Help
EspañolHow a study gets made
From the brief to the report. What you decide at each step, what to avoid, and what the platform does not do today.
The path, end to end
- 1Create the study. Name, objective, market and language. The market cannot be changed later; language and study type can.
- 2The audience. Who you want to talk to. Describe it in plain language and the AI proposes the segments; you edit them.
- 3The guide. The questions. The AI drafts them from your objective.
- 4Review. The platform reviews the guide before launch and flags what will cause trouble.
- 5Launch. Interviews run. You can watch the conversations as they happen.
- 6Read. Analysis, findings with their verbatim quote, and a downloadable report.
What each plan includes
This is not just billing: it changes what the screen lets you do. If something does not appear or bounces back, it is almost always the plan and not a bug.
| Free | Studio | Intelligence | |
|---|---|---|---|
| Interviews per month | 10 | 50 | 300 |
| People per audience | 10 | 50 | 100 |
| Segments | 2 | 3 | 10 |
| Questions per guide | 8 | 20 | 30 |
| AI analysis | — | Yes | Yes |
| PPTX and PDF report | — | Yes | Yes |
| Segment Intelligence | — | — | Yes |
| Copilot over the study | — | — | Yes |
Writing a segment that holds who you say it holds
This is the decision that changes the result the most, and the one most often got wrong. What follows is not craft advice: every rule comes from a measurement of ours, and carries its number so you can judge it.
1. Write it in the positive, not by what the person does not do
An affirmative condition lands well: "has just come back from a trip" holds in 5 of 5. A negation dilutes: "does not travel" holds in 2 of 5. The nuance matters — none of the five said the opposite: the generator does not invert the condition, it forgets it. If you need the absence of something, say it through what that person does instead.
2. Say what the segment IS, not the context it lives in
Naming a sector works (5 of 5) but it overruns the occupation: you get chimeras like "a lab technician working at a four-star hotel". If your segment is a role, say it as a role. If it is a behaviour — you play padel, you listen to rap — do not give it an occupation or an age: a behaviour segment holds without either, measured across four transport modes, 5 of 5 in all four, with ages from 29 to 83.
3. If a segment comes back with no fixed vector, pick an archetype
The AI proposes a psychometric vector when the brief supports it, but it does not always manage. A segment with no fixed vector gets a sampled one, and two sampled segments may not differ at all — the opposite of what you want when comparing groups. The review screen tells you which ones lack it: that is where you can give them an archetype.
4. We check that every person qualifies, and replace those who do not
Before anyone is interviewed, each biography is judged against the condition your segment declares, and whoever contradicts it is dropped. In the worst case measured — that negation landing in 2 of 5 — asking for 5 people with the gate on delivers 5 of 5 confirmed. It costs one extra generation round.
Only those who contradict are dropped, never those where it "does not say": that is why the gate cannot empty a segment on you. And if your condition leaves no trace in a biography — the channel someone buys through, say — the gate cannot see it, and we tell you the verdict anyway on each person.
What this does NOT mean. A segment controls who gets in. Whether the panel composition changes what gets decided inside the interview is a different claim, and we do not treat it as proven.
How many people per segment
- Split so each segment has enough. 6 people across 4 segments is four groups of one or two: you are comparing nothing.
- To measure divergence you need 2 segments and at least 3 people in each. Below that, the report cannot claim two groups think differently — and it will tell you why it is not claiming it, instead of leaving you a blank.
The interview guide
The AI drafts from your objective and you edit it. Before launch the platform reviews it and flags what usually goes wrong: closed questions answered with a yes, double-barrelled questions, and questions that suggest the answer.
If you want Segment Intelligence, your guide needs at least one bounded-choice question — between two and five options. With an open question there is nothing to diverge between, and the report will say so rather than invent a difference.
What you CANNOT do today
We would rather you knew before paying than after trying.
- Study markets outside the eight supported
- Spain · Mexico · United States · United Kingdom · France · Germany · Italy · Brazil.
- Change the market of a study already created
- It is chosen at creation. Language and study type can be changed later; the market cannot.
- Measure divergence with an open question
- It needs a bounded choice of two to five options, and at least 2 segments with 3 people each.
- Hand-edit a person already generated
- You regenerate them or re-run their interview; you do not edit them field by field.
Things that keep happening
- "It will not let me add more segments"
- That is your plan cap. The notice tells you how many it allows.
- "The study will not start"
- It has no guide yet.
- "I asked for 50 people and got 10"
- Your plan people cap, or the monthly quota.
- "The report has no segment section"
- It needs Intelligence and, on top of that, a choice question with 2 × 3 people.
- "The people sound Spanish and my study is UK"
- Wrong market chosen at creation.
- "I cannot see Insights"
- AI analysis is not included in Free.