Guide, 12 October 2026, by Thilo Tamme
How many people do you need for interview results to be anonymous and useful?
A private equity operating team asked us this about 100 team leads. Their worry was fair: with so few people, would anyone be recognisable, and would there be enough to see a pattern? Two numbers decide it. Results only show for groups of five people or more, and the more ways you want to split the results, the more people each split needs.
The rule: five or more per group
Every figure you see comes from at least five people. Below five, it's withheld on every screen, export, report and API response. A team of four still counts in the totals; you just can't look at it on its own, and you can't stack filters until one person is left.
Count the groups before you count the people
The total matters less than the smallest group you want to look at. If you want to compare four regions, each region needs five people who took part. If you want regions by seniority, each of those combinations needs five.
Not everyone you invite will take part. Participation was 87.5% in one deployment; plan with a lower figure until you know your own.
Worked examples
Example only, not a client result. Each assumes 70% of invited people take part.
| Invited | Taking part | What you can see |
|---|---|---|
| 50 | 35 | Themes across everyone, and two or three large groups such as office and frontline |
| 100 | 70 | Themes by four to six groups, for example by region or function |
| 200 | 140 | Themes by team and by seniority, and most combinations of the two |
For a first project, we plan with about 200 people when you want results by team. From about 50 people the themes are useful, with fewer ways to split them.
When a theme counts
Size also decides when a theme is solid enough to act on. Ten agents code each statement, and a theme counts when six of them agree. Next to every theme you see how many people raised it, for example "29 of 120", so a theme from three people doesn't read like one from thirty.
If you interview the same people again
When you ask again after acting on the results, keep the groups the same so you can compare them. If a team shrinks below five, it drops out of the comparison and stays in the totals.
A checklist before you invite anyone
- Write down the groups you'll look at, and the smallest one.
- Check each group can reach five people who take part, with room to spare.
- Agree the groups with your works council if you have one.
- Tell people in the invitation that results only appear for groups of five or more.
More on what happens to what people say: Are AI voice interviews anonymous?