More brands are blending deterministic and probabilistic data for hybrid targeting approaches
By Sam Bradley • April 3, 2025 •
Ivy Liu
Advertisers are looking again at the use of probabilistic data for audience targeting, and, in some cases, are seeing it as a potential solution to addressability concerns.
It’s the latest installment in marketing’s tug of war between precise targeting and broad reach — and an indirect consequence of third-party cookie deprecation.
“It’s not an either-or scenario. There’s no favorite child,” said Kumar Amrendra, head of digital marketing at U.K.-based pay-TV broadcaster Sky.
Deterministic data — information that can be used to directly refer to an audience or customer — is often first-party data (for example, an email address). By contrast, probabilistic data (an IP address) is used to infer information about an audience. Though many digital marketers prefer to trust the precision of the former, some are turning to a blended approach, employing both kinds of data in their audience targeting — in the hope they can enjoy both reliability and reach.
Take Sky, for example. It’s been working to blend its own first-party data with probabilistic data from publishers held in ad tech provider Permutive’s data clean room.
This is known as “lookalike modeling,” a well-established practice that uses one dataset (a deterministic one) and matches attributes within it to a larger audience database. The matches provide a marketer with a dataset of web users similar to their existing customers who they can then target.
“Essentially they have an identity of that person, and they’re trying to use those characteristics to find like-minded people,” said Tim Lathrop, vp of platform digital at Mediassociates.
According to Amrendra, the approach resulted in a “new headroom” for Sky in the form of a 121% rise in incremental sales, and a 55% reduction in cost-per-acquisition. He didn’t provide the dollar CPA.
“When you move into this way of buying, everything changes radically,” said Permutive CEO Joe Root.
Sky began testing the new targeting approach in October 2023 and, eighteen months on, Amrendra said “very little” of its digital marketing budget is not invested this way.
As a subscription business, Sky’s in a good position to take advantage of such an approach — it gathers a lot more first-party data itself than a packaged goods brand might, for instance. That doesn’t mean other clients or agencies can’t reproduce its framework, though, especially as AI tools developed by ad tech players proliferate.
John Gladysz, head of product at indie media shop Noble People, told Digiday that the agency was testing a related approach using Claritas, an AI audience optimization tool. Meanwhile, tech firm Resonate launched an “Audience Builder” tool this February, which promises to “instantly create highly targeted audience segments for insights and activation,” per a press release — while Yahoo’s demand-side platform (DSP) offers “next generation” tools that include lookalike modeling.
“I really do hope more brands get on board with probabilistic … the scalability of deterministic ebbs and flows. I think the best approach is to take a hybrid approach,” said Alice Beecroft, senior director of global strategy and partnerships at Yahoo DSP.
‘Too many buyers are stuck in this legacy mindset’
Why is this relevant now? Perhaps it’s best to think of it as one of the many long-tail consequences of cookie deprecation. When the third-party cookie was advertising’s global currency, deterministic approaches edged out the use of probabilistic data. Why guess when you can pinpoint, went the idea.
The years-long cookie deprecation journey led many marketers and media agencies to look toward first-party data as a cleaner, better way of targeting users on the web. In time, they’ve found limitations to that alternative, namely size and cost. “Deterministic is very hard to scale,” said Matt Wilke, head of programmatic at Mediaplus U.K.
Alternative identifiers, meanwhile, haven’t yet reached ubiquity among media buyers and marketers, despite their availability (and aren’t free of suspicion since the Colossus ID spoofing episode).
It’s those limitations, and the emergence of AI applications in the ad tech space, that have caused some to look back toward probabilistic solutions.
Ad tech companies are bullish on the idea. “AI only accelerates the opportunity with modeled and probabilistic datasets. But too many agencies and buyers are stuck in this legacy, old-school mindset,” said Permutive’s Root.
Media buyers seem less enchanted. They’re careful to prescribe hybrid approaches that use each type of data like opposing jigsaw pieces.
“We would love it if we could only rely on deterministic data,” said Gladysz. “The reality is that probabilistic is more scalable and cheaper and easier to deploy.”
“Deterministic [data] is critical for a lot of clients, but we’re using a hybrid approach to make sure that we’re getting precision as well as scale for our clients,” added Lathrop.
As Charles Simon, vp of private advertising standards at DSP provider RTB House, put it: “Addressability is going to get complicated, but the stuff that’s going to win out is going to be some material hodgepodge of deterministic and probabilistic signals.”
https://digiday.com/?p=574070
