By Tim Peterson • September 25, 2024 •
Ivy Liu
This article is part of Digiday’s coverage of its Digiday Publishing Summit. More from the series →
In July, when Google announced its decision not to deprecate third-party cookies in Chrome by default, publishers were faced with reevaluating their post-cookie plans in light of a post-cookie landscape not necessarily being guaranteed. Hearst Magazines was among those publishers, having introduced its first-party targeting tool Aura a month prior.
“We paused for about five minutes [after Google’s announcement], and we just said, ‘Oh.’ But the plans were in place, and we were already on the path of a first-party targeting tool. I mean, we all believe this is the right thing for us,” said Jen Dorre, svp of ad products and data at Hearst Magazines, on stage during the Digiday Publishing Summit in Key Biscayne, Florida, on Sept. 24.
Rather than reverse course as Google has done, Hearst Magazines has pressed on with its post-cookie plans – and now with the benefit of those plans not needing to be entirely post-cookie.
Google not doing away with third-party cookies in Chrome “means for us that there’s a little more seed data, a little bit of deterministic data” that Hearst Magazines can use to to build out its probabilistic models for cookie-less (meaning fewer cookies, not no cookies) targeting and measurement, Dorre said.
Because that’s ultimately how the landscape is changing. It’s not so much going from a cookie-based to a cookie-less model for ad targeting and measurement but from a deterministic to a probabilistic one. The trick, though, is having the deterministic data as the foundation for the probabilistic models and having thresholds in place to preserve the integrity of the underlying data signals as they are projected across a broader audience.
Case in point: Hearst Magazine’s Aura is rooted in deterministic data – its first-party audience – and then combines that with contextual data regarding the types of content that people consume across its properties. Aura uses that combination of deterministic and contextual data to project audience profiles across the rest of its audience base, for which it may not have data.
This type of projection can mean that advertisers have to make a leap of faith of sorts when targeting ads based on the probabilistic model. But targeting isn’t the only side of the coin turning probabilistic.
“We’re all trying to figure out measurement. And so I think of measurement in two parts: us as a publisher knowing that we’re delivering what we say we will deliver, and then there’s what the advertisers are willing to accept for measurement,” Dorre said.
One form of measurement that Hearst Magazines is exploring is attention measurement. Companies like Adelaide provide proprietary metrics for measuring attention from a sample of people and projecting those measurement across a publisher’s audience. Effectively panel-based measurement, this type of measurement is very much probabilistic and not ready to be transacted upon, but it’s part of the process for getting advertisers to adapt to cookie-less measurement models.
So are data clean rooms. Hearst Magazines is exploring data clean rooms as a means of attributing the performance of advertiser campaigns to the publisher. These clean rooms enable a publisher and an advertiser to match data to connect the dots on whether a publisher’s site visitor ended up purchasing a product from the advertiser. But that only works if both sides have that deterministic data by which to match the ad exposure to the sale transaction. Once again, enter a probabilistic approach.
“If you have a small seed, you start from there, and if you can join that, and then you could build models from that. So I still believe you have to have a small seed of deterministic data,” Dorre said.
How small of a seed? Generally speaking, Dorre said she has seen success when modeling against at least 10% of users and at least 100,000 ad impressions.
Those aren’t the only thresholds Hearst Magazines has put in place with regard to its cookie-less modeling. With Aura, the publisher is looking to keep the number of different audience segments to the range of 30 to 50. And on the contextual data side, it is trying to stick to three to five top-level categories with no more than five sub-categories apiece.
“Beyond five did not make sense…because we found too many overlaps,” said Dorre.
This article has been updated to reflect that Dorre was not speaking to specific thresholds that Hearst has in place with respect to seed data sizes.
https://digiday.com/?p=556374
