How I Used Claude + Mixpanel to Stop Designing on a Guess

You're about to redesign something.

A stakeholder told you it "feels off."
You have your design instincts.
Maybe a heatmap if you're lucky.
Maybe some user feedback from six months ago.

That's it. That's what most redesigns start with.

And that's exactly how designers end up solving the wrong problem.

Here's how it usually goes: you make an educated guess about where the friction is, you spend two weeks in Figma, you ship it, and then you wait. Maybe it helped. Maybe it didn't. You're not really sure. Neither is anyone else.

That's not a design process. That's a hope strategy.

I found a better way this week. It took less than an hour.

I was starting a redesign on a specific section of our app. Before I opened Figma, I ran a workflow I hadn't tried before.

Step 1: I connected Claude to our GitHub repo and asked it to inspect the codebase for that specific page. Specifically, which Mixpanel tracking events were already firing on it.

Claude surfaced the exact event names. No guessing what data existed. No digging through docs. No asking the engineering team to pull a report. Just the tags, right there.

This part alone saved me an hour. If you've ever tried to figure out what's actually being tracked in a mature codebase, you know what I mean.

Step 2: I took those tag names into Mixpanel's new agent. It's an open text field where you prompt it in plain English. I asked it to show me the high-usage funnels and drop-off points on that flow using the exact event names Claude had just surfaced.

No SQL. No analyst request. No waiting until next sprint.

Two back-and-forths. That's it.

I had actual user engagement data for the exact section I was about to redesign.

Here's where it got interesting.

My initial assumption about where the friction was?

Wrong.

The users weren't dropping off where I thought. They were getting stuck somewhere else entirely. The section I was planning to redesign wasn't even the bottleneck. The real problem was one step earlier in the flow.

If I had skipped this and gone straight to Figma, I would have spent two weeks building a polished solution to a problem that wasn't the problem.

One path is two weeks of work in the wrong direction.
The other is a 45-minute conversation with your own data.

Why this works (and why most designers skip it)

The data isn't missing. It's already there.

Most apps have years of behavioral tracking sitting in their analytics platform. The problem isn't access. It's that nobody gave designers a fast way to surface it without writing queries or waiting on data teams.

Mixpanel's agent changes that. Claude connecting it to your actual codebase tags makes it precise. You're not asking vague questions and hoping the agent figures out what you mean. You're feeding it the exact event names that live in the code and asking it what those events show.

That's a very different conversation. And it gives you a very different answer.

Here's how to try it yourself:

Step 1: Open Claude and connect it to your repo. Ask it to list the analytics events on the specific page you're working on.

Step 2: Copy those event names. Open Mixpanel's agent. Ask it in plain English what those events show you about user behavior.

Step 3: Let what you find change your starting assumption before you open Figma.

That last step is the one most designers skip. They get the data, nod at it, and then design what they were already going to design anyway. Don't do that. Let the data actually move you.

If you want to get sharper at turning data like this into design decisions you can defend and tie to outcomes your stakeholders care about, that's exactly what Builder Kickoff is built for.

It's July 4th. Most people are off. I'm not closing this.

But spots are limited and I'm cutting access off today. If you've been sitting on this, now is the time.

​Grab your spot before they're gone.​

Talk soon,

PS. One more thing: starting with this email, you'll see me coming from tyler@mrtylerwhite.com. Same me, new home. Add it to your contacts so these keep landing.

PPS. The fastest way to earn trust as a designer isn't to have the best instincts. It's to be the person in the room who already looked at the data before anyone asked.
​