The method

We model both sides of the relationship.

Your brand has a personality. Your customers have measured ones. Nothing else in your stack knows either. This page states exactly what we model, with which evidence, and where the evidence stops.

Audience side

Measured, with validated psychology

Real customer segments profiled on the Big Five — the most empirically validated taxonomy of human personality in the field. This is where OCEAN belongs, and where it is bulletproof.

Brand side

Compiled, on brand-appropriate dimensions

A brand does not meaningfully score on neuroticism. The brand side is compiled on dimensions built for brands, expressed as voice, principles and behaviour — versioned like code.

The match

How this brand should speak to that person

The product is neither side alone. It is the match between them, compiled into an object every system in your stack can read at request time.

The audience side: OCEAN, used where it holds

The Big Five — openness, conscientiousness, extraversion, agreeableness, neuroticism — is the most empirically validated taxonomy of human personality in psychology. Decades of replication across languages and cultures sit behind it. On people, the claim is safe, and we make it without hedging.

We profile real customer segments on those dimensions using your own customer research and behavioural data. That profile is what makes a synthetic test credible: it is a temperament, not an age bracket with a job title attached.

The brand side: not the same axes

It is tempting to score a brand on OCEAN too. We do not, and the reason is a finding rather than a preference. In 1997 Jennifer Aaker at Stanford deliberately mirrored the Big Five method on brands and produced a Brand Personality Scale with five different dimensions: sincerity, excitement, competence, sophistication and ruggedness.

That difference is the point. Brands are not perceived along the same axes as people. A brand does not meaningfully score on neuroticism, and a model that claims otherwise is borrowing credibility it has not earned.

So the brand side is compiled on brand-appropriate dimensions, stated explicitly in the output, and expressed as something operational: voice, principles, behaviour, and the words the brand does not use. A person reads it and recognises the brand. A model reads it and can act as the brand.

The match is the product

Most tools treat brand personality as a mood board and customer psychology as segmentation, and never connect the two. We model both sides and compile the relationship between them: how a brand with this personality should speak to a customer with that profile.

That object is what AI Agents, Dialog Center, Playbook OS and Synthetic Sandbox read. It is also what makes a synthetic test worth running — the persona is conditioned on a specific brand and specific customers rather than a population average.

Synthetic respondents: the honest state of the art

The supporting evidence is real. A 2024 Stanford study simulated a thousand real individuals as language-model personas and reproduced their attitudes and behavioural patterns more accurately than demographic baselines. Google researchers in 2026 tested synthetic personas on previously unseen topics and covered over 80% of the range of human opinion.

The counter-evidence is equally real, and we would rather you heard it from us. A 2026 controlled comparison of 117 real interviews against synthetic ones found three frontier models converging on identical themes and modal answers — averaging models gives you three phrasings of one consensus, not the disagreement or the outliers. In the Rival Group 2026 trends report, 42.75% of market researchers said they were not excited about synthetic respondents.

Our position is the same as the consistent professional position: synthetic research is a complement, not a replacement, and should be validated against a small real sample. What we add is conditioning. Generic vendors assemble personas from demographics and public data, which is precisely why the models converge — the personas are not distinguishable enough to disagree. Profiles built from a specific brand's real customers can disagree, because they were built from different people rather than the same average.

Stated plainly

What we claim, and what we do not.

We claim

  • We claim the Big Five is validated on humans, and we use it on the audience side only.
  • We claim the brand side is compiled on brand-appropriate dimensions, and we name them in the output.
  • We claim synthetic testing is a filter that runs before real research.
  • We claim the identity object is versioned, exportable and readable by any system you point at it.

We do not claim

  • We do not claim brands score on human personality factors.
  • We do not claim synthetic respondents replace talking to real customers.
  • We do not claim a personality model predicts revenue.
  • We do not claim a compiled identity is a substitute for having something worth saying.