How many interviews are enough? Sample size in qualitative research

Twelve people in Madrid. That was the whole sample of the customer journey study we ran for Amazon in Spain, and if you’ve ever presented a qualitative study to a finance team, you can guess the first question. Twelve? Our last survey had two thousand people in it.

It’s a fair question, and it has a better answer than “qualitative is different, trust us”. The research on qualitative research sample size is more specific than you might expect, and it explains why our published case studies range from 12 interviews to 109.

Why this is a challenge

A survey and a set of interviews measure different things, so their samples follow different rules.

A survey estimates how many people think or do something, and its precision depends on how many people answer. Pew Research Center’s explainer puts it simply: a simple random sample of 1,067 people has a margin of error of about three percentage points either way, and a subgroup of around 160 people inside it has a margin of about eight.¹ That’s why surveys need hundreds or thousands of answers.

Interviews map which reasons exist and how they work: why people buy, hesitate, switch or stay. Here the useful question is whether one more interview would still tell you something new. Researchers call the point where it stops doing so saturation, and it has been tested directly.

What the evidence says

• In 2006, Greg Guest, Arwen Bunce and Laura Johnson coded 60 in-depth interviews with women in two West African countries and tracked when new themes stopped appearing. Saturation came within the first twelve interviews, and the basic elements of the main themes were already there by the sixth.² The group was fairly similar and the topic narrow, which matters for what follows.
• In 2022, Monique Hennink and Bonnie Kaiser reviewed 23 studies that had tested saturation. Most of the 17 that used real interview or focus group data reached it within 9 to 17 interviews, or 4 to 8 focus groups. Those studies mostly had fairly similar participants and narrow aims, and mostly tracked whether new themes kept appearing. The exceptions, such as multi-country studies and studies of deeper meaning, needed more.³
• Saturation also depends on what you need from the data. In a 2017 study based on 25 interviews with US veterans living with HIV, Hennink, Kaiser and Vincent Marconi found that the range of issues was covered after nine interviews, while understanding those issues properly took 16 to 24, and one issue was still showing new sides at the last interview.⁴ It’s the difference between having heard it all and understanding it all.
• Comparing groups pushes the number up. In a cross-cultural study of 132 people in four sites, Ashley Hagaman and Amber Wutich found that 16 or fewer interviews were enough to find the common themes in sites where people were fairly similar. Finding the themes shared across all four sites took 20 to 40 interviews per site.⁵
• Market research has its own classic. In “The Voice of the Customer”, published in 1993, Abbie Griffin and John Hauser interviewed 30 potential buyers of coolers, picnic baskets and similar food-carrying products, and found that any 20 of them, on average, surfaced more than 90% of the needs that all 30 did. Their working estimate was that 20 to 30 interviews per segment capture 90% to 95% of customer needs.⁶

Most of these tests come from health research, where the question and the people tend to be tightly defined, and the one classic from market research landed higher. None of this gives you a magic number, and the researchers behind these studies say so. Kirsti Malterud and colleagues argued in 2016 that the right size depends on what they call information power: how narrow the aim is, how specific the participants are, how much established theory guides the study, how good the conversations are and how the analysis is done.⁷ Virginia Braun and Victoria Clarke went further, arguing that saturation doesn’t fit reflexive thematic analysis, the interpretive approach they developed, because the analyst builds meaning from the data instead of digging it up. In their view, how many interviews is enough can’t be fully settled before the analysis starts.⁸

What the evidence does give you is a direction. Around a dozen interviews can be enough to hear the main themes from one fairly similar group with one focused question. Understanding those themes in depth, or listing nearly every customer need, has taken nearer 20 to 30, and every extra group you want to compare adds more.

Saturation tests show when new themes stop turning up among the people you’ve interviewed, so they can’t warn you about a view that none of them happened to hold. Andrew Fugard and Henry Potts proposed a simple way to think about this.⁹ If one customer in ten holds a view and people are picked at random, twelve interviews give you about a 72% chance of hearing it even once, and you’d need 29 to be 95% sure. If a small group matters to the decision, recruit it on purpose.

Four projects, four sample sizes

This is how that played out in four of our projects, all of them published as case studies.

• Amazon, 12 shoppers. One city and one focused question: how a purchase in Spain moves between the shop and the screen. Instead of adding people, we added depth. Each shopper kept a mobile diary for three weeks, we went to the shops they chose with every one of them, and we interviewed them after they paid. Twelve people followed that closely give you far more information each than a one-hour conversation would.
• Finom, 24 interviews. The value proposition study for Finom compared two markets, so the sample was 12 in Germany and 12 in France: 20 entrepreneurs and small business owners, plus two specialists in entrepreneurship and fintech and two startup investors to place what users said within the wider market.
• A care app, 100 interviews. The care app study had to hear everyone around an older person’s care. About a third of the interviews were with families, a third with care homes and a third with municipal social services, plus care organisations such as the Red Cross and older people themselves, across Asturias and Alicante, in cities and rural areas.
• Ironhack, 109 interviews. The State of Tech reports for Ironhack covered nine countries, with the same six voices in each: three tech leaders, three career changers, two key opinion leaders, two Ironhack students, a gender expert and an Ironhack outcome manager. That’s twelve people per country. One hundred and nine sounds like a lot until you see it’s nine sets of twelve, plus one interview with the founders.

At that size, a project is really several studies run side by side, one per group or per country, and that’s how both were designed. Count the groups honestly, too. Each Finom and Ironhack country mixed several profiles, so twelve per country gave a solid local picture without saturating every profile, and the Ironhack reports also drew on more than 1,300 published sources. If a study has to compare themes across countries in depth, the evidence points nearer 20 to 40 interviews per country.⁵

The lesson is the same in all four: count the interviews per group you need to understand, not in total.

What to ask about the sample in a proposal

If a proposal on your desk includes interviews, these questions will tell you whether the number has been reasoned or simply copied from the last project:

• How many distinct groups will be compared, and how many interviews does each one get?
• Which small but important groups will be recruited on purpose?
• How will people be chosen, and who checks they qualify?
• How focused is the question? A narrow question needs fewer conversations than a broad one.
• How deep is each conversation: a 45-minute video call, a two-hour visit at home or three weeks of diary?
• What will the team do if new themes are still appearing at the end of fieldwork?
• Who analyses the material, and how many people read it?

The last question matters more than it looks. In Griffin and Hauser’s study, seven readers coded the same transcripts: an experienced qualitative analyst, two students and four engineering teams. Each found on average 54% of the customer needs, the expert was at the low end, and together they found far more.⁶ Different readers catch different things, which is one reason our quality standards have a second senior expert read every project cold before it reaches you.

And if what you need is a percentage, such as how many customers feel each need or which group is biggest, interviews won’t give it to you at any sample size. That’s a job for a survey, ideally written after the interviews, so it asks about the reasons people actually gave. Interviews, a survey and social listening are the three lenses our H3™ method fuses into one reading.

Count the groups before you count the interviews, and the number starts to explain itself.