How Nala reads an interview
You’ve run the interviews, the decision is due this week, and someone will ask “how many people actually said that?” This page shows how Nala answers.
“I don’t trust a number unless I can see where it came from.”
7 of the 12 people you spoke to
That’s the shape of every answer Nala gives: people, out of everyone you interviewed, each counted once however many times they said it. Three things could make a number like that wrong. Here is what Nala does about each.
In your study, every count, like 7 of 12, opens to those seven people and what each of them said.
How we know
- A chatty interview can’t outvote a quiet one.Five mentions from one person is still one person.
- The denominator is everyone in the study,not just the people who talked about the topic. Your interviewer, and anyone you’ve marked as not a customer, are never in it.
- It says who raised a point, not how common it is.Nala won’t word a count as if it were a market share, so you won’t overclaim in the room.
Only real evidence is counted
Small talk, your own questions and passing remarks never reach a count. Everything a customer says is split into single points, and each one is judged on the same terms.
01Each answer is split into single points
The best thing a customer said is usually buried in a long, wandering answer. Nala splits every answer into passages, short stretches that each make one point, so that line is evidence on its own. Names come out before any of this, so a quote you share carries a P-number, never a name.
In your study, open any transcript to see the whole conversation, with everything Nala left out marked.
How we know
- Long answers split where the point changes.One rambling answer can’t be read as one opinion.
- Each passage keeps the question that prompted it.A short “yes, every month” still counts as an answer to its question, and your own questions never count, so a leading question can’t plant a theme.
- Filler is trimmed, in the open.“Mm-hm” and “yeah, totally” come out, and you can see exactly what came out and why.
- Only the name changes.The quote still says exactly what they said, in every export and link you share.
02Every passage gets the same questions
Read twelve transcripts yourself and you won’t read the twelfth the way you read the first. Nala puts every passage through the same few questions, answered by a model built to judge, not to write. It can’t paraphrase, embellish or invent.
In your study, every quote shows its kind and strength, and a hypothetical is marked as one.
How we know
- Is this evidence at all?Small talk, scheduling and job titles don’t pad out your findings.
- What kind?Pains are kept apart from workarounds and asks, so you can tell what hurts from what people already do about it.
- How strongly, and is it real?A heated complaint and a polite “might be nice” don’t weigh the same, and something that could happen isn’t mistaken for something that has.
- Only the clearest passages can start a theme,so a passing remark can’t become a headline.
A theme needs more than one person
Themes form only where different people make the same point. The two figures below are the same nine passages, before and after they’re grouped.
03Passages are compared two at a time
Two customers can use the same words for different problems, and different words for the same one. Nala asks, pair by pair, whether two passages make the same point.
How we know
- Each passage is compared only with the few most like it,so no comparison is a long shot.
- Only passages from two different people are compared.Your loudest customer can’t manufacture a pattern by repeating themselves.
- Each match records how sure it is,and surer matches count for more.
04Themes form where people agree
Ask a chatbot for themes and you get tidy headings, and half the quotes under them don’t fit. In Nala a theme is where many small agreements hold together, and no list of themes is set in advance.
In your study, open a theme to read every passage in it, in the speaker’s own words.
How we know
- A theme needs at least two people.A single interview never becomes a finding.
- Two themes merge only if every passage agrees across them,so themes don’t sprawl into vague catch-alls you can’t act on.
- Close calls come to you.Themes that nearly matched are offered as possible merges, so the judgment calls are yours, not a hidden setting’s.
- A passage that didn’t start a theme goes where it fits best,or is left out rather than forced in.
Every count is checked, and stays put
Nothing reaches you as a count until each person in it has been checked, and once it’s counted, it doesn’t move.
05Everyone in a count actually said it
“Did they really say that, or just brush past it?” Before anyone is counted, Nala checks each person in a theme: do they actually make the point? Someone who only touched on it stays visible beside the theme, but isn’t in the number.
In your study, quotes that didn’t count are still on the theme, marked Not counted.
How we know
- If fewer than two people pass that check, the theme is a lead.Shown to you, never counted, so you see the hunch without mistaking it for a finding.
- Who you interviewed isn’t a finding.A theme that only describes people’s role or company is set aside as background.
- One stray comment doesn’t add a person.A single passage counts only if it clearly belongs.
06The same interviews give the same answer
If the same interviews gave different themes each time, nothing you shared would hold. Every judgment is stored against its exact input, so you get the same themes and counts every time.
In your study, run the same interviews again and compare.
How we know
- Run it again and nothing moves.Nala reuses each stored answer rather than asking afresh.
- Nala never swaps its models under you,so an update somewhere else can’t shift your numbers.
- Names and summaries come last,written only from that theme’s own passages, with every summary citing its quotes. Your workspace can turn summaries off.
What you can promise your team
Nala doesn’t chat with your data. No text-writing AI judges a passage or counts a person. A writing model does one job, last: it names each theme and summarises it from that theme’s own quotes, and cites them.
Your own AI is held to the same quotes
Connect the AI assistant your team already writes with and it can read everything Nala found. A claim it makes that no quote supports is flagged, not smoothed over.
How we know
- It suggests, never edits.What your AI writes lands as a suggestion someone on your team keeps or dismisses.
- Its numbers are Nala’s numbers.A drafted “nine in ten” is held back until it says the 8 of 10 your interviews support.
- 1Run it again, get the same answer.
- 2Every number opens to the quotes behind it.
- 3We count people, never mentions.
Bring this week’s interviews.
See what your customers agree on before you decide.
Cancel any time, and ask for a refund within 14 days of any payment. Your data stays in London.