Discovery Used to Take Weeks. Now I Do It While Users Are Still Typing.

I used to think good discovery meant more user interviews

At Finvestfx, I had a problem. We were a treasury management SaaS serving 20+ enterprise clients, and every implementation felt like starting from scratch. Different pain points, different workflows, different languages (literally, half our clients spoke in Hindi and Kannada mixed with English).

So I did what every PM handbook says. I scheduled user calls. Took notes. Built journey maps. Synthesized themes in a Miro board that looked impressive but took me four days to make sense of.

The discovery cycle was 3-4 weeks minimum. By the time I'd synthesized insights, half the context was stale.

Then I started using AI tools at Sonic Linker, and I realized I'd been doing discovery in slow motion.

AI doesn't replace talking to users. It makes you talk to 10x more of them.

Here's what actually changed. At Sonic Linker, we were building an AI SaaS platform from scratch. Founding team, no playbook, moving fast. I needed to understand what our early users were struggling with, and I needed to understand it *now*.

I started feeding our support tickets, onboarding call transcripts, and Slack messages into Claude. Not to replace analysis. To surface patterns faster than I ever could manually.

In one afternoon, I processed 6 months of user feedback. Claude flagged three things:

1. Users kept asking "how do I know this is working?" in 14 different ways
2. The word "confused" showed up 47 times, almost always in the first 10 minutes of using the product
3. Power users never mentioned our core feature. They talked about a side workflow we'd built as an afterthought.

I would've caught #1 eventually. Maybe #2. I would've completely missed #3 because I was too close to the product.

The next week, I had follow-up calls with 8 users. But this time, I wasn't fishing for problems. I was validating hypotheses. The conversations were sharper, shorter, and way more useful.

Speed is great. But the real shift is *breadth*.

I used to do discovery in batches. Interview 5 users. Synthesize. Build. Repeat.

Now I'm running discovery continuously. AI lets me analyze every single customer interaction, not just the ones I have time to manually review. Support tickets, sales calls, product usage logs, even the stuff users say in Slack when they think we're not listening.

At NJ Group, I was coaching 60 insurance advisors and IFAs on product adoption. Before AI, I'd pick 10 of them, do deep dives, and assume the insights applied to the other 50. It mostly worked, but I was always guessing.

Now? I can see patterns across all 60. Who's struggling with the same workflow. Who's found a workaround we didn't design. Who's about to churn because they haven't logged in for 9 days.

I'm not smarter. I just have better peripheral vision.

The hard part isn't the AI. It's knowing what questions to ask.

Here's where most PMs screw this up (I did too, initially). They think AI is a magic insight machine. You dump data in, insights come out.

That's not how it works.

AI is a tool for exploring hypotheses faster. But you still need the hypotheses. You still need to know your product, your market, and your users well enough to ask the right questions.

When I was analyzing support tickets at Sonic Linker, my first pass was useless. I asked Claude to "summarize the main pain points." It gave me generic platitudes. "Users want better onboarding." Great, so does everyone.

The second pass was better. I asked: "Show me all tickets where users mentioned success or failure in the first sentence. What were they trying to do right before they reached out?"

That question came from experience, not AI. The AI just helped me answer it 50x faster.

What I actually do differently now

Discovery at Sonic Linker looks like this:

1. I spend 30 minutes every Monday feeding the previous week's support tickets, user calls, and analytics anomalies into Claude
2. I look for patterns, contradictions, and outliers
3. I write down 3 hypotheses worth testing
4. I talk to 5-8 users that week. Not to discover. To validate.
5. I ship small changes fast, and the cycle repeats

The whole process takes half the time it used to. But the quality is better because I'm not guessing anymore. I'm testing.

The real unlock: AI gives you permission to be wrong faster

I used to treat discovery like research. Slow, careful, comprehensive. I was terrified of missing something.

Now I treat it like iteration. I make a guess, test it with AI-assisted analysis, validate it with a few users, and move on. If I'm wrong, I find out in days, not months.

At Finvestfx, I once spent 6 weeks building a feature based on 4 user interviews. It flopped. If I'd had AI tools back then, I would've caught the warning signs in week 2.

AI didn't make me better at discovery. It just made me faster at being wrong. And in product, that's the same thing as being right.