Advancing the frontier of persuasion.

01 · The problem

Frontier models can prove theorems and write working software. They still cannot persuade a person: they cannot tell which message someone will act on, or say something that changes what they believe.

Persuasion is as much an intelligence task as mathematics or code. What it has lacked is something to learn from.

02 · Taste it

Two messages for the same product. Which one did people act on?

Turn the sound on. Then pick the one you would act on.

Expert marketers can tell. Frontier models can’t.

03 · For labs

Expert data, environments and evaluations for marketing.

Data

Expert human data from the people who run marketing: their decisions and their reasoning, joined to what was spent and what customers did.

Environments

RL environments built on real marketing work: write the brief, ship the creative, move the budget. Graded by real-world outcomes.

Evaluations

Frontier evaluations for persuasion: can a model tell which message people will act on, and win without deceiving them?

For labs

Partner with us