Know why they hesitated, before the next cut.
BuyerLens is a synthetic buyer interview engine. Run a structured panel of 5–15 personas against your creative or concept, surface the specific resistance, and decide which objection to remove next.
The score tells you how strong. The interviews tell you why not.
RoastIQ predicts how an ad will perform. BuyerLens predicts the specific reason a real buyer would hesitate, the line, the missing reassurance, the wrong reference point.
Synthetic interviews are not a real consumer panel. They are scenario simulation, structured against the same persona dimensions Kantar uses, and validated by inter-rater agreement against real interviews. We are explicit about both ends.
- Old way · SurveyRecruit, screen, field, code$8k–$40k. 3–6 weeks per round.
- Old way · Focus group8 people, 90 minutesOne room, one moderator, one narrative.
- Now · BuyerLens10-persona synthetic panel5 minutes. Tagged resistance, not a transcript.
- After launchIn-market validationReal signal closes the loop, on a smaller bet.
From one brief to a tagged resistance map, in four passes.
A frozen interview spine, a structured persona panel, deterministic tagging. The interviewer changes nothing about the question. Only the audience changes.
Set the brief
Define the audience (category, region, life stage), pick the interview spine (Conversion, Resistance, Concept, Pricing, Switch), upload the creative or paste the concept. That's the entire input.
Build the persona panel
5–15 personas drawn from a pre-built persona library, balanced across role, life stage, brand familiarity, price sensitivity, and category fluency. Each persona is a frozen JSON spec, not a re-rolled prompt.
Run the interview spine
The same 6–10 questions per spine, in the same order, against every persona. No model creativity on the question side, that's how the same answer means the same thing twice.
Tag the resistance
Each answer gets coded to a frozen taxonomy: Trust gap, Price uncertainty, Wrong reference, Missing proof, Format friction, and so on. The output is a resistance map with frequencies, not a wall of quotes.
Five axes. Balanced panels. One resistance map.
Each panel is balanced across the same five persona axes, so a 10-persona run is not 10 versions of the same person. The axes are frozen, panel composition is reproducible.
Three resistance states. One ladder. One next move.
Every BuyerLens panel lands on exactly one of these. The rule is mechanical.
Synthetic, but reproducible. Here's how we keep it honest.
Two runs on the same brief should return the same resistance map within tolerance, and the tags we generate should agree with what a trained human coder writes down. We measure both.
What BuyerLens is, and what it isn't.
Synthetic interviews work because they are reproducible and structured. They fail when they are read as a real panel. We are explicit about which is which.
BuyerLens does
- Run a balanced, frozen-persona panel against your creative or concept.
- Apply the same five interview spines in the same order, every run.
- Tag every answer to an 11-tag resistance taxonomy with ~85% human agreement.
- Return a resistance map, dominant tag, and one mechanical verdict (Convert / Tension / Block).
- Re-run on demand against a single edit, so you can watch the dominant tag fall.
BuyerLens does not
- Run a real consumer panel, every persona is a frozen scenario spec, not a human.
- Measure brand recall, that requires recall-survey methodology.
- Predict sales lift, ROAS, or attributed conversion.
- Generate creative ideas, the spine asks, it doesn't author the answer for the brand.
- Replace post-launch listening. The map runs before the spend, not instead of it.
Hear the panel before you re-brief.
Upload the cut, pick the spine, get the tagged resistance map in five minutes. Decide which objection to remove next, not whether one exists.