The ad tech industry is in its AI era. Every major player is leaning into large language models, creative agents, or conversational workflows designed to supercharge media buying. But while attention is focused squarely on the buy side and buzzwords dominate earnings calls, the sell side, especially app publishers, continues to face an old, persistent challenge: understanding their users beyond surface-level behaviors, and addressing them meaningfully.
Imagine you’re at a bookstore. A shopper heads straight for the gaming section. To the store staff, they’re a “gamer.” But what if this same person also reads travel guides, buys parenting books, and always pays in cash after 8 p.m.? You’re looking at a much more nuanced individual than just a gaming enthusiast. These behavioral nuances can speak volumes – far beyond what a publisher or app marketer, in the context of advertising, can infer from within the boundaries of their own app.
What you can’t see will cost you
The deprecation of device identifiers like IDFA has only made matters worse. Without cross-app signals, publishers are flying blind. Their understanding of audience segments becomes shallow, and their ability to price or personalize inventory weakens. Most audience enrichment efforts today come from external data providers, which introduces latency, potential privacy risks, and cost.
Meanwhile, advertisers expect precision. They want to know whether they’re reaching a parent, a business traveler, or a health-conscious consumer and they want that context in real time. This creates a growing gap between what advertisers demand and what publishers can confidently offer.
Adtech’s AI obsession, and what it overlooks
Much of the industry is solving for the buy side: optimizing bids, improving creative generation, or building media plans through agent-like interfaces. These are meaningful advancements, but they rarely help publishers answer a more fundamental question: “Who is my user, really?”
This is where machine learning, particularly when embedded directly on-device, becomes compelling — not flashy, but functional. Unlike generative AI, this form of intelligence doesn’t write scripts or generate headlines. Instead, it quietly processes patterns, usage habits, and context signals to build user cohorts with greater accuracy, all while respecting user privacy.
On-device cohort modeling: a quiet revolution
When it comes to the challenge of poorly understood mobile audiences, on-device cohort modeling represents an effective solution. This technology, embedded within an app, can interpret user behavior locally, without sending personal data to external servers. This approach respects platform privacy norms, doesn’t rely on third-party cookies or universal IDs, and offers real-time cohort assignment based on granular, behavioral signals.
On-device cohort modeling, when deployed across a significant enough number of apps, can be leveraged to define significantly more active user cohorts, and in greater detail. For example, a gaming app user previously labeled generically as a “puzzle player” could be reclassified into a more nuanced cohort — “a frequent traveler, a late-night player, a high income earner with a high attention span.” These are insights that are generated in real time, at programmatic speed, and this approach reshapes how advertisers and publishers communicate with real people. These aren’t just one-size-fits-all app labels.
For publishers, on-device cohort modeling provides the ability to:
- Personalize in-app experiences more meaningfully, based on the user’s location context or previous app actions.
- Create better pricing tiers for inventory based on inferred user value.
- Monetize without exposing user data or relying on shared identifiers.
For advertisers and their DSP partners, it means better bidding predictability and a more consistent performance across environments using both ID Solutions and ID-Less ones, as outlined in IAB Tech Lab’s guidance releases. Cohort-based audience is, in fact, one of the recommendations in the IAB Tech Lab’s ID-less guidance.
The future doesn’t need to feel like science fiction. It just needs to work better.
On-device intelligence doesn’t make headlines like generative AI does. Nor does it spark LinkedIn threads. But its potential for publishers and the ecosystem as a whole is significant.
As the industry celebrates creative agents and AI-powered workflows, it’s worth asking: Are we overlooking the foundational intelligence that could actually improve results? In an industry eager for breakthroughs, sometimes the best innovations are the ones closest to the user and furthest from the hype.
Anish Aravindakshan
Head of Product Marketing
Verve