PersonaCraft: Consumer Personas, Built Three Ways

Personas get invented in a workshop and then believed. Six months later a team is arguing about what a 34-year-old aspirational buyer would do, and nobody in the room remembers that he was made up on a Tuesday afternoon by people who had not spoken to a customer that quarter.

The failure is not that personas are fiction. It is that the fiction loses its label.

An application that produces a persona by one of three routes, and marks every persona with the route it came from.

• Generated. A demographic profile goes in: age, location, income band, occupation, gender. A language model returns a full persona with psychographics, goals, pain points, behaviours, a representative quote and a short backstory.
• From a question flow. The same structure, assembled from answers to a guided set of questions, so a team’s own assumptions are made explicit rather than borrowed from a model.
• From interviews. Built from what real people actually said.

Every card carries its badge. A compare view puts two personas side by side.

The personas are written for the market they describe rather than translated into it. The current set runs through Kolkata neighbourhoods, family businesses, and the gap between a traditional bazaar and an online one.

118 people have used it, and the live set holds 41 personas.

What I have not measured is the thing that matters: whether a team working from labelled personas decides better than a team working from unlabelled ones. That needs a design I have not run.

The badge does more work than the model does. Watching a room, a persona marked interview based gets quoted as evidence, and the same persona marked AI generated gets questioned. That is the right instinct. Software can either support it or hide it, and most persona tools hide it.