Cohort analysis for product managers and marketers

When I went into this cohort analysis lesson, my hands were literally shaking. The commercial director had told me point‑blank: “Our current CAC is a disaster. If nothing changes, in a couple of months we’ll have to part ways.” I had already dug through all my channel reports, built pretty dashboards, but I still had no real understanding of where exactly we were burning money. During Alexey’s lesson, for the first time it became genuinely clear how to break the audience into cohorts so that the picture stops being chaos and turns into a story where you can see what happens to people after the first touchpoint.

What I really liked is that it wasn’t some boring academic breakdown. Alexey speaks like a normal person, without “analytics snobbery” and endless formulas. He shows everything with examples: here are cohorts by month of registration, here by month of payment, here’s how they behave across channels, and here is the real leak — not where people usually try to look for it. At some point I literally felt relieved: the problem wasn’t that I’m a “bad marketer”, but that I’d been looking at the data from a weird angle. In his explanation, cohort analysis stopped being a scary buzzword from books and turned into a working tool I could open the next morning and apply directly to my numbers.

After the lesson I just sat down and rebuilt my reports: split users into cohorts by first payment and source, checked where people start dropping off massively, and suddenly a lot of things made sense. We found channels that had a “nice” CAC, but after a couple of months almost none of those customers stayed — and that was exactly what I hadn’t seen before. I went back to the commercial director not with excuses, but with a clear picture and concrete suggestions on what to cut and what to scale. Honestly, the only downside is that sometimes you have to pause the video and rewind - there’s a lot of material — but this is one of those rare cases when you feel you’re not just learning for the sake of it, you’re actually saving your marketing (and your job in the company).

«Ran pricing models over and over, couldn’t get to the right number. Looked at segments, looked at cohorts – Boom!»

What for. I own the P&L and pricing, and my task was simple on paper: find a price level where both revenue and margin grow without a spike in churn. In practice, I kept running into an “average hospital temperature”: one LTV, one CAC, one set of plans that were supposed to fit everyone but never gave a clear picture.

How I use it. After the cohort analysis lesson, I rebuilt the whole approach: first I segment customers by product use cases, acquisition channel, and initial plan, and only then I calculate unit economics inside each cohort. Now a price test is not one big experiment on the whole base, but a series of targeted hypotheses for specific clusters. This drastically reduces the cost of mistakes and speeds up decision cycles.

Did it pay off. Yes, and fairly quickly. In some cohorts we raised prices by 10–15% with almost no impact on churn; in others we made the entry softer and won on customer lifetime. Most importantly, the gap disappeared between “the model looks great” and a P&L that behaves unpredictably: now it’s clear which segments actually benefit from price changes and which are better left alone.

Pros and cons. The upside is that the cohort approach gives a transparent link between pricing decisions and real customer behavior over time, and it’s easier to talk to finance and sales when you show dynamics for specific groups. Another plus is that the tool scales well: you set up the slices once and then just plug in new pricing hypotheses. The downside is the entry barrier: until you flip your thinking from averages to cohorts, it feels like unnecessary complexity — but after the first successful iteration, you really don’t want to go back.

Honestly, I signed up for this mostly to tick a box. I’m a Senior PM in big tech, surrounded by dashboards, my own analyst, retention squads - felt like I’d seen it all. But at some point it got embarrassing to admit: we’d been shipping “retention features” for six months, and I still couldn’t say clearly *who* was dropping off and *why*. After going through Alexey’s cohort breakdown, it became uncomfortably obvious that we’d been staring at an average retention curve and making decisions based on a “median user” who doesn’t actually exist.

Since then, I talk to the data in a completely different way: not “retention went down”, but “this cohort, with this entry scenario, starts falling apart in week three after this specific change”. It’s much easier for the team to prioritize because we see which hypotheses actually move cohorts and which ones just add noise to the overall chart. I won’t pretend everything is perfect now, but that feeling of banging my head against the wall is gone — at least we finally know *which* part of the wall to hit.

“Reviewed financial models in an accelerator. God, why do so few people know about cohorts?”

I ran a venture fund, and I’m honestly blown away by the financial models founders send over. In 90% of cases it’s a pretty spreadsheet with “we’ll grow X times in 3 years,” built on faith‑based assumptions. No cohorts, no attempt to understand how the audience will change over time, just generic MRR curves and the immortal “we’ll bring churn down to 2%.” In the end it’s not a model, it’s a slide deck of hopes.

If someone sends me a financial model without cohorts, I send it back to be redone. If a founder can’t show how users from different channels and product scenarios behave by cohort, how retention works, and what *actually* drives revenue, that’s a red flag for me already at pre‑seed/seed, let alone Series A. A cohort‑based approach is not a luxury for later stages, it’s a must‑have for anyone asking for serious capital: it forces you to think in a driver‑based way instead of drawing pretty lines and hoping “it’ll somehow work out.”