top of page
  • Writer's pictureAdvanced Contextual

Audience Modeling in the Aftermath of the Third-Party Cookie

The third-party cookie gave us predetermined audience models most of the time. Brands and agencies trusted that a segment was about what it was named and that they could easily monitor their campaign’s performance and scale goals using those segments.  

This all goes away in 2024 when Google retires the third-party cookie. What replaces it is modeling from a smaller number of IDs for walled gardens and inventory modeling in digital. 

Audience modeling of converters is fairly straightforward. A brand has a customer segment that has achieved a certain KPI, and they want to expand their reach. That cohort can be applied to native lookalike models that exist today in the walled gardens. They’re good models and are built for the environment they exist in, and they help with very low-funnel KPIs. Think of it as the new retargeting.

But you also need to build prospecting models. That isn’t as easy as the lower-funnel use of converted IDs. Enter programmatic contextual. 

Advanced Contextual’s solution powers both audience and inventory models. It all starts with using topics to identify example pages that a prospect is likely reading or watching. Instead of domains or channel fronts, we prioritize article and video pages that represent topics.

We understand the topic of the content, not just the keywords . Because without precise content classification, you risk appearing on pages that are unsuitable for advertising. For instance, imprecise categorization could lead to your travel ads winding up on a news article about people fleeing a natural or manmade disaster. Our proprietary page reader carefully classifies content to prevent these mishaps.

After we put seed articles, or articles verified to genuinely represent given topics, into our platform and hit go, we ask our platform to find all the content that matches the seed topics from the 3.5 billion pages in our contextual index. And since we’re built in front of auctions, we also pull back any IDs attached to the content that has the right topic.

Our segments contain a list of pages as well as a list of IDs that visited that content. The IDs can be exported to walled garden models in a privacy-safe fashion so you can put prospects in the top of the funnel. As for the open web, because you know the exact pages you want to target, you can push those to your DSP and use our model to bid on those pages.

We can take this same data and use it to build audience extension models. By understanding the topic that led us to target a cohort of IDs for modeling and seeing what other topics those IDs visited, we can create a segment-level consumption graph. This is a simple query into our search engine with 3.5 billion pages: we ask the engine what other topics (and specific pages) a cohort of IDs visited in the last 30 days. We then receive a list of pages and segments ranked by comp index so you can see consumption concentration. This list can be ported over to DSPs for an audience extension buy. You can also take those same high comp index targets and model audiences using walled garden lookalike models.

In short, in 2024, the mechanics of programmatic will shift from largely deterministic targeting to probabilistic models. We can use context to build those models, both for the open web and the walled gardens. So, audience data may be going away, but scalable advertising performance doesn’t have to follow suit.

7 views0 comments

Recent Posts

See All

How Advanced Contextual Uses AI

Every adtech company says they use AI. But what really matters is what they can actually do with AI that they couldn't do otherwise and how it affects performance for advertisers. To understand how AI


bottom of page