Power Levels of Context Metadata

Figuring out the system designs will be fun and the winners highly rewarded.

One of the most pivotal things to remember in this Second Coming of Context Targeting is that not all metadata are created equal.

Let's take Google as an example. As the giant bank vault doors started to close on ID targeting, Google came up with the idea of FLoCs -- meaning Federated Learning of Cohorts. Meaning that Google would create logical audiences and sell them en bloc and no one would need to know the IDs involved. In other words, the ultimate walled garden wet dream. Who can blame them?

It's really when one thinks about the post evaluation of ROI that the FLoCs idea becomes unappealing. ROI estimation methods (other than experiments) are already noisy enough with IDs let alone with FLoCs.

A long time ago Tracey Scheppach asked me to compare the value of zip-level vs. household-level targeting. She was shocked to see how poorly zip-level targeting performs compared to deterministic household targeting. TRA knew because we had both Experian data sets for car brand registrations. The correlations between the networks you'd buy were close to random between zip level and household level, except for very high-end cars.

One thing that will be lost as we shift weight from the ID leg to the context leg is the gullible use of lookalikes, which are so diluted as to be associated with single-digit lifts at best.

So now in response to marketplace demand, FLoCs are out the window. Google instead has come up with a new idea: Topics.

Each user will be automatically coded based on the three topics most looked at per week.

When I first got into the business this type of data was called "interests." Magazine publishers made great use of interests and so did direct mail. In the early 00s Dave Morgan's TACODA had the pole position on interests. It was measured based on behavior -- website visit behavior.

I was privileged, then as now, to be great friends with Dave, and in a consultation, he had no objection to publish the good with the bad. I found cases where context targeting was as good or better than interests targeting. There were categories such as video games where the interest in videogames was a gold mine, and a much larger array of other product categories where interests were only weakly predictive of potential purchase.

Many DMPs and others in today's game have been doing good work with "interests" data.

The specific approach that Google is planning, however, is going to have some sour consequences.

The user is going to be blitzed each week with hordes of ads directed to their latest interests. If you read my previous article in this series, you know that I consider the user bombardment phenomenon one of the worst side effects of stalker marketing.

What then should we be using for targeting if not lookalikes and not interests?

Targeting known purchasers of a product category has always been a good idea. Not necessarily targeting the purchasers of one's own brand. Byron Sharp and Larry Light/Mike Donahue will be debating the wisdom of targeting one's own loyal purchasers until the laws of physics prevent it. TRA proved to me and at least 77 major brand clients that targeting heavy swing purchasers (category heavy purchasers who have bought your brand at least once but have not been loyal) makes you the most money.

TRA did this with linear TV mostly and so it was all context based, rather than ID based, although in the CPG category, IRI, Nielsen, Catalina, 8451 all have frequent shopper card data so that HSP targeting can be executed in ID-based addressable media.

A stronger level of impact amplification is available when using the targeting of HSPs with the targeting of people who are motivationally resonant with the brand's current creative – the work of "my" company RMT with partner Semasio.

Outside of CPG where HSP proof has not yet been adequately produced, RMT's ability to identify people and contexts who are likely to respond well to a specific ad appears based on all of today's evidence to be the best metadata to use for ID targeting and for context targeting.

This is not to say that new forms of metadata are not coming that will add to the industry's ability to make profits on advertising investments using both ID and context targeting. Adelaide is one company that is building out from digital to all other media and using fact-based data (ad clutter for example) to calibrate a form of metadata which predicts the likelihood of attention regardless of the ad creative -- and someday hopes to include adjustments for that as well.

All proven solutions ought to be integrated together without regard to whether you invented them or someone else did.

Click the social buttons to share this content with your friends and colleagues.

The opinions and points of view expressed in this content are exclusively the views of the author and/or subject(s) and do not necessarily represent the views of MediaVillage.com/MyersBizNet, Inc. management or associated writers.