Every clinic owner I spoke to before building Callidus, the multi-tenant clinic platform I built on React, Firebase and Stripe Connect, told me the same thing about reporting. They needed it. The tool they were on made it painful, and they described that pain in detail, unprompted, with examples that had dates attached. So reporting shipped in the first release.
For about two months, almost nobody opened it. What they did open every morning, before the first appointment, was tomorrow's schedule, to check whether the day was full. Not one person had mentioned that. It never came up because it was not a complaint. It was just Tuesday.
That gap is not a listening failure. It is the default output of a conversation where one person is visibly hoping for a particular answer and the other person can see them hoping. A SaaS customer interview collects what somebody is willing to say about themselves in thirty minutes, to a near-stranger with an obvious stake in the reply. Getting truth out of that setup is mostly a question of what you refuse to ask.
What is a SaaS customer interview supposed to produce?
A customer interview should produce evidence about what someone already did, not a verdict on the product you are hoping to build. That single distinction does most of the work. Alex Osterwalder named soliciting opinions instead of facts as the biggest error in customer interviews back in 2018, and his reasoning is blunt: people rarely do what they say.
An opinion is free. "Yes, I'd use that" costs the speaker nothing, warms the room, and is forgotten by both of you within the hour. A fact costs them effort, because they have to go and retrieve it. That retrieval is the whole signal. When someone has to stop and think about the last time a thing actually happened, you are no longer measuring their generosity toward you.
Here is the test I run on every question before a call: could the answer be checked by somebody else? "How do you handle that today" can be checked. "Would you switch to a tool that did it for you" cannot be checked by anyone, including the person answering.
If you sell B2B SaaS, a second filter matters just as much. The person who feels the pain and the person who releases the budget are frequently not the same human, and the second one is harder to book. Interviewing ten enthusiastic end users and zero approvers produces a beautifully consistent story about a purchase that will never happen.
Ask about their life, not your idea
Rob Fitzpatrick's The Mom Test compresses the whole discipline into three rules. Ask about their life instead of your idea. Ask about specifics in the past instead of generics about the future. Then the hard one: listen more than you talk. The book has been on accelerator reading lists for a decade, at Seedcamp and Microsoft Ventures among others, which is a decent proxy for it surviving contact with reality.
Rules are easy to nod at and hard to hold under pressure, so it helps to see the substitutions written down.
| What you asked | What it actually measures | What to ask instead |
|---|---|---|
| Would you use this? | Politeness | Walk me through the last time you dealt with this. |
| How much would you pay? | Imagination | What are you paying for it now, and who signed that off? |
| Do you like the idea? | Rapport with you | What have you already tried in order to fix it? |
| Is this a big problem? | Willingness to agree | What did it cost you the last time it happened? |
The third rule is the one that breaks people, mine included. Silence in a call feels like a fault you are responsible for repairing, and the fastest repair available is to start explaining your product. Every second you spend explaining is a second you spend teaching the other person which answers will please you.
Problem interviews and solution interviews are different jobs
Run them on separate days, with separate people, and never inside the same call. The moment you show a mockup, you have converted a research conversation into a demo, and everything the person says afterwards is colored by having seen your work and formed a view about you.
A problem interview has no artifact in the room. You are mapping how the work gets done now, what the workaround costs, who else is involved, and what happened the last time it broke. Nothing you learn there depends on your product existing. That is exactly why the findings survive a pivot, and why they are the only research worth doing before you scope a minimum viable product.
A solution interview needs the artifact and needs a different frame: put the thing in front of someone, shut up, and watch where they stall. One writer in June 2026 put the tension well in a piece arguing that discovery interviews optimize for confidence rather than accuracy — the gap between what a customer said they wanted and what they did with a prototype is where the useful information lives.
Actually, that overstates it slightly. Problem interviews are not worthless because they measure stated preference; they are excellent at finding out what already happened, which is a fact. They are only worthless when you ask them to predict. Keep them in their lane and they hold up fine.
How do you ask about price without getting a polite number?
Ask what they pay for the current fix, who approved that spend, and what happened the last time a request like it got refused. Every one of those has already occurred, so none of them requires the person to imagine themselves into a future purchase.
The structured methods still have their place. The Van Westendorp Price Sensitivity Meter has been around since 1976 and its four questions do bracket a plausible range, and a modern write-up of willingness-to-pay research at scale puts a full study at 150 to 300 interviews with a useful pilot at 20 to 30. The same piece cites McKinsey's finding that a 1% price improvement lifts operating profit by roughly 8.7% on average, which is why this question deserves more than the eight minutes most founders give it.
But a gauge tells you the range, never the reason the top of the range exists. For that, run the conversation in this order:
- Establish the current spend on this problem, including the salary cost of whoever does it manually today.
- Find out who signs. Ask for their title, then ask what that person rejected most recently.
- Ask for a commitment rather than a number. A signed letter of intent, a pre-pay at a discount, or an introduction to that approver all cost the customer something real.
- Watch what happens in the following week. A warm "send me pricing" that generates no reply is a data point, and it is a more honest one than the number they said on the call.
You have had this call. They told you the price sounded fair, everyone smiled, and the thread went quiet. Nothing about that outcome was ambiguous.
On BookBed, the property-management SaaS I built on Flutter, Firebase and Stripe, the pricing conversations that meant anything were the ones where an owner started comparing the figure against the channel commission they were already bleeding every month. They had a live number to hold it next to. The ones who had no such reference point gave me a shrug wrapped in a compliment.
The question I stopped asking
I no longer ask anyone what their biggest challenge is. Everyone has an answer ready for that one, it is the answer they give at conferences, and it has been repeated often enough to have come loose from their actual week. Ask what they did on the Monday just gone instead. The rehearsed answer has no version of Monday.
What the cancelled customer knows that your survey does not
A churned customer can describe the decision, while your cancellation survey only ever captures the reason, and a decision has a date and a trigger, with another vendor usually somewhere inside it. Reasons are what people write in a text box to get past it.
Klue's guide to churn interviews, written by Adam McQueen in February 2025, makes a point worth stealing: 75% of tech businesses evaluate alternatives before they renew, so by the time somebody cancels they have generally been living with a comparison for weeks. That comparison is the interview. Ask what the other tool showed them that yours did not, and ask when they first went looking.
The framing that gets these calls accepted is humility rather than negotiation. You are not there to win them back, and the call stops being research the second you offer a discount. Thirty days after the fact, when the replacement has disappointed them at least once, is the sweet spot.
Who is actually running AI-moderated interviews?
Nobody outside vendor marketing can tell you, because the widely quoted 2026 adoption figures come from companies selling the tools and no sample size was ever published. That is worth dwelling on in a post about interviewing for truth.
The report doing the rounds this year claims AI customer research is now the default for 81% of research teams, 73% of UX, 67% of PM and 51% of CS, with completion rates above 85% against 22% for long-form surveys and median time-to-insight falling from 26 days to 3.2. Read the methodology section and there is none. The article refers to "our 2026 sample" repeatedly without disclosing a respondent count or a sampling frame. It is published by a vendor whose product is AI-moderated interviews.
The numbers may well be directionally right. I have no way to know, and neither does anyone quoting them, which is the point. Repeating an unsourced vendor statistic inside an article about how customers lie to you would collapse the whole argument.
What I do believe from that material is the mechanism. Async interviewing is genuinely good at breadth on a question you already know how to phrase, and genuinely bad at the moment where a human hears something odd and abandons the script to chase it. Use it after your problem interviews have told you what to ask, not instead of them.
How many should you run before you build?
Twenty conversations with real buyers, before you write a line of code, is still the right target. Jason Lemkin's 20 interview rule has held up since 2014, and it is phased: the first five map the market from the buyer's side, the next five confirm the pattern you think you spotted, and the last ten sharpen how you talk about it. Twenty real buyers, not twenty friendly users. If you are still choosing tooling while you do this, the stack decisions for a SaaS MVP can wait until interview twelve, when the pattern is visible.
Open your billing dashboard, filter to accounts that cancelled in the last ninety days, and pick three whose logo you would still be glad to have on the site. Send each of them two sentences asking for twenty minutes to learn what went wrong, with no pitch attached. If all three name the same trigger, and it is not on your roadmap, what are you going to do about it this quarter?
