We spent thirty minutes explaining a Miro board to a client.
Not learning about their problems. Not mapping their workflow. Not asking a single useful question.
Thirty minutes explaining the tool we brought to capture their answers.
The client sat there politely. Nodding. Waiting for us to finish.
We never did find out what was actually keeping them up at night. We ran out of time.
That was our introduction to discovery. And it taught us more about building AI products than any framework, any course, any product book ever did.
The mistake most PMs make before the call even starts
When you don't know what you're doing, you over-prepare.
You build structured templates. You create elaborate frameworks. You prepare questions in neat categories, pain points, current workflow, desired outcomes. You show up with a colour-coded Miro board and an impact versus effort matrix and an agenda that covers everything.
And then you spend the entire meeting explaining your agenda instead of listening to the customer.
The client came to tell you about their world. You made them sit through a tour of yours.
The elaborate setup becomes the meeting. The customer's real problems never make it into the room.
Every PM does this. We confuse preparation with understanding. We mistake structure for insight. We walk in with so many hypotheses to validate that we stop being curious and start being selective, only really hearing the things that confirm what we already thought.
That's not discovery. That's a very expensive way to stay wrong. And the worst part? You leave the meeting convinced you did a great job.
The only rule that actually matters
After enough sessions, enough missed insights, enough products built on half-understood problems, we landed on one principle that cut through everything else.
You know you've done discovery wrong when you are doing more talking than the customer.
That's it. One rule. Embarrassingly simple.
Client talks 80% of the time. You talk 20%. If that ratio flips at any point — if you're explaining, presenting, walking through anything you prepared — stop. You've lost the thread.
For product managers specifically, this is harder than it sounds. We go into customer conversations with things we want to validate. Features we want reactions to. Roadmap decisions we're quietly hoping to confirm. And without realising it, the entire conversation bends toward our agenda instead of theirs.
You leave the meeting feeling productive. You go back to your desk and write up your notes. You put together a requirements doc that feels thorough and well-evidenced.
And somewhere in there, the actual problem, the one the customer almost told you about but never quite got to, sits untouched.
Discovery done right feels different. The customer is talking and you're genuinely surprised by half of what comes out. That surprise is the signal. That's where the real product work starts.
The one tactic that changed everything
Ask the customer to share their screen.
Not metaphorically. Literally. Right now, in this meeting, can you share your screen and show me how this actually works?
Nine times out of ten they'll say yes. And the moment they do, something shifts. They stop describing their workflow and start showing it. You stop listening to a summary and start seeing the reality.
And the reality is almost never what the summary said it was.
There are workarounds nobody mentioned. Spreadsheets running alongside the tools they're supposed to be using. Manual steps that three people do every day that nobody documented because everyone assumed someone else was going to fix it eventually.
The real workflow and the documented workflow are never the same. The screen share shows you the real one.
We once watched a customer navigate their CRM process and noticed they were running everything through a Google Sheet alongside Salesforce. You already pay for Salesforce. Why the Google Sheet?
The answer: "It's free."
Two words. Said almost apologetically. Like they knew it sounded bad.
But that answer told us everything, about their budget sensitivity, about how they evaluate tools, about what our solution needed to cost to even get a serious conversation. None of that was in the brief. None of it would have come up if we'd stuck to our prepared questions.
It only came out because we asked why at exactly the right moment.
Ask why. Then ask it again.
Once the customer is talking, your only job is to listen and ask why.
Not aggressively. Curiously. Like you genuinely can't believe the process works this way and you want to understand it better.
Why does the process look like this? Why hasn't anyone changed it? Why are you using that tool if it causes this much friction every single week?
The first answer is what the customer thinks the problem is. The third answer is what the problem actually is. The fifth answer is what you build the product around.
Most PMs stop at the first answer. They write it down, nod, move to the next question. They leave with a list of symptoms and call it requirements.
Every layer of why peels back something the customer has normalized, a workaround they stopped noticing, a constraint they assumed was permanent, a trust issue with a previous tool that still shapes how they behave today.
And when you don't understand the industry jargon, and you often won't, don't pretend. Ask them to clarify. Summarize back what you heard. Say "just want to make sure I'm understanding this correctly, what I'm hearing is..." and let them correct you.
That moment feels vulnerable. It's actually your sharpest tool. Because when you force someone to explain something simply, they often realize for the first time exactly why it's broken.
What AI quietly changed about all of this
There's one part of discovery that used to make deep listening almost impossible.
Note-taking.
You'd be trying to hear what someone was saying while simultaneously writing down what they just said. You'd miss the tone, the hesitation, the thing they mentioned quietly like they weren't sure they should say it. Half your brain was in the conversation. The other half was doing admin.
You'd end up with detailed notes and a shallow understanding. The words were there. The meaning was somewhere on the floor of that meeting room.
Now your notetaker handles the transcript. You just listen.
That sounds like a small thing. It isn't.
AI didn't just change what we can build for customers. It changed how we learn what to build in the first place.
Full presence in a discovery call, no pen, no keyboard, just watching their screen and asking why, is a completely different experience. You catch things you never would have caught before. And those things become the product decisions that actually matter.
What this taught me about product management at an AI startup
Every AI agent we've built that actually works started with a discovery conversation where we shut up long enough to find the real problem.
Product management in a startup doesn't happen in a room somewhere removed from the customer.
You're in the discovery call. You're watching the screen share. You're the one asking why when something doesn't add up. Every observation you make in that conversation is product data.
The Google Sheet sitting next to the expensive CRM — product signal.
The workaround that's been running for three years — product signal.
The reason they won't hand control of a process to an automated system — the most important product signal of all.
You can have the best AI model in the world. You can have the most elegant architecture, the cleanest UI, the most impressive demo. And if you built it around the wrong problem, the stated problem instead of the real one, none of it matters.
Discovery isn't something that happens before product work begins.
Discovery is where product work begins.
The best product requirement I ever wrote came from watching a customer share their screen and asking why three times. No template. No framework. No Miro board.
Just curiosity. And the discipline to stay quiet long enough to actually hear the answer.
