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

The problem hit me on a Friday afternoon

We'd just shipped Sonic Linker's core AI linking feature. Users were signing up. Some were creating projects. But when our lead investor asked "what's your Day 7 retention?", I had absolutely no idea.

I didn't have a data engineer. I didn't have a budget for Amplitude or Heap. And I definitely didn't have time to wait three months for someone to build a proper data warehouse.

What I did have was a weekend, Mixpanel's free tier (up to 100k events per month), and a growing suspicion that most early-stage products don't actually need fancy infrastructure. They just need answers to five critical questions.

Start with the five questions that actually matter

Before touching any tools, I wrote down what I needed to know:

1. Are people activating? (Did they complete their first AI link?)
2. Are they coming back? (Day 1, Day 7 retention)
3. Where are they dropping off? (Which step in the flow kills them?)
4. What features correlate with retention? (Do people who use X stay longer?)
5. Is our funnel broken? (Signup to first value)

That's it. Not "let's track everything". Not "we need a complete behavioral model". Just five questions I could answer with simple event tracking.

I've seen teams at Finvestfx spend months building dashboards they never look at. The goal isn't perfect data. It's actionable data you'll actually use to make decisions.

My actual setup (you can copy this tomorrow)

I installed Mixpanel's JavaScript SDK and set up exactly six events: - User signed up - User created project - User generated first AI link (our activation moment) - User invited team member - User ran advanced query (power user signal) - User hit paywall

Each event included basic properties: user ID, plan type, project count, and timestamp. Nothing fancy. I hardcoded the API calls right into our frontend because we needed speed, not elegance.

Mixpanel's free tier gave me funnels, retention curves, and cohort analysis. That covered questions 1, 2, 3, and 4.

For question 5 (is the funnel broken?), I needed more context than events alone could give me. So I built a daily automated export.

I used Mixpanel's API to pull a daily summary (new signups, activations, feature usage) into a Google Sheet via Apps Script. Then I added manual columns for qualitative data: - Which acquisition channel (I literally asked users in onboarding) - First support ticket topic - Whether they responded to my welcome email

This combo of quantitative and qualitative data is what actually helped me diagnose problems. Pure event tracking told me people were dropping off after signup. The Sheet told me why (they didn't understand what "AI linking" meant).

Here's the thing nobody tells you: analytics tools don't help unless you have a routine for actually looking at them.

Every Monday morning, I spent 30 minutes answering my five questions. I didn't build elaborate dashboards. I didn't automate Slack reports. I just opened Mixpanel, checked the numbers, updated my Sheet, and wrote down one insight and one action item.

Sometimes the insight was huge ("people who invite a team member in Week 1 have 4x better retention"). Sometimes it was small ("our signup button is broken on mobile Safari"). But I always had something to act on.

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

If I were doing this again tomorrow, I'd add one thing: session recordings. Tools like Hotjar or Microsoft Clarity are free and show you *why* people are confused, not just that they are.

But I wouldn't change the core approach. I've worked with teams that waited months for "proper analytics infrastructure" while making completely blind product decisions. At Finvestfx, we had enterprise clients asking for features I couldn't validate because our tracking was broken for weeks at a time.

Starting scrappy beats starting never.

The real unlock

The best part of this setup wasn't the tools. It was that I could answer investor questions in real time, validate feature ideas in days instead of weeks, and kill bad bets before wasting engineering time.

When we eventually hired a data person six months later, they didn't tear down what I'd built. They just made it faster and added more nuance. The foundation (tracking activation, retention, and drop-off) stayed exactly the same.

You don't need a data engineer to make data-driven decisions. You just need to be honest about what questions actually matter, and scrappy enough to answer them with whatever tools you've got.