← Back to Writing

If you are doing "customer discovery" but it feels like random calls, vague notes, and fake validation, you are not alone.

Most teams fail for one reason: they collect opinions instead of evidence. Customer discovery is not "Do you like my idea?" It is a system for reducing uncertainty about who your customer is, what problem is real, and what you should build next.

This guide gives you that system.

What customer discovery actually is (and what it is not)

Customer discovery is a loop:
Assumption → Interview → Evidence → Decision → Next assumption.

It is not:

• Asking users to design your product
• Pitching your idea
• Collecting feature requests
• Hunting for compliments

Your job is to learn. You are looking for clues that confirm or deny your assumptions.

Start with assumptions or you will waste weeks

Founders, let's face it - we love jumping into calls.

The problem is simple: without assumptions, you do not know what you are trying to learn. Write 3–5 assumptions as your discovery backlog, keep them short.

The minimum assumption set

• ICP: who has the pain
• Problem: what pain is real and frequent
• Workflow/Alternatives: what they do today
• Blockers: what stops them from solving it
• Differentiation: why you vs status quo
• Monetization: what "worth paying" means
• Acquisition: how you will reach them

For each assumption, add: "If this is false, we will…".

Example:

• Assumption: "Ops managers track vendor compliance in spreadsheets and hate audits."
• If false, we will: "Stop building audit features and focus on onboarding automation."

Learn more on resources like talk2user.com!

Pick what to test next (so discovery compounds)

You only need one rule: test the riskiest assumption first. Risk means if wrong, it kills or radically changes your plan.

A simple status set that works:

• Testing (we do not know yet)
• Confirmed (enough consistent evidence)
• Rejected (enough consistent contradiction)

Do not overthink scoring early. Focus on whether evidence is repeating.

Who to talk to (and how many interviews you actually need)

B2B: start in the middle. Talk to people who feel the pain daily and will actually answer you. Mid-level managers and ICs are usually better than C-suite early on.

Volume matters more than you think

You are not looking for a single "yes." You are looking for patterns.

• Consumer: expect noisy behavior. Talk to 70–100 people to see patterns.
• Early B2B: plan for 20–30 as a bare minimum to get signal (and cancellations).

If you have done 7 calls and feel "validated," you are probably just excited.

How to find interviewees fast (without being annoying)

Use three channels

1. Your network: friends, investors, operators. Warm intros compound.
2. Where they already hang out: Slack groups, Discord, Reddit, conferences.
3. LinkedIn targeting: role + company type + geography.

The outreach message that gets replies

Ask for advice. Be clear you are not selling.

Template:

"I am researching how [role] handles [job]. I would love to learn from your experience. Could I ask you a few questions on a 20-minute call? I am not selling anything."

The referral question that scales discovery

End every call with:
"Is there anyone else you think I should talk to?"

Aim for 1–2 intros per interview. That is how you go from "hard" to "automatic."

Designing questions that do not produce lies

Most bad discovery comes from one mistake: You ask for opinions about the future. People are terrible at predicting future behavior. They are great at describing what they already did. So your default question format is:

"Tell me about the last time you…"

Rules that keep questions clean

• Ask open-ended questions. Talk little. Let them talk.
• Prefer who / what / why / how. Avoid is / are / would / do you.
• Avoid "magic wand" questions. Customers cannot design solutions well.

A repeatable question stack (use this every time)

Pick one assumption. Then ask:

1. Trigger story
   "Tell me about the last time [problem] happened."
2. Workflow
   "Walk me through what you did, step by step."
3. Pain + intensity
   "What was hardest? How often does this happen?"
4. Current alternatives
   "How do you solve it today? What did you try before?"
5. Cost of the problem
   "What does this cost you: time, money, risk, stress, reputation?"
6. Decision criteria
   "When you choose a tool/process here, what matters most?"
7. Constraints and blockers
   "What stops you from fixing this properly today?"
8. Close
   "What should I have asked you that I did not?"

That stack reliably produces real data.

Running the interview (the 30-minute playbook)

Before

• Prepare an outline, not a script.
• Record if allowed. Get transcript.
• Block 30 minutes. No distractions.

During

Your only job: listen more than you talk.

Tactics that work:

• Do not pitch your product. Stay in their world.
• Double down with "why" when something matters.
• Repeat back what you heard. Let them correct you. Listen!
• Watch for emotions and workarounds. Hacks signal they have real pain.
• Avoid hypotheticals. Ask what they do now, or did last time.

After

Insights come from patterns, not one call.

• Write key takeaways immediately.
• Send a thank-you and keep them updated. Early interviewees become early users.
• Share notes with your team. Do not trap learning in one person's brain.

Synthesis that leads to decisions (not a notes graveyard)

Most teams fail here. They do calls, collect notes, and still do not know what to do next.

Fix it by forcing structure.

For each assumption, write four things

1. Evidence bullets: 2–4 bullets grounded in what was said
2. Contradictions: explicit counter-evidence (with quotes)
3. Known vs Unknown: what you learned vs what is still unclear
4. Next test: the next interview focus

If you do this every time, your discovery stops being random. It becomes a compounding system.

Track "signal," not vibes

A simple per-assumption label works:

• Supports
• Mixed
• Contradicts
• No evidence

Don't forget about sample size. If only 2 people said it, it is not relevant.

The mistakes that quietly ruin discovery

These show up in almost every startup:

• Pitching instead of learning
• Leading questions
• Yes/no questions without follow-ups
• Asking about future behavior
• Talking too much
• Ignoring negative reactions
• Interviewing the wrong people
• Group interviews
• Not synthesizing what you learned

If you fix only one: stop asking "Would you use {your product description}?"

Close

Customer discovery is not about being "good at interviews." It is about building a system that prevents you from wasting months building the wrong thing.

If you try one thing after reading this, do this:
Write 3 assumptions, then schedule 5 calls this week to kill the riskiest one.