top of page
  • Writer's pictureAdvanced Contextual

Topics from Google - Eureka!!!

By now we’ve all read about Google’s latest announcement about how they’ll target audiences in the post-cookie/3P data world. For once we find ourselves applauding Google for moving to topics. Topics have been a fundamental concept in our platform at Advanced Contextual since we released it.

Here our product and engineering team, all SMEs in topics and programmatic, provide their take on the Google announcement.

We have to initially point out that while Google announced the leveraging of topics, the topics are applied behaviorally. As a product and engineering team looking at a bidstream where 50% of the impressions don’t have an ID, we wonder if behavioral topics are practical for Google or for anyone else. People increasingly don’t want to be tracked and behavioral topics won’t change that rising tide. The newly proposed legislation backs that up.

We’ve learned and innovated a lot on topics and contextual understanding of content, and we have these observations and questions:

1. 300 topics isn’t enough to eliminate ambiguity, and Google knows that, but they’ll have a hard time expanding to a deeper taxonomy. Any of us who’ve been in search know how difficult it is to manage a portfolio of keywords and phrases to eliminate ambiguity. We have a topic model of 1,500 topics and we openly integrate other topic models from vertical experts like B2B companies. In expanding their behavioral topic taxonomy, Google will likely run afoul of their own benchmark of maintaining high k-anonymity. They won’t achieve cohorts big enough and they’ll be stuck at 300 topics.

2. Site-based vs page-based. Putting people into buckets based simply on the domain they visit is a surefire way to introduce ambiguity. Sites aren’t nearly precise enough to telegraph interests in any meaningful way.

3. Transparency in topic definition. Will Google make transparent the classification methodology they use to establish their topics? Transparency is key in establishing a robust topic framework. We know this and our customers know this, and as such we built complete transparency into our platform, including sophisticated ways to leverage KWs, sentiment, psychographics as well as first party audiences.

4. One topic a week from the last three weeks seems overly arbitrary. How do we know this? We map content consumption through our 18 month index of billions of open web pages and videos. For purposes of activation, we use all topics consumed in the last 30 days.

5. How will Google handle transparent audience extension providing brands and publishers with control? It’s not mentioned by them, but being able to extend an audience by matching non-endemic consumption to endemic consumption is table stakes for leveraging topics, which is what we do today. But it has to be done transparently and has to also be informed by a brand’s 1P data.

What does brand suitability mean for your brand? Recently, we noticed an explosion of really great travel keywords available in our ecosystem. When we dug into our programmatic analytics, we found that all of the instances of increased consumption of these keywords occurred within articles about the Afghanistan evacuation. This was a perfect example of having exactly the right keywords in precisely the wrong place. Learn how Advanced Contextual can solve for brand suitability here.

With FLoCs now Topics we think progress has been made. No one understood how FLoCs would be identified and it all seemed overly complex. But we wonder why they are spending any time at all on behavioral targeting. That doesn’t seem like a permanent solution given the way the wind is blowing on it.

And will they make all of this transparent? Of course they won’t and that’s unfortunate. If there’s one thing we’ve learned in building the best of breed contextual 2.0 platform it’s that transparency results in better scale, precision and performance. There shouldn’t be anything to hide from in contextual analysis.

Unless you have something to hide…

68 views0 comments
bottom of page