I Set Up Product Analytics in One Weekend Using Google Sheets and Mixpanel's Free Tier

The problem: you're shipping blind

At Sonic Linker, we were three months into building an AI SaaS platform. We had users. We had features. What we didn't have was any clue which features people actually used.

No data engineer. No analytics stack. Just a React app, a backend API, and a lot of guessing.

I've seen this exact situation at three different companies now. The data team is "coming soon." Meanwhile, you're making decisions based on support tickets and vibes. That's not a strategy, that's a coin flip.

So I spent one weekend setting up analytics that actually told me what mattered. No fancy infrastructure, no six-month data roadmap. Just the minimum viable setup to stop flying blind.

Start with events, not dashboards

The mistake most PMs make is thinking they need a perfect data model before they start tracking anything. Wrong.

I picked five events that mattered: - User signed up - User created their first project - User ran an AI workflow - User invited a team member - User came back the next day

That's it. Not fifteen events. Not every click. Five things that showed me if the product was working.

I used Mixpanel's free tier (generous limits, dead simple SDK). Took me two hours to add the tracking code to our frontend. Another hour to test it. By Saturday evening, I was watching real events roll in.

The key was defining these events with the engineering lead, not solo. I wrote down exactly what each event meant ("workflow run" = when the AI actually processes something, not when they click the button). He implemented it. No ambiguity, no guessing later.

Build your first dashboard in Google Sheets

Here's the part that sounds ridiculous but worked perfectly.

Mixpanel had the raw event data. But I needed to answer specific questions: What's our day 1 to day 7 retention? How many users are getting to their first workflow? Which signup sources convert better?

I exported a week's worth of event data to CSV. Dropped it into Google Sheets. Built pivot tables and basic formulas.

This sounds manual because it is. But here's why it worked: I could iterate on the questions faster than any BI tool would let me. I'd look at the data, realize I was asking the wrong question, tweak the formula, and try again. In a proper analytics platform, that would mean tickets, wait times, and someone else's interpretation of what I needed.

At Finvestfx, I did the same thing with client activity data. The Treasury team wanted a "usage dashboard." I gave them a Google Sheet that updated daily via Zapier. They loved it because they could filter and sort however they wanted. No training needed.

The one API call that changed everything

Two weeks in, I realized I needed one more thing: a daily snapshot of key metrics sent to Slack.

I wrote a simple script (okay, ChatGPT wrote 80% of it) that hit Mixpanel's API every morning, pulled the previous day's numbers, and posted them to our team channel: - Signups: X - First workflows: Y - Day 1 retention: Z%

Took maybe three hours to set up. Ran on a free Heroku dyno.

Suddenly, the whole team was looking at the same numbers every day. When retention dropped, we noticed immediately. When a new signup source started converting better, we doubled down. The data became part of our daily rhythm, not a monthly report nobody read.

What I'd do differently (and what I wouldn't)

The temptation is to upgrade too fast. "We need Amplitude, we need a data warehouse, we need proper ETL."

Maybe eventually. But not yet.

The scrappy setup forced me to be disciplined. I couldn't track everything, so I tracked what mattered. I couldn't build complex funnels, so I focused on the three steps that actually predicted success.

When we finally did hire a data person at Sonic Linker (five months later), they thanked me. The event taxonomy was clean. The questions we cared about were well-defined. We weren't starting from zero, we were scaling what worked.

The one thing I'd add earlier: user-level cohorts. Tracking aggregates is good. Tracking specific user journeys is better. Tools like June or PostHog have free tiers that make this easier now.

The real unlock isn't the tools

Here's what actually mattered: I stopped waiting for permission to understand our users.

No data engineer meant no excuse to delay. No budget meant I had to be creative. No perfect setup meant I shipped something imperfect and iterated.

The analytics I set up that weekend weren't impressive. But they were running on Monday. And by Wednesday, we'd already killed a feature nobody used and doubled down on one that was working.

That's the bar. Not perfection. Just enough signal to stop guessing.

If you're in this situation right now, pick your five events today. Set up Mixpanel or PostHog tomorrow. Build your first dashboard this weekend. You don't need a data team. You need to start looking.