wikipedia page features a bunch of european BT companies. WunderLoop and nugg.ad look interesting! I'll check them out at some point. Tacoda doesn't seem to be licensing their technology right now, so let's skip them altogether.

If I get some time over the weekend, I'll put together some really sketchy notes on what's really happening under the hood with some of these behavioral targeting systems.