Better Than Targeting: Aampe’s Agentic Approach to Learning From Users

Marketers often talk about “learning what works” as if that means identifying which types of users respond best. The logic goes: find the attributes that correlate with success, target those users, and scale.

That approach feels data-driven. It isn’t.

A recent International Journal of Research in Marketing paper makes the weakness visible. The authors tested hundreds of user segments across Facebook and Spotify and quantified the tradeoff between precision and reach. The pattern was stark: as segments became narrower - for example, moving from “people interested in cars” to “people interested in compact electric cars of a specific brand” - the improvement in click-through rate had to rise non-linearly just to keep profits flat. A 5%-reach segment needed roughly a 150% performance boost to match an untargeted campaign. Those gains are rarely achievable in practice.

That math exposes the hidden fragility of attribute-based optimization. Each time you slice more finely by demographics, interests, or behavior, you reduce the surface area for discovery. Your model starts learning only about the users it already understands, and becomes blind to everyone else. When data quality drops (as it did for online advertisers after Apple’s App Tracking Transparency rules came out), those hyper-narrow segments collapse first. Their apparent efficiency was a byproduct of over-fitting to a shrinking pool of predictable users.

Correlation-based micro-segmentation isn’t learning. When you “learn” that 35-year-old suburban parents convert better, you haven’t discovered what makes them respond - you’ve just described a cluster of historical coincidences. You can’t infer why something worked or how to improve it next time. You’re optimizing over artifacts, not behavior.

Real learning comes from interaction. You try something with a user, observe the response, update your understanding, and try again. Over time, you build a model of how to help this person decide — not which demographic bucket they sit in.

Agentic infrastructure makes that possible by treats the conversation itself as a first-class citizen in learning. Each exchange refines your model of that individual’s context, goals, and responsiveness.

The future isn’t better targeting. It’s better interaction - systems that learn from what they do with people, not from what they believe about them.