Short version
Recruit ten people who had the problem in the last 30 days, ask only about what they already did, take verbatim notes, and score each interview against a written assumption. Stop when three interviews in a row tell you the same thing.
1. Give your first ten conversations a decision to inform
Your first ten customer interviews are not a vote on whether your idea sounds clever. They are a way to discover whether a particular group has a problem worth changing its behaviour to solve. A focused batch can fit into a week once recruiting is underway, and can prevent months of building around a misunderstanding.
Before booking calls, finish this sentence: 'After these interviews, I will decide whether to test this problem with a paid manual service, change the segment, or stop.' You are looking for workarounds, money spent, tools abandoned and time lost. Enthusiasm can make a call pleasant, but it cannot make that decision for you.
Throughout this guide, imagine you are exploring a scheduling service for independent physio clinics. Your starting assumption is that clinic managers spend time every week filling last-minute cancellations by contacting patients manually. That is a claim you can investigate without showing anyone a product.
2. Recruit people who encountered the problem recently
Start with recency rather than a job title alone. A clinic manager who dealt with an empty appointment yesterday can describe the sequence much better than someone giving a general opinion about scheduling. Ask whether the relevant event happened within the last 30 days; for seasonal problems, use the last relevant cycle and record when it occurred.
Use a short screener: 'Do you manage appointment bookings at an independent physio clinic? Have you handled a cancellation with less than a day's notice in the last month?' Include people who handled it easily. Screening only for frustration would build your desired answer into the sample.
Keep a recruiting log with contact source, date, qualification and booking status. Cap people who already like you at three of the first ten, and seek participants from more than one source. A friend can introduce you to an unfamiliar clinic, but note the connection rather than pretending it is independent evidence.
- Screen for the role, recent event and actual involvement in the task.
- Record nonresponses and disqualifications so recruiting does not become invisible selection.
- Ask each participant for a referral to someone who handles the same work.
3. Send an invitation that does not sell the answer
A good invitation names the work you want to understand, asks for a small amount of time and avoids promising a solution. If you advertise an exciting new automation product, you will attract people interested in automation rather than necessarily people who experience the problem. Keep the request specific enough that someone can recognise whether they qualify.
Try this email: 'Hi Sam, I'm researching how independent physio clinics handle short-notice cancellations. I'm looking to speak with people who managed one in the last month. Could I ask you about the most recent example in a 25-minute call? This is research, not a product demo. If this is someone else's responsibility, who would be best to contact?'
Offer two possible times or ask for their preference. One brief follow-up is reasonable; repeated chasing can turn a reluctant favour into misleading participation. If you offer an incentive, keep it fixed regardless of their answers and record it separately from evidence of demand. Attendance bought with a research incentive is not willingness to pay for your product.
4. Write five to eight questions before writing the product
Choose one risky assumption for each call and write it at the top of your notes. For the clinic example: 'Managers already spend staff time contacting patients to refill cancellations each week.' A short script keeps you focused, but follow concrete details rather than reading every question mechanically.
Use the interview question generator to draft questions for your context, then remove wording that introduces your proposed solution. Check whether someone could answer honestly without agreeing that a problem exists. 'How frustrating is your current process?' assumes frustration; 'What happened next?' does not.
Keep the main interview backward-looking. Asking about a recent purchasing decision is more useful than asking whether they would buy your idea. Hypothetical consequences can be useful follow-ups, but label them as estimates rather than treating them as observed costs.
- Tell me about the last short-notice cancellation you handled.
- What happened from the moment you found out?
- What did you try, and what happened after that?
- Roughly how much staff time or money did that take?
- What other approaches have you tried and stopped using?
- Who chose or approved the current approach?
- Could you walk me through a recent example or a redacted record?
Free tool: Customer interview question generatorPaste your startup idea and get eight non-leading customer discovery questions, each mapped to the assumption it tests, plus the leading questions to avoid. Free with a Vouch account.Run it on your idea
5. Run a calm call with more listening than talking
Open with a simple permission statement: 'I'm trying to understand how this works today, not persuade you to change it. It's useful to hear when something works well too. Is it okay if I take notes?' If you want to record, ask before starting and confirm permission on the recording. Explain who will have access and respect a refusal.
For a 30-minute call, reserve a couple of minutes for context, roughly twenty for one detailed incident, and the remainder for buying context and a referral. Aim to listen for most of the conversation; an 80% listening target is a useful reminder, not a measurement to optimise while someone is speaking.
When an answer is vague, pause before filling the silence. Then ask, 'Can you remember the most recent time?' If they cannot, write that down and move on. Do not rescue your assumption by supplying a story for them to agree with. End with, 'Who else should I talk to about this?'
6. Follow the story until you understand the workaround
Suppose a manager says, 'Cancellations are a nightmare.' That tells you how they describe the situation, but not what happened. Ask, 'What happened with the last one?' Imagine they explain that a therapist texted reception, reception called four patients, and one took the slot. Now you can examine an actual workflow.
Ask which steps required attention, which tools were involved and what happened when nobody answered. If they say the process took an hour, clarify whether that was elapsed time or an hour of active staff work. Those imply different costs and different opportunities.
Explore abandoned alternatives too: 'You mentioned trying an automated waitlist. When was that, and why did you stop?' Abandonment reveals constraints that enthusiasm conceals. Perhaps the clinic needs a clinician to approve suitable patients before outreach. That detail could change the proposed service completely, even if every participant agrees that cancellations are annoying.
7. Capture quotes without turning interpretation into fact
Take verbatim notes on the phrases that matter, alongside the sequence, workaround, replaced tool and current cost. Separate what someone said from your interpretation. 'Reception called four patients' is an account of behaviour; 'clinic urgently needs automation' is your inference, even if it feels obvious.
Use a compact record after each call: participant code, segment fit, recent incident, actions taken, cost, buying participants, exact quote and unanswered questions. Add the recording timestamp if you have permission to retain a recording. Do not put unnecessary patient details into your notes; ask for redacted examples rather than identifiable records.
Quotes make later conversations with a co-founder or investor more concrete, but they are not proof by themselves. Keep the surrounding context, including anything that contradicts your interpretation. If the manager also says the process works well and rarely interrupts other work, that belongs beside the colourful complaint, not in a forgotten note at the bottom.
8. Score every interview against the written assumption
Immediately after the call, mark the assumption supported, contradicted or unclear. Add one quote or observation explaining the label. An interview that did not reach the relevant topic is unclear, not supportive. If a person did not fit the segment, preserve the notes but do not quietly count them toward the segment's evidence.
In the clinic example, a manager describing weekly manual outreach supports the workflow assumption. A manager showing that existing software fills slots automatically contradicts it. Someone recalling a cancellation from two years ago provides weak evidence for a claim about current weekly work.
The validation checklist can keep the assumption and evidence together. After ten scored calls, a pattern becomes easier to discuss without relying on whoever remembers the most enthusiastic quote. If seven contradict the claim, reconsider it. If seven support it, that may justify offering a paid pilot; it does not prove they will buy.
Track the strength of evidence separately from the label. A detailed recent account with a redacted schedule is stronger than a vague recollection, but neither guarantees future adoption. Keep problem evidence distinct from evidence about budget, authority and willingness to switch.
Free tool: Startup idea validation checklistA 21-step checklist that takes an idea from assumption to go-to-market: write and rank assumptions, interview ten customers about past behaviour, ask for money, measure product-market fit, then turn the evidence into positioning and a first channel.Run it on your idea
9. Correct the mistakes that make every interview sound positive
The most common mistake is slipping into a pitch halfway through. Once you explain your clever approach, people start reviewing it instead of describing their work. If you catch yourself doing this, acknowledge it: 'I've jumped ahead. Before we discuss solutions, can we return to what happened last week?' Save a demo for a separate follow-up.
Another mistake is treating yes-or-no questions as discovery. 'Would reminders help?' provides little context. 'What reminders went out before the last cancellation?' gives you something to examine. Closed questions are fine for clarifying facts such as role or date; they are weak substitutes for the story.
Finally, do not combine unlike participants to manufacture a pattern. A solo practitioner, a hospital scheduler and an independent clinic manager may all discuss appointments while facing different constraints. Separate their evidence. If you change the segment after several calls, write down why and treat the new segment as a new line of inquiry.
10. What to do this week
On Monday, choose one segment, one risky assumption and a concrete decision. Draft your screener, invitation and short script. Start outreach early enough that booking delays do not tempt you to replace qualified participants with available friends.
Across the next few days, run calls in small batches and score each one immediately. Adjust confusing wording between batches, but do not secretly change the assumption to match the answers. Keep a brief change log so you know which participants answered which version.
By the end of the week, review the pattern. Three consecutive qualified interviews telling the same story can be a practical signal to stop repeating that question, provided they are not all from one narrow source. It is a saturation cue, not statistical certainty. If the stories conflict, inspect segment differences before booking another ten.
Choose the next action: offer a clearly scoped paid manual pilot, investigate a specific unresolved assumption, or drop the current problem framing. Ten is a useful starting target, not a certificate. The useful outcome is knowing what you will test next and what evidence made you choose it.
Questions founders ask
How many customer interviews are enough?Ten focused interviews per risky assumption is a useful starting plan. Three consecutive qualified calls repeating the same pattern can signal saturation. Check that the pattern spans more than one recruiting source, and move to a different test when more conversation is unlikely to change your decision.What makes a question leading?It pushes someone toward your preferred answer by introducing your solution, assuming frustration or inviting agreement about a hypothetical future. 'Would you pay for easier scheduling?' is weak discovery. 'What did you do after the last cancellation?' leaves room for an inconvenient answer.Do I need to pay interviewees?Not always. A modest fixed incentive may help recruit consumers or busy specialists; some B2B participants may prefer an anonymised findings summary. Check employer policies, avoid incentives tied to positive feedback and never count research participation as demand for the product.Should I show a prototype during these interviews?Keep initial discovery focused on existing behaviour. If a prototype could answer a separate question about workflow or usability, schedule that test afterward and record its evidence separately. Otherwise you risk confusing reactions to your design with evidence that the underlying problem matters.What if someone says they do not have the problem?Ask how they handle the situation today and whether anything changed recently. Their existing solution may explain the absence of pain. Record the contradiction honestly; do not argue, suggest frustrations or remove a qualified participant simply because the answer was inconvenient.Can I interview the user instead of the buyer?Yes, for understanding the work. Ask who selected the current approach and who approves spending, then seek a separate conversation with that person. User pain supports a problem assumption; it does not establish purchasing authority or available budget.
This story was originally published on 2 June 2026 at https://vouch.specky.space/blog/how-to-run-your-first-10-customer-interviews.
Top comments (0)