Customer discovery interviews: 25 questions to ask and how to run them

What customer discovery is

Customer discovery is the first stage of building a product: you find the people with a problem and learn how they experience it today, before you design a solution. Steve Blank named it as the first of four steps in his Customer Development method, the start of what he calls the search for a repeatable business model. The method rests on one habit, which Blank puts bluntly:

Discovery is about the problem, not your answer to it. It sits before two other kinds of research that are often mixed up with it:

Until recently, running discovery meant choosing. Calls get you the full story, but every one takes scheduling, and a week of them is a week gone. A survey is fast, but the answers come back too thin to learn from. What makes an interview work is the questions and the follow-ups, and those no longer need a call: people can answer by voice in their own time, with an AI asking the follow-ups.

The three rules behind every good discovery question

A good discovery question is about the other person’s life, asks about a specific moment in the past, and leaves them doing most of the talking. These are the three rules of Rob Fitzpatrick’s The Mom Test, named for the idea that even your mother should not be able to lie to you if you ask well. Eric Migicovsky, a Y Combinator partner, warns against the same three mistakes in his talk on how to talk to users: pitching your idea, asking about hypotheticals and talking too much.

Break a rule and you collect one of three kinds of bad data, as Fitzpatrick sorts them:

• Compliments. “That sounds really useful.” Pleasant, free to give, and no evidence of anything.
• Fluff. Generic claims (“I usually…”), hypotheticals (“I would…”) and promises about the future (“I’d definitely try it”). They describe the person someone hopes to be.
• Ideas. Feature requests and suggestions. Worth writing down, but dig for the problem behind each one rather than adding it to a list.

Teresa Torres, who coaches product teams on continuous discovery, explains why the past works better: when you ask what people typically do, they describe what they aspire to do. A specific story shows what they actually do.

The 25 customer discovery questions

Ask them in this order: context, one recent story, what they do today, what it costs, and a close that asks for a small commitment. You will not use all 25 with one person. Pick eight to ten, whether you ask them on a call or in a study. Replace the words in brackets with the task you are researching. Most work as spoken answers in an async study too; the ones marked “Best live” need a person who can follow the thread.

Context: who they are

Start with a little context, so you can place each person in a segment later. Keep it brief; the story that follows matters more.

1. 1

   “What is your role, and what are you responsible for?”

   One fact that places the person in a segment. Keep it short; this is not the interesting part.

   Short text

2. 2

   “Walk me through last week. Where did [the task] come in?”

   Last week is a real week. A “typical” week is a tidy story that leaves out the mess you are looking for.

   Best spoken · Instead of “What does a typical week look like?”

3. 3

   “Who else is involved when [the task] happens?”

   It maps the people around the problem: who does the work, who waits for it and who pays for any fix.

   Best spoken

The last time it happened

This is the heart of the interview. Get one specific, recent story and stay with it.

1. 4

   “Tell me about the last time you [did the task]. What happened?”

   The single most useful question in discovery. A real story brings the details that a general answer smooths away.

   Best spoken · Instead of “How do you usually handle [the task]?”

2. 5

   “What happened just before that?”

   The moment before a task is often the trigger, and the trigger tells you when someone would reach for a solution.

   Best spoken

3. 6

   “What was the hardest part?”

   Adapted from Eric Migicovsky’s list. It assumes something was hard, so it gets a specific answer rather than a yes.

   Best spoken · Instead of “Is [the task] hard for you?”

4. 7

   “Why was that hard?”

   The first answer names the symptom. The reason behind it is usually where the opportunity is. Ask again if the answer stays vague.

   Best spoken

5. 8

   “What happened after that?”

   Consequences show whether the problem matters. If nothing happened after, it may be an annoyance rather than a problem.

   Best spoken

What they do about it today

Then find out how they handle the problem now. Their workaround is the competition you will be judged against.

1. 9

   “What do you do about it today?”

   The current fix is your real competition, and it is often a spreadsheet, a colleague or putting up with it.

   Best spoken

2. 10

   “What, if anything, have you tried to solve it?”

   From Migicovsky’s list. “If anything” lets people say nothing, which is an important answer: a problem nobody tries to fix is rarely painful.

   Best spoken

3. 11

   “What bothers you about the things you have tried?”

   Adapted from Migicovsky. It finds the gap in current options without suggesting what the gap is.

   Best spoken · Instead of “Don’t you hate the tools out there?”

4. 12

   “Show me how you handle it today.”

   Seeing the spreadsheet, the inbox or the checklist shows steps that people forget to mention because they no longer notice them.

   Screen share

5. 13

   “What happens when it does not get done?”

   A fact-based way to learn what is at stake, without asking people to rate how important it is.

   Best spoken

What it costs and where it ranks

Now size the problem: how often it happens, what it costs and how it ranks against everything else on their plate.

1. 14

   “How many times did this come up in the last month?”

   A count for a fixed period is a fact. Offer options such as not at all, once or twice, about once a week, several times a week.

   Multiple choice · Instead of “How often do you deal with this?”

2. 15

   “What did the last one cost you, in hours or money?”

   Today’s cost is a fact you can build a price on. A hypothetical price is a guess, and usually a generous one.

   Best spoken · Instead of “How much would you pay to fix this?”

3. 16

   “Where does this rank among the problems on your plate right now?”

   A real problem can still lose to five bigger ones. If it is not near the top, people will not make time for a new solution.

   Best spoken

4. 17

   “What budget or time, if any, has gone into fixing it?”

   Money or hours already spent are the strongest sign a problem is worth solving. A budget that already exists is easier to win than a new one.

   Best spoken · Instead of “Would you budget for a solution?”

If they already switched: the story of the change

Some people will already have changed how they handle the problem. Their story of that change is the most valuable thing you can hear, and it takes good follow-up questions to draw out. There is more on this below.

1. 18

   “When did you first feel the old way was not working?”

   The first thought often comes months before any action. The gap between the two tells you how much pain people tolerate.

   Best live

2. 19

   “What happened that made you start looking for real?”

   The trigger event. It shows when people are ready to change, which is when your product has to be in front of them.

   Best live

3. 20

   “What did you look at, and what did you rule out?”

   The options they considered, including the ones they rejected, are the alternatives you will be judged against.

   Best live

4. 21

   “What almost stopped you from switching?”

   Anxiety and habit, the forces that hold people back. They explain why people who clearly have the problem still do not switch.

   Best live

5. 22

   “What is different now?”

   What progress meant to them, in their own words. Often it is not the feature you would have guessed.

   Best live

The close

End by asking for a small commitment and for the next person to talk to.

1. 23

   “What should I have asked that I did not?”

   People often keep the most important thing for last. An open question gets a real answer where a yes-or-no one gets a no.

   Best spoken · Instead of “Is there anything else?”

2. 24

   “Who else should I talk to about this?”

   An introduction costs the person some reputation, so it is a small commitment. It is also the best source of your next interview.

   Best spoken

3. 25

   “May I follow up when we have something to show?”

   Ask plainly, and make it easy to say no. A yes gives you another conversation, which is useful, but it is politeness, not proof of demand. Time, an introduction or money are the stronger signals.

   Multiple choice

Check your own questions

Most weak discovery questions break one of seven rules. Run each question you write against this list before you send the study.

• Ask about the past, not the future

  People describe the person they hope to be, not what they do.

  Instead of: “Would you use an app that chases late invoices for you?”

  Ask: “Tell me about the last time a client paid late.”

• Leave your idea out

  Once your product is on the table, people review it instead of describing their week.

  Instead of: “What do you think of our idea for a smarter invoicing tool?”

  Ask: “What do you use today to keep track of invoices?”

• Stay neutral

  A leading question hands over the answer you hope for.

  Instead of: “Isn’t it frustrating to chase invoices by hand?”

  Ask: “What was hardest about the last invoice you chased?”

• Ask about a real moment, not a habit

  General questions invite the tidy version of someone’s month.

  Instead of: “Do you usually struggle with getting paid on time?”

  Ask: “How many invoices were paid late last month?”

• Keep it open

  A yes or no ends the story before it starts.

  Instead of: “Is invoicing a problem for you?”

  Ask: “What happens when a client pays late?”

• Ask one thing at a time

  People answer the easier half, and you cannot tell which.

  Instead of: “What did you try, and did it work?”

  Ask: “What have you tried so far?”

• Ask what it costs today, not what they would pay

  Hypothetical prices tend to run high. Today’s cost is a fact.

  Instead of: “How much would you pay to fix this?”

  Ask: “What did the last late payment cost you?”

How to run customer discovery in Feedbaq

Feedbaq gives you the depth of an interview at the pace of a survey. You send one link instead of booking calls. People tell you about the last time the problem happened, in their own words and their own time, and an AI asks the follow-up a form never could.

Why it works better than a traditional survey

• People tell the story out loud. Spoken answers are about three times longer than typed ones: 44 words against 14 in a 2024 study with 1,001 people. A story needs that room.
• An AI asks the follow-up. It asks what happened just before, or what it cost them, guided by a line you write for each question.
• People can show you. With a screen share, they open the spreadsheet or tool they use today and talk you through it.
• No calls to schedule. People answer when it suits them, across time zones, so five conversations take days, not weeks.
• Answers are ready to read. Every spoken answer is transcribed and labelled, with labels such as workaround, use case and competitor, and the most insightful answers come first.
• Your AI agent can do the work. Connect Claude, ChatGPT or another agent, and it can draft the study from your sales call notes and read every answer for you.

Set it up in four steps

1. Start from the template. The customer discovery template asks about their role, the last time the problem happened, the hardest part, what they do today, a screen share of how they handle it, and how often it comes up.
2. Replace the task in brackets with the problem you are researching, and swap in questions from this guide if you need them.
3. Send the link to people who pass your screening questions. They need no account.
4. Read the answers. For the people whose answers surprised you, send a short second study that goes deeper into their story, or call them.

Here is what one spoken answer and its AI follow-up look like. The follow-up asks for a number, which turns a complaint into a cost.

Building a study is free and you can use our customer discovery template to start collecting feedback today.

Dig into one story: the switch timeline

When someone has already changed how they deal with a problem, walk through that change from the first thought to today. Bob Moesta and Chris Spiek teach this part of Jobs to be Done as a timeline: a first thought that something is wrong, a stretch of passive looking, an event that makes it urgent, active looking, a decision, and then life with the new way. Each moment has its own questions.

At every moment, listen for four forces. Two push people toward a change and two hold them back. Interviews tend to hear only about the problem, the first force. People rarely volunteer the forces against a switch, so ask about them directly.

You do not need a switcher for every interview. But two or three of these stories, from people who already paid to fix the problem, will teach you more about what to build, and how to sell it, than a dozen interviews with people who only complain about it.

How many customer discovery interviews you need

Start with five, plan for about twelve per type of customer, and then keep going at a steady pace. Two studies anchor the number twelve. Guest, Bunce and Johnson coded 60 interviews and found that by the twelfth, 88% of the themes and 97% of the important themes had already appeared. A 2022 review by Hennink and Kaiser of 23 studies found saturation between 9 and 17 interviews, with a median of 12 to 13. Both are summarized by User Interviews. Twelve interviews used to mean twelve slots in your calendar; as a study, it is one link sent to the right people.

Those numbers hold for a fairly similar group of people. A founder at a seed-stage startup and a product manager at a bank are different segments, even if both “manage feedback”, so count again for each. A practical rule of thumb: stop a round when two or three interviews in a row teach you nothing new.

Discovery also does not end. Teresa Torres recommends interviewing at least one customer every week, for as long as you build the product. A weekly habit catches changes that a single big round cannot.

Finding people to talk to

Start with the people closest to you and ask every person for an introduction to the next; cold outreach works, but slowly. Finding people is the hard part of discovery, and most guides skip it. Here are the options, closest first.

1. People you already know. Migicovsky’s advice is to begin with yourself, then friends and coworkers who have the problem. You do not need thousands of people.
2. Your users and your waitlist. If you have any, they already raised their hand. Torres suggests recruiting people while they use your product and asking customer-facing colleagues to pass names along.
3. Introductions. Question 24 is your recruiting engine. One interview that ends with two names is worth more than a hundred cold messages.
4. Where they already gather. Communities, forums, meetups and events for the role. Migicovsky tells founders to show up in person where their users are. Take part before you ask for anything.
5. Cold messages. Expect most to go unanswered. Belkins measured an average reply rate of 0.45% across 7.5 million cold sales emails in 2025. A request for advice is not a sales email, and no one publishes a reliable rate for it, but plan for silence and keep the ask small.

Keep the ask small in every channel. Torres suggests asking for about 20 minutes rather than an hour. A study is smaller still: a few spoken answers, whenever it suits them. On LinkedIn, a free account can add a personal note to three invitations a month, of up to 200 characters, so the note has to be short:

Should you pay people? Friends, users and people with an urgent problem often talk for free, because they want it solved. Strangers usually need a reason. For reference, User Interviews reports an average incentive of $85 an hour for remote B2B sessions and $80 for consumers, from nearly 20,000 projects. For a 10-minute study, that is about $14.

Check they have the problem first

Good questions are wasted on someone who has not done the task lately. Before you send the study, ask two or three neutral screening questions, and talk only to people with recent, first-hand experience:

1. “When did you last [do the task]?” In the last week, in the last month, in the last three months, longer ago, never. Keep the first two or three.
2. “What is your role?” So you can split answers by segment later.
3. “Which of these do you use for [the task] today?” A short list of tools, including “a spreadsheet”, “nothing in particular” and “other”.

Do not reveal which answers you are looking for. People who want the incentive will pick them.

How to read what you hear

Read every story, count facts rather than feelings, and decide per segment whether the problem is worth solving.

1. Get each story on one page. Feedbaq transcribes and labels every answer, and your AI agent can turn each response into what Torres calls an interview snapshot: the story in order, the forces you heard, what they do today and what it costs, in their own words.
2. Count facts. How many people had the problem in the last month? How many built a workaround? How many spent money or searched for a fix? These counts beat any rating.
3. Split by segment. A problem that is lukewarm across everyone is often acute for one kind of customer. That group is where to start.
4. Find the trigger. Look for the event that turned a background annoyance into action. It tells you when to reach people and what to say.
5. Decide. Keep going, narrow to one segment, or move to another problem. If you keep going, the next step is a concept test.

Run your first round this week

A first round of discovery fits in a week: a morning to prepare, a few days of answers, and an afternoon to read them.

1. Monday morning: write the problem and who has it, one sentence each. Pick eight to ten questions from this guide, with the story questions at the center.
2. Monday afternoon: set up the study from the customer discovery template, and send it to the closest people who have the problem.
3. Tuesday to Thursday: answers arrive in people’s own time. Read them as they come in, and send the study on to the people they introduce.
4. Friday: read the answers together. Count the facts, split by segment and decide. Then send the study to five more people.

Questions people ask

Sources

1. Steve Blank, “Profound beliefs” 2023.
2. Steve Blank, “The non-dummies guide to customer discovery” 2010.
3. Y Combinator, “Startup School week 1 recap: Kevin Hale and Eric Migicovsky” Eric Migicovsky’s five questions and three mistakes.
4. Rob Fitzpatrick, The Mom Test
5. Unusual Ventures, “Rob Fitzpatrick’s Mom Test” A summary of the rules and the kinds of bad data.
6. Teresa Torres, “Customer interviews” Product Talk.
7. Jobs to be Done, “The forces of progress and the switch timeline” As taught by Bob Moesta and Chris Spiek.
8. Lenzner, Höhne and Gavras, spoken versus typed answers to open-ended probes Journal of Survey Statistics and Methodology, 2024.
9. User Interviews, “Qualitative research sample sizes” Summarizes Guest, Bunce and Johnson (2006) and Hennink and Kaiser (2022).
10. User Interviews, “Research incentives report” Updated January 2025.
11. LinkedIn Help, “Personalize invitations to connect”
12. Belkins, “Cold email response rates” Sales email data for 2025.
13. Nielsen Norman Group, “Leading questions”