Some Context On Buying Context Programmatically

What will the programmatic world look like as third-party cookies start to decline? Not as different as you might think. That’s because programmatic platforms, in their little Silicon hearts, are trying to buy an impression that works with a client’s creative. You don’t need a cookie to skin that cat.

There will be two general forms of adaptation to cookie deprecation. One will be the use of alternate IDs or first-party data, both mapped into ID graphs. These alternative IDs will be privacy-friendly, but may lack some capabilities. To the extent that this approach continues to allow servers to serve ads to people rather than to contexts, there is not too much change. 

The other will be a return to the idea that the context of an ad is the best indication of the audience attributes (people who like plumbing fixtures go to sites about plumbing fixtures, blah blah). It’s old-dog thinking, but technology may bring some new tricks.



What’s interesting is that we have the opportunity to use a huge palette of technology to optimize the placement of creatives, impression by impression. As third-party cookies fade, platforms are left with a vacuum for data. How can they assess which creative should go to which impression?  The simple idea of using context is a great start at filling that vacuum, and we have 20 years of tech advancement to reinvent how context buying is done.

What tech advancements? Here are a few examples:

AI. AI can be used to assess the interests that a piece of content might relate to, ergo which creatives should be placed in it. AI can do this sort of analysis for speech, images, video, text, etc. AI, within the domain of a publisher, can examine how individual users behave and build profiles that reveal how content parts within a publisher relate. If you think about cookie deprecation as loss of signal, AI can help extract the remaining signal from the noise. 

Taxonomies. In the bad old days, taxonomies were rigid, incomplete, and had trouble with ambiguity, etc. But we have come a long way. In case you think this is academic, the core idea underneath Yahoo was to build a taxonomy of websites. That did pretty well starting almost 30 years ago. Now we have much thinner but faster systems of ascribing meaning or organization to unstructured information, across more languages. Better taxonomies will be required to make the most of matching classifications across placement and creative.  Try the demo at to get a more visceral understanding of this.

RTB systems. Some of these engines can evaluate bid requests at over 3 million per second. That is ridiculously fast by any standard of computing and communications, and implausible 20 years ago. Imagine being able to evaluate every context change on the web continuously.

Attention Measurement. Measurement systems tell us how humans responded to advertising or content. Imagine being able to look at how responses to different kinds of content correlate to outcomes, and using that data to feed buying engines.

So, the cookie model uses context to define a user’s interests, but is able to serve appropriate ads to the user wherever and whenever they show up.  The context model will similarly infer interest from context, but, lacking a user ID, will not be able to store that information for use across the web. People go across sites, but content generally does not.

Related, the idea that walled gardens will see no change is easy to believe, but not entirely true. For example, the ultimate walled garden, Facebook, took several cookies from any page containing a Facebook widget. Thus, Facebook had all the first-party data generated by behaviors on Facebook, and a huge amount of third-party data as well. No more.

To understand the new landscape from the point of view of buy-side platforms, remember that buying protocols use cookies, but don’t depend entirely on them. The RTB protocol itself allows for a huge amount of publisher-provided information about anything. Publishers can present any information they think will help a buying platform make a better decision. In a context-based world, this should probably include a set of references to a standard taxonomy.

Over time, programmatic has somehow become the blame-object (or guilt by association) for any number of industry maladies, for example the “ad-tech tax” and “fraud.” However, pricing structures, accountability systems, and fraud itself were all creations of humans (those present excluded, of course). The technology only does what it programmed to do.

So, as we re-architect the ad ecosystem, we should be learning from the past. I think you all know what happens if we don’t.

Next story loading loading..