The Craft of the Mom Test | How to Get Honest Answers in Customer Interviews

Hello, this is Ryuta Hamamoto from TIMEWELL.

Whenever I advise people on new ventures, there is one warning sign that scares me most: the report that goes, "We asked customers, and everyone said it sounds great." The person's face is bright. They feel like they're onto something. But every time I hear this, a chill runs down my spine. I have seen the same failure over and over: a new business with hundreds of millions of yen poured in, every single person interviewed was enthusiastic, and yet six months later the number of people who actually paid was zero.

The problem is in how the questions were asked. People are kind, so when someone in front of them has worked hard on something, they will say "that's great." The moment you mistake that social courtesy for evidence for your business, your validation stops. Today I want to talk about the technique for slipping past that flattery and reaching the truth: the Mom Test¹. For those who want to build a new business or start a company with AI, I'll write it out in a way you can act on tomorrow, from how to craft your questions all the way through analyzing them with AI.

Here are the three bare bones of this article up front.

• "That's great" is not evidence. What counts as evidence is past behavior, meaning money actually paid or traces of a homegrown workaround.
• Good questions have a three-beat rhythm of "past, specific, and other-person-centered." Start by deleting every "would you use it if it existed."
• AI cannot stand in for a customer. The practical move is to use it as a partner that makes transcription and first-pass analysis ten times faster.

If you're curious how far you're already able to use AI in building your own business, running the AI Literacy Check first to take stock of where you stand, which takes about ten minutes, will make it easier to see exactly where in this article you can apply what.

Why You Must Not Rejoice at "That's Great"

Let me start with the single most important point in this article. Answers like "that's great," "seems handy," and "I'd use it if it existed" are the worst possible replies in a customer interview. It runs counter to intuition that positive words could be the worst, so let me break the reason into three.

First, those words cost nothing to say. Neither the person's wallet nor their calendar has moved a single millimeter. Saying "that's great" is free, and it's actually convenient as social courtesy that keeps the mood pleasant. Agreement that involves none of the pain of paying money, freeing up time, or persuading a boss is meaningless.

Second, it's a statement about the future, and nothing is less reliable than your future self. Of ten people who say "I might use it," at best one actually will. People paint their future selves as more conscientious, more decisive, and more impressive than they really are. Everyone has bought a piece of exercise equipment and never used it. Intentions about the future almost never play out as stated.

Third, and this is the scariest, mistaking praise for evidence stops your learning cold. The moment someone is satisfied that "everyone said it was good," they start moving to the next stage. On to development, or off to fundraising. Even though in truth nothing has been validated yet, they push ahead convinced they've confirmed it. This is the textbook failure that melts hundreds of millions of yen. Precisely because everyone was favorable, nobody paused.

That's why, when "that's great" comes up in an interview, I actually become more wary inside. Instead of rejoicing, I steer the conversation back to past behavior: "So, have you tried anything similar recently?" I'll cover this countermove in detail later.

What the Mom Test Is, and Its Three Principles

The name Mom Test comes from an idea popularized by the entrepreneur Rob Fitzpatrick in his book of the same title¹. The concept is simple. Your mother will say "that's lovely, wonderful" no matter what you show her. Because she loves you, she won't give you an honest evaluation. So how do you ask questions in a way that gets even your mother to speak the truth? If you can design your questions to that level, you can slip past flattery no matter who you're talking to. Hence the paradoxical name.

There are three principles to the Mom Test.

First, ask about past behavior, not a hypothetical. Instead of "would you use something like this if it existed?", ask "when was the last time you did that, and what did you do?" Intentions about the future are prone to becoming lies, but past behavior is fact.

Second, ask about specifics, not abstractions. Ask "is there anything you're struggling with at work?" and you'll get nothing but vague, inoffensive answers. Ask "during last month's month-end close, which task took the most time?", tied to a specific scene, and vivid stories come out.

Third, keep your own idea hidden for the first thirty minutes. This is the one people fail to follow most. The instant you talk about your idea, the other person shifts into a mode of evaluating your idea rather than their own life. And the kinder they are, the more they say "that's great." So at first, ask only about the reality of their life and work. Bring up your idea at the very end, once you thoroughly understand their reality.

Since words alone are hard to grasp, here's a quick reference of NG and OK questions.

Common NG question	OK question that follows the Mom Test
This task is a hassle, right?	When did you do that task yesterday, and how many minutes did it take?
Do you think a tool like this would be handy?	Have you tried anything to make that task easier before? Why didn't it stick?
Would you pay 5,000 yen a month?	Have you paid for any tool or service around that area recently? How much was it?
Would you want to use a service like this in the future?	How are you actually getting that job done right now?

What the left column shares are three poisons: the future, the abstract, and your own idea. The right column commits entirely to the past, the specific, and the other person as the protagonist. That gap becomes, directly, the gap in the quality of information you get.

The Five Arrows for Rewriting Your Questions

Even so, memorizing the principles isn't enough; the moment you sit down to draft a question list, poison creeps in. So here's the check I use: the five arrows. Just rewrite the questions in your hands to point in these five directions, and the quality rises by an order of magnitude.

From opinion to behavior. Not "what do you think?" but "what did you do last time?" From future to past. Not "would you use it?" but "have you tried it?" From abstract to specific. Not "are you struggling?" but "last week, when, where, and with what were you struggling?" From yourself to the other person. Check every time whether you're asking for an evaluation of your own idea. And from a verbal Yes to a behavioral Yes. Not "could you pay?" but "how much have you paid recently?"

Let's look at this concretely with a fictional business. Suppose you're thinking of a SaaS called "GenbaAI" that, for site foremen at regional small and midsize construction and renovation companies, uses site photos and voice memos to automatically generate work logs and customer-facing reports. The target is foremen in their forties to sixties at companies with annual revenue of roughly 300 million to 3 billion yen. Many of them still write their work logs by hand or in Excel.

Here, let me line up the questions you tend to draft against their rewrites. "Writing work logs is a hassle, right?" is the worst kind of question, combining a leading premise with abstraction. Rewritten: "When and where did you write yesterday's log? From the moment you started to when you finished, how many minutes did it take?" "Don't you think it'd be handy if AI could automate it?" carries the double poison of a future hypothetical and your own idea. Rewritten: "Have you tried anything to make logs easier before, like an app, voice input, or asking an office clerk? Why didn't it stick, or why did it?" "Would you pay 5,000 yen a month?" only elicits a verbal Yes. Rewritten: "Have you paid for any tool or service around site management recently? How much, and whose budget did it come out of?"

This rewriting work is exactly where sparring with AI is fast. Paste in your own question list as is and have it diagnosed. Try dropping the following prompt into Claude or ChatGPT.

Below is a customer interview question list I wrote. Critique it from the
perspective of Rob Fitzpatrick's The Mom Test and rewrite it.

# Criteria (judge by these five arrows)
1. Opinion -> behavior (Is it "what did you do last time?" not "what do you think?")
2. Future -> past (Is it "have you tried it?" not "would you use it?")
3. Abstract -> specific (Is it "when/where/with what last week?" not "are you struggling?")
4. Yourself -> the other person (Are you asking for an evaluation of your own idea?)
5. Verbal Yes -> behavioral Yes (Is it "how much did you pay recently?" not "could you pay?")

# Output
A table "| Original question | Problem (which arrow it breaks) | Rewrite |" for every question.
Also, across the whole list, point out the percentage of "questions that end in Yes/No"
and "leading questions (isn't it, right?)."

# My question list
[paste here]

When you have AI help you detoxify like this, leading phrasing and abstractions you'd never notice yourself get flushed into the open. But the final judgment is yours. AI will mechanically flag "this question is leading," but only you know what you truly wanted to confirm with that question.

Digging for the Real Job with JTBD

Once you've pulled your questions toward the past and the specific, the next step is the direction you dig. What works here is JTBD, the Jobs to Be Done theory². The core can be stated in a single line: customers don't "buy" a product, they "hire" it. They hire the product to get a certain job done.

Jobs have three layers: functional, emotional, and social. Most entrepreneurs only see the functional job. "I want to make writing work logs faster" is a functional job. But satisfy only the function and it stalls at "handy" and doesn't sell. Money actually moves, usually, at the emotional and social layers.

A few classic examples. When someone investigated why milkshakes sold so well in a certain morning time slot, it turned out they were being hired as "a companion that relieves the boredom of a dull commute and staves off hunger until lunch"³. It wasn't improving the flavor but understanding this job that grew sales. A lacquered bowl can be hired not merely as tableware but as "a medium for handing love to your family." Some parents who hire a babysitter are hiring not just the function of childcare but a way to "dispel the guilt of wondering whether they're a bad parent." The real value lies outside the function.

Thinking about GenbaAI, the functional job is "write logs faster." But emotionally it might be "I don't want to be chased by logs at home in the evening; I want to spend time with my family." Socially it might be "I want the younger crew to see me as a foreman who runs a tight operation, not one who preaches the old-school grind." Dig this far and the words of your pitch change.

Translate the job you've dug up into the smallest unit, the Job Story. The format has three parts: "When (situation or trigger), I want to (motivation), so I can (expected outcome)." For example, "In the evening when I get back to the office after the site wraps up (When), I want the log to take shape just by taking photos (I want to), so I can leave on time and make it home before my kids go to sleep (so I can)." It's a handy vessel that translates a customer's actual words straight into a design blueprint for your business.

This three-layer JTBD breakdown is another place where having AI draft it broadens your thinking.

You are an expert in Jobs to Be Done. Break down the job my prospective customer
truly wants to "hire" within my product category into three layers: functional,
emotional, and social. Dig especially deep into the emotional and social roles
that most entrepreneurs overlook, because satisfying only the function stops at
"handy" and doesn't sell.

# Input
- Prospective customer: [write here]
- Product/service: [write here]
- How the customer currently gets that job done (the alternative): [write here]

# Output
1. Jobs for each of the three layers (2-3 each, in the customer's own words)
2. The psychological block that keeps it stuck at "handy," seen from the emotional/social roles
3. Three Job Stories (in the form "When [situation], I want to [motivation], so I can [outcome]," one each for Functional/Emotional/Social)
4. Two interview questions to validate each of those emotional/social roles

But let me stress this: the three layers AI produces are only hypotheses. Whether the emotion truly lies behind them can only be confirmed by actually meeting people. This connects seamlessly to the next step, How to Find Customer Problems and the Empathy Map. There are many tools for digging out the job, but the destination never changes: "a real person's real behavior."

Spotting the Polite No and Making Silence Your Ally

Even when you ask about the past and the specific and dig for the job, people wrap the truth in kindness. This is the Polite No. When phrases like these come up, take them as almost certainly a No: "That's interesting," "definitely, if the chance comes up," "we're a special case," "I'll take it back to my superiors," "personally, I think it's good." Each is just a way of declining without causing offense, and the wallet stays shut.

When a Polite No appears, being disappointed and backing off is the biggest waste. Steer the conversation back to past behavior here with "have you tried anything similar recently?" and the truth comes out. To the person who said "if the chance comes up," follow with "have you ever tried to create that chance before?" and you get "well, actually, we brought in a free one a while back, but nobody ended up using it," a real-life failure story. That one sentence is worth more than a hundred "that's greats."

One more thing that works in practice is silence. More than half of the insightful remarks come after a silence. Even when the other person seems to have finished answering, stay quiet for just three seconds and wait. They'll start to say something a step deeper to fill the gap. If the interviewer immediately piles on the next question, this deeper layer never surfaces. The ratio of talking should be two to eight between interviewer and interviewee. Let them do eighty percent of the talking. To that end, quietly insert three prompts: "why?", "specifically?", and "what else?"

That said, don't think silence is a cure-all. With a parent in the middle of raising kids, or a craftsman working with their hands on site, three seconds of silence won't hold. They're busy, and if a gap opens they'll bail with "well, that's it for me." For people like that, rather than pressing with silence, capping the time at thirty minutes, or asking bit by bit over several days via a messaging app, is far more genuine consideration. You match their life's rhythm. In the end, that's what draws out the deepest conversation.

The Validation Workflow and Division of Labor in the AI Era

Here's the main event for those who want to move a business forward with AI. The interview itself is human work, but everything around it gets dramatically faster with AI. The overall flow is this: first document your hypotheses, then record the interview, transcribe it with AI, run first-pass analysis with AI, and turn it into cards to check against your hypotheses. When you can run this loop in a single day, the speed of your validation changes.

The starting point is documenting your hypotheses before you go out to ask. Write down three hypotheses you want to validate, plus the disconfirming condition, "what would have to happen for me to know this hypothesis is wrong?", in advance. If you can't write it, that's a sign you're underprepared. For GenbaAI, the hypotheses might be "they spend 30 minutes or more a day writing logs," "they take logs home to write at night or on weekends," and "they feel formatting the customer report is a burden." The disconfirming conditions might be "log writing averages under 15 minutes" or "5 or more out of 8 say it isn't a particular burden." Only after deciding this far do you have AI create your interview guide.

You are a pro at designing customer interviews for new ventures. Based on the
business hypotheses I give you, create a 60-minute Problem Interview guide that
strictly follows the three principles of Rob Fitzpatrick's The Mom Test:
(1) ask about past behavior, not a hypothetical, (2) ask about specifics, not
abstractions, and (3) keep your own idea hidden for the first thirty minutes.

# My information
- Business idea: [write here]
- Prospective customer (ICP): [write here]
- Three hypotheses to validate: [write here]
- Disconfirming conditions that would show the hypotheses were "wrong": [write here]

# Output
1. A 60-minute agenda (time allocation for opening/context/deep-dive on past behavior/alternatives and pain/closing and referral request)
2. 15 questions, all in the past, specific, other-person-centered form like "when was the last time you..." and "what did you do then." Do not include a single "would you use it if it existed?"
3. One line on which hypothesis each question validates
4. An explicit note on when it's OK to talk about my idea
5. Three example countermoves for when the person says "that's great"

Once the interview is over, transcribe the recording with tools like Notta, Whisper, or tl;dv, and have AI analyze that transcript. What matters here is not letting it summarize or paraphrase, but having it handle the statements verbatim, word for word. Whether you can quote the customer's actual voice directly into a pitch or plan document pays off later.

You are a customer interview analysis assistant. Read the transcript below and
structure it. Do not summarize or paraphrase; handle the statements verbatim,
word for word.

# Context
- Interview type: Problem / Solution / Pricing, which one -> [specify]
- My prior hypotheses: H1 [ ] / H2 [ ] / H3 [ ]
- Prospective customer attributes: [ ]

# Output
1. Five verbatim quotes, ordered by strength of emotional movement (with one line each on "why it matters")
2. Job Stories, one each for Functional/Emotional/Social (When..., I want to..., so I can...)
3. A pain score (1-10) with three supporting statements (frequency, impact, and emotion each)
4. Judge each hypothesis as "confirmed / disconfirmed / neutral / not mentioned" with supporting statements
5. [Important] At least two statements that contradict the hypotheses (disconfirming). If there are none, state "none found" explicitly; do not silently omit
6. Point out whether any Polite No is included, such as "that's interesting" or "if the chance comes up"
7. Three questions to confirm in the next interview

# Transcript
"""
[paste here]
"""

And this may be the most powerful of all: deliberately casting AI as the disconfirmer, the devil's advocate. People interpret things in their own favor. Even I, if I let my guard down, take "that's great" favorably. So hand the results you've gathered to AI and have it disconfirm harshly.

You are the "disconfirmer (devil's advocate)" who breaks my confirmation bias.
I tend to take "that's great" favorably and interpret things in my own favor.
Deliberately and harshly disconfirm the interview results below.

# What I'm giving you
- Business hypothesis: [ ]
- Summary of interview results: [ ] (include favorable reactions, as is)

# Your tasks
1. Surface every "that's great / seems handy" as a Polite No and explain why it isn't a real Yes
2. Point out where I'm picking things up through confirmation bias (interpreting in my own favor)
3. Present three ways this hypothesis could be completely wrong, with concrete scenarios
4. Judge whether the results contain evidence of a "behavioral Yes (past payment, homegrown workaround, referral, prepayment)." If none, conclude "the pain is shallow"
5. Five Disconfirming Questions I must ask in the next interview

Having handed all this to AI, it's odd for me to say so, but get the division of labor right, no matter what. What AI can do goes as far as transcription and first-pass analysis. "What pain does this person truly feel?" and "so, what do I go verify next?" can only be decided by a human. The hypotheses AI produces are no more than a rough draft of the map. The sound of the wind blowing through a bamboo grove isn't on the other side of the screen. Meet people, listen, interpret. The moment you let go of that core, the business becomes someone else's affair.

Turning this sense of "keeping AI as a partner while never letting go of judgment" from an individual's knack into a team's standard equipment is the work we do together in our AI consulting service, WARP. Rather than handing out prompts and calling it done, we design together which task at which stage of building a business to entrust to AI, and where humans keep their grip.

A Collection of Anti-Patterns to Avoid

Finally, let me gather the landmines people frequently step on in the field. If even one rings a bell, fix it in your next interview.

Explaining your own idea at the very start. This puts the other person in evaluation mode and starts the flattery. Sealing it for the first thirty minutes is the iron rule. Asking about the future with "would you use it if it existed?" People paint their future selves as impressive, so it's almost always a lie. Asking in the abstract with "are you struggling?" You'll only get abstract answers. Mistaking "that's great" or a Polite No for a real Yes and advancing to the next stage. This is the most expensive mistake of all. Picking up only the statements that fit your hypothesis and ignoring the disconfirming ones. This is the textbook confirmation bias, which is exactly why, during analysis, you must always ask yourself "what statements contradict this hypothesis?"

There are other traps too: the interviewer talking too much and failing to keep the two-to-eight ratio, being unable to wait through silence and killing the deeper layer, summarizing statements into your own words and losing the verbatim, and being satisfied after asking only acquaintances and relatives. Ask acquaintances at most three people. Beyond that, put your questions to strangers who properly match your ICP. Friends are too kind, and they'll say exactly what your mother would.

And the trap of over-delegating to AI. Having AI answer even "what's the real emotion behind this Job?" and "what do I go verify next?", then feeling like you understand. AI is a ten-times booster, not the owner of the business. Judgment, at least, must always stay in your own hands.

Summary and Three Actions for Tomorrow

We've run through the craft of customer interviews in one go, with the Mom Test as the axis. Finally, let me leave you the key points paired with your next step.

• "That's great" is not evidence. Evidence is a behavioral Yes: past payment, a homegrown workaround, a referral, a prepayment.
• Rewrite your questions with the five arrows: opinion to behavior, future to past, abstract to specific, yourself to the other person, verbal Yes to behavioral Yes.
• With JTBD, dig the three layers of functional, emotional, and social, translate them into Job Stories, and carry them over into your business blueprint.
• Use AI for transcription and first-pass analysis, and as the disconfirmer. Keep interpretation and the next move in human hands.

Just three concrete actions for tomorrow. First, pick three people close to your ICP from the contacts on your phone. Next, message one of them to say "not a sales pitch, but there's something I'd love to learn from you." And print out the NG/OK quick reference from this article and stick it on the wall in front of your desk. That alone will change your next conversation.

The big-picture view of building a new business is laid out in the Complete Guide to the New Business Framework, how to decide the customers you target in How to Decide Your Customer Segment, and how to organize the problems you've unearthed in How to Find Customer Problems and the Empathy Map. Read them together and the flow connects, from asking questions to landing what you heard into the business, rather than ending at "we asked."

If you'd like to spar in concrete terms, tailored to building your own business, on everything from designing questions that draw out the customer's true feelings to drawing the line on how far to use AI, please tell us about it at a WARP one-on-one consultation. And if you'd first like to measure where your company stands on using AI, start with the AI Literacy Check.

---

References

Footnotes

1. The original source of the Mom Test. Rob Fitzpatrick, The Mom Test: How to Talk to Customers & Learn if Your Business is a Good Idea When Everyone is Lying to You (2013). ↩ ↩²

2. Jobs to Be Done. Clayton M. Christensen et al., "Know Your Customers' Jobs to Be Done," Harvard Business Review (September 2016), and Competing Against Luck (2016). ↩

3. The milkshake case. A representative case study of Jobs to Be Done theory, widely introduced in the lectures and writings of Clayton M. Christensen. ↩