Cohort analysis and retention consultant

Every struggling growth story I've been called into has the same scene: a dashboard full of up-and-to-the-right charts, and a company that somehow isn't compounding. The charts are blended — new user growth papering over the fact that each month's cohort quietly dies. Cohort analysis is just the refusal to let averages lie to you, and it's the highest-leverage analytical habit a product team can build.

How I read a retention problem

• Curve shape first. A cohort curve that flattens — at any level — means a core of users found durable value; the job is widening that core. A curve that decays to zero means no floor exists yet, and no acquisition spend should be scaled onto it. These are different diseases with different treatments, and blended metrics can't tell them apart.
• Segment the flatteners. Which acquisition channel, use case, or onboarding path do the retained users share? At CaaStle, cohort-level reads on subscription behavior — not blended churn — were what let our experimentation program find $2.1M in ARR savings; the money was hiding in differences between cohorts that the average erased.
• Find the cliff, then interview it. Most products lose users at one or two specific moments — after first value, at the first renewal, when the novelty loop ends. The cohort table locates the cliff; churned-user interviews explain it; the roadmap fixes it.
• Instrument for the next question. Retention analysis dies when every question needs a data engineer. Part of every engagement is leaving behind self-serve cohort views your PMs can slice without help.

Founder-tested, not just enterprise-tested

I apply the same discipline to my own companies. At WisOwl AI, weekly recruiter-cohort retention — not our 5,000+ signups — is the number I actually steer by, because signups are what happened to us and retention is what we earned. At Medzin, repeat-usage cohorts were what proved a healthcare discovery product had a business underneath it, long before the revenue chart looked like anything.

Typical engagement: a two-to-three-week retention diagnosis — cohort rebuild, cliff identification, churn interviews, ranked fix list — with optional fractional support to run the retention roadmap that follows.

Frequently asked questions

What retention rate is "good" for our product?

Benchmarks mislead more than they help — a daily social product and an annual tax tool live on different curves. The honest standard: your curve should flatten, and each quarter's cohorts should flatten higher than the last. Direction beats benchmark.

We're pre-revenue. Is cohort analysis premature?

It's most valuable exactly then — usage cohorts are the earliest honest evidence of product-market fit you can get, long before revenue data exists. A flattening week-four usage curve is worth more than any survey.

What tools do you need us to have?

Whatever you've got — a product analytics tool is convenient, but I've rebuilt cohort truth from raw event tables and SQL plenty of times. The blocker is almost never tooling; it's event definitions nobody has audited.

How is this different from hiring a data analyst?

An analyst produces the tables; I produce the decisions — which cliff to attack, what to build, what to stop spending on — and leave your team able to run the analysis themselves. It's product judgment applied to the data, not reporting.

Related pages