Customer Discovery 2.0: Using AI to Get Better Interviews

AI didn’t kill customer discovery — it exposed who was bad at it

There’s a quiet anti-pattern spreading among founders:

“We don’t need interviews anymore — we use AI.”

This is backwards.

AI didn’t replace customer discovery. It raised the standard for it.

The founders who are winning right now are still talking to customers — but they’re doing it with more structure, less bias, and far faster synthesis.

Customer Discovery 2.0 is not about fewer conversations. It’s about better ones.

The old discovery problem: vibes, bias, and selective memory

Traditional customer discovery fails for three reasons:

1. Founders ask leading questions
2. Founders hear what they want to hear
3. Founders remember anecdotes, not patterns

AI didn’t introduce these problems — it made them more visible.

When you re-read transcripts, cluster feedback, and compare conversations side by side, it becomes obvious how often “positive feedback” was meaningless.

Discovery 2.0 fixes this by separating:

• collection (human)
• synthesis (AI)
• judgment (founder)

Each has a role. None can replace the others.

What customer discovery is actually for

Customer discovery is not about:

• selling your idea
• pitching features
• collecting compliments

It is about answering three questions:

1. What problem does this person already experience?
2. How painful is it compared to everything else in their day?
3. What do they do today instead?

If you can’t answer those without referencing your solution, discovery failed.

The Discovery 2.0 workflow

Here’s how strong founders run discovery today.

Step 1: Frame the interview as archaeology, not persuasion

Your job is not to convince. Your job is to excavate reality.

Good framing sounds like:

• “I’m not selling anything.”
• “This may not become a product.”
• “I’m trying to understand how you work today.”

This disarms politeness and reduces false positives.

Step 2: Ask about the past, not the future

People are bad at predicting what they’ll do. They’re very good at describing what they already did.

Bad question:

“Would you use a tool that does X?”

Good questions:

• “When was the last time this happened?”
• “What did you do next?”
• “What broke or felt frustrating?”

Future-facing answers lie. Past behaviour tells the truth.

Step 3: Record everything (then get out of the way)

AI shines here — but only after the conversation.

Use tools to:

• transcribe interviews
• tag emotions and objections
• cluster repeated phrases
• surface contradictions

Do not rely on your memory. Your brain edits in real time.

How AI actually helps (when used correctly)

AI is exceptional at synthesis, not sense-making.

Use it to:

• identify repeated pain language
• group objections across interviews
• detect emotional spikes (stress, frustration, relief)
• summarise “what people actually do today”

But never ask AI:

“Is this a good idea?”

Instead ask:

• “What problems appear most frequently?”
• “Where do users sound annoyed or resigned?”
• “What workarounds are people maintaining?”

AI shows you patterns. You decide what they mean.

The biggest Discovery 2.0 mistake: synthetic users too early

AI personas and simulated users can be useful — but only after real discovery.

If you generate personas before talking to humans, you’re just role-playing with yourself.

Synthetic users are helpful for:

• stress-testing messaging
• exploring edge cases
• preparing follow-up experiments

They are not a substitute for:
• hearing frustration
• understanding context
• observing real constraints

Use AI to amplify truth, not invent it.

A simple discovery quality check

After 5–10 interviews, you should be able to answer:

• Who feels this pain most intensely?
• When does it show up?
• What do they sacrifice to avoid it?
• Why existing solutions fall short
• What would make them change behaviour

If your notes are mostly feature ideas, restart.