Run Customer Discovery Interviews That Actually Inform Your GTM

Most founders skip this step or do it badly. The skipped version produces ICPs and value props pulled from imagination. The bad version produces "we asked 30 people if they liked the idea and they all said yes" — which is the most expensive non-signal in startup history.

Done right, twenty hour-long conversations with the right people in two weeks gives you the raw material for every other artifact in Day 1: Position. The ICP writes itself. The value prop's exact phrasing comes from a transcript. The pricing range is bounded by what you heard. Skip this and you guess. Do it and you watch the rest of the week move at 3×.

Why Most Discovery Interviews Fail

Three failure modes account for almost every botched interview round:

Asking about the future. "Would you use a tool that does X?" People are uniformly terrible at predicting their own future behaviour. The answer is always "sure, that sounds useful," and it correlates with nothing real.
Pitching instead of listening. Founders cannot resist describing the product. The interview becomes a demo. The customer becomes polite. You learn nothing.
Talking to the wrong people. Interviewing your friends, your fellow founders, or the most enthusiastic respondents on a waitlist will give you encouraging answers and a misleading picture. You need to talk to people who actively have the problem you are trying to solve, ideally people who currently pay someone else (or do it themselves the hard way) to solve it.

Rob Fitzpatrick's The Mom Test is the canonical short read on this. The principle: ask about the customer's life, not your idea. The technique: probe specifics, not generalities. Both are easier to write than to do.

What You Are Building This Week

By the end of this work, you have:

15–25 completed 30–60 minute interviews with people who match (or test) your ICP hypothesis
A single transcript repository, searchable, tagged by theme
A "voice of customer" doc with the exact phrasings buyers use to describe their pain, their current workarounds, and their desired outcomes
A bounded pricing range (not a single number — a range you heard)
A list of the 3–5 features and the 1–2 anti-features that emerged
Three pre-validated buyer quotes you can use, with permission, on the landing page

Everything in this list directly feeds the ICP, Value Proposition, Pricing Strategy, and Landing Page Copy work.

1. Recruit the Right People

The hardest part of discovery interviews is not the interview — it is finding the people. Cast a wider net than you think you need; expect a 10–20% positive response rate.

Three pools, in order of signal quality:

Pool A: People who actively have the problem (best signal)

Find them by searching for public statements of the problem. Reddit posts, X complaints, LinkedIn posts venting about the existing tools, Hacker News threads under your category. Each one is a candidate. Reach out:

Subject: Saw your [post / comment / tweet] about [specific problem]

Hey [Name] — I read your [link] about [the specific frustration they expressed]. I'm researching exactly that problem for [one-line context], and you sound like you've thought about it more than most. Would you be open to a 30-minute conversation? I'm not selling anything; I just want to understand how you currently deal with it.

Happy to send a $50 [Amazon / Stripe gift] for your time.

The $50 incentive is fair — you are taking 30 minutes of a working person's day. It does not skew the conversation if the person had a real frustration.

Pool B: People who pay for an adjacent solution (good signal)

Customers of incumbent tools you would compete with. Find them through G2 reviews, Capterra reviews, LinkedIn (people who list the incumbent in their stack), or partnerships with adjacent SaaS founders who can intro a few of their power users.

Pool C: People who fit demographically (weakest signal)

Roles, industries, company sizes that match the ICP hypothesis without active evidence of the problem. Useful as a control group. Recruit them through LinkedIn outreach, Polywork-style platforms, or industry communities. Do not over-rely on this pool — you will get polite confirmation of your hypothesis and miss every signal that actually matters.

Aim for: 60% Pool A, 30% Pool B, 10% Pool C.

2. Set Up the Interview Logistics

Before the first call:

A scheduling link (Cal.com, SavvyCal, Calendly) with 45-minute slots. Pad in 15 minutes between for notes — not back-to-back.
A consent template sent in the calendar invite: "I'd like to record this call for my own notes. The recording stays private; I will only quote you with explicit permission."
A recording tool that transcribes automatically (Granola, Tactiq, Otter, Notion AI, fireflies.ai). The transcript is the artifact, not your scrawled notes.
A single doc for all interviews — one tab per interview, plus a synthesis tab. Notion, Coda, or a markdown folder all work; tooling is not the bottleneck.

Block 5 hours per week for interviews + 5 hours for synthesis. Twenty interviews in two weeks at this cadence is realistic.

3. The Interview Structure

Forty-five minutes, broken into four phases. The phasing matters: skip phase one and the answers you get to phase three are uncalibrated.

Minute 0–10 — Their world

Before you mention the problem, before you mention your product, understand the texture of how they work. Open-ended questions:

"Walk me through what you've been working on this week."
"When was the last time [related-but-broader thing] came up in your day?"
"What part of your job takes more time than it should right now?"

You are listening for organic mentions of the pain you are trying to solve. If they bring it up unprompted, you have a strong signal. If they never get there, that is also a signal — the pain is less salient than you assumed.

Minute 10–25 — The specific problem

Now narrow to the problem space. Still no mention of your product. Probe for specificity:

"When did you last have to [specific instance of the problem]?" — get them in time, not abstract.
"Walk me through what you actually did. Step by step." — get them in specifics, not summary.
"What did that cost you — time, money, customers, sleep?" — quantify.
"What did you try before settling on [current solution]?" — competitive landscape from their POV.
"What's the most painful part?" — let them rank.

Two ratios you are watching for:

Pain frequency: how often does this happen? Daily, weekly, monthly, never?
Pain magnitude: how much does it cost when it does happen?

The product opportunity is where both are high. If pain is frequent but trivial, no one will pay. If it is severe but rare, no one will remember to buy.

Minute 25–40 — Their workaround and willingness

Only now, with the texture of their world and the specifics of the problem in hand, get into solution and pricing.

"What are you using to handle this today?"
"Why that, and not [obvious alternative]?"
"What would have to be true for you to switch?"
"If a tool [described in plain words, not your product name, in their own phrasings from earlier] existed, what would it be worth to you per [month / project / outcome]?"
"Who would have to approve a purchase like that? Is that you?"

Watch for the hesitation between "this would be worth $X" and "I would actually pay $X." The first is a willingness signal; the second is a buying signal. They are not the same.

Minute 40–45 — Wrap and ask for two more

Always end with two specific asks:

"Would you be open to seeing what we're building once it's ready, and giving me 15 minutes of feedback?" (Builds your beta list.)
"Who else do you know who has this problem worse than you do?" (Builds your interview list. Snowball recruiting beats cold outreach 5×.)

4. The Five Questions That Carry the Round

If you only got to ask five questions, these five carry the most information. Use them as a fallback when an interview is going sideways and you need to recover:

"Walk me through the last time [problem] happened. What did you do, step by step?"
"What did that cost you?"
"What did you try before what you do now?"
"What would have to be true for you to switch?"
"Who else do you know dealing with this?"

Notice what is missing: any version of "would you use it" or "do you like the idea." Those questions belong in usability testing, not discovery.

5. Synthesise as You Go, Not at the End

The single biggest mistake after recruiting is to do all 20 interviews and then sit down to synthesise. Memory degrades, themes blur, and by interview 20 you have lost interview 3. Synthesise after every conversation, in 15–20 minutes:

For each interview, capture:

1. **Five-line summary** of who they are, role, company, current setup
2. **Pain quotes (verbatim)** — every sentence where they used their own words to describe the problem. These become landing-page copy. Quote literally, do not paraphrase.
3. **Workaround** — what tool, process, or hack they use today. Names of competitors / alternatives mentioned.
4. **Pain frequency and magnitude** — your read, on a 1–5 scale each. Note your reasoning.
5. **Willingness signal** — what they said about cost, value, willingness to switch. Verbatim where possible.
6. **Buy authority** — could this person sign the cheque? Who else needs to approve?
7. **Open question** — anything you didn't ask but wish you had. Add to the next interview's script.

After every five interviews, run a synthesis pass:

Across the last [5 / 10 / 20] interviews, surface:

1. **Themes mentioned by 3+ interviewees** — these are real signals. List the verbatim phrasings.
2. **Themes mentioned by 1–2 interviewees** — interesting, possibly niche, possibly noise.
3. **Themes I expected but never heard** — the most informative output. Where my hypothesis was wrong.
4. **Pricing range I heard** — lowest "this would be worth" to highest, plus the most-common range.
5. **Anti-features** — things people explicitly said they did NOT want, or warned against.
6. **Strongest verbatim quotes** for each major theme — for use in landing copy with permission.

Output as a single doc I can paste into the [Value Proposition](value-proposition.md) and [Landing Page Copy](landing-page-copy.md) work this week.

The synthesis doc is the most valuable artifact in your business at this stage. Keep it living — every new interview updates it.

6. Read the Negative Signals Honestly

The hardest part of discovery is hearing things you did not want to hear, and not flinching past them. Watch for:

Politeness. "Yeah, that sounds useful" with no follow-up enthusiasm is a no.
Solutions in search of a problem. If interviewees describe a different problem than the one you set out to solve, that is the signal — pivot toward what they are actually saying.
Workaround already adequate. "Honestly, I just use a spreadsheet and it's fine" means the willingness to pay is low even if the demographic fit is high.
Buyer ≠ user. A user who loves the idea but cannot buy without three layers of approval is a long sales cycle. Account for it.
The "interesting" trap. Interviewees frequently say "that's an interesting idea" when they mean "I do not have this problem." Treat interesting as a polite no.

If 15 interviews in, the picture you are seeing differs from the picture you started with — update. The product positioning that ships in week three is allowed to be different from the one you walked into the round with. That is the entire point.

7. Common Failure Modes

Confirmation-mode interviews. You wanted to validate a hypothesis, you asked leading questions, the customer agreed, you concluded the hypothesis was right. The fix: every interview, ask at least three questions you genuinely cannot predict the answer to. If you can predict every answer, you are not interviewing — you are pitching.

Sample size of three. Three interviews is a great start; it is a terrible decision basis. Anything you decide on the basis of three conversations should be tagged "tentative" until twenty more conversations support it.

No-show waste. Recruit 30 people to get 20 interviews. The 33% no-show rate is normal. Build the buffer into your timeline.

Friend interviews. People who like you give you false signal. They want you to succeed; they will say what helps you, not what is true. Use friends sparingly, and only as warm-ups to test the script before the real interviews.

Skipping synthesis. Twenty interviews and no synthesis is twenty interviews wasted. The transcripts are inputs; the synthesis is the artifact. If you only have time for one, choose synthesis and do fewer interviews.

Quoting without permission. Every quote you use on the landing page or in a pitch deck needs explicit permission, with attribution scope. A separate one-line ask after the interview ("Mind if I quote that one line about X anonymously / with your name?") suffices and protects the relationship.

How This Feeds the Rest of Day 1

Each Day 1 artifact has a direct dependency on what you learn in this work:

Artifact	What discovery interviews supply
Ideal Customer Profile	The seven attributes are filled with names, examples, and observed behaviour rather than guesses
Value Proposition	The exact phrasings of the pain — your copy quotes the customer back to themselves
Pricing Strategy	A bounded range from real "what would this be worth" answers, not a number pulled from a competitor's site
Landing Page Copy	Three pre-permissioned customer quotes, the verbatim pain language, the rejected anti-features
Brand Voice	The vocabulary your buyers actually use — anchored in transcripts, not your aspirations
Pitch Deck	A real customer story arc to anchor the deck, not a generic problem slide

Founders who skip discovery work do all six of those artifacts on imagination. They cohere with each other but not with the market. Founders who do discovery first watch the rest of the week assemble itself.

Deliverable

15–25 completed interviews with people in Pools A and B (with at least 60% Pool A)
A transcript repository, tagged by theme, searchable
A voice-of-customer synthesis doc updated after every interview
A pricing range bounded by what was heard
3–5 pre-permissioned customer quotes ready for the landing page
A list of validated features and explicit anti-features
A snowball recruit list for round two (post-MVP usability testing)

What's Next

With discovery in hand, move to Define Your ICP for an AI Company — armed with the names, behaviours, and quotes you collected here. Then Craft Your Value Proposition by literally quoting the verbatim pain language back at the buyer.