Retention is the truest signal of product and business health. Learn how cohort analysis works and how to run it in plain English on your own data.
Most dashboards show you totals: total signups, total revenue, total active users. Totals can rise even while your product is quietly failing — new users pour in the top while existing ones leak out the bottom, and the headline number hides it.
Cohort analysis fixes that. You group users by when they joined — their "cohort" — and then track how each group behaves over time. It answers the question totals cannot: of the people who joined in January, how many are still active three months later? That number is retention, and it is the truest signal of whether you are building something people keep using.
Cohort analysis is simpler than it sounds. You need two things:
- A join/start date per user — when they signed up, made their first purchase, or first became active.
- Activity events with dates — logins, orders, visits, or any action that signals they are still around.
That is it. From those, you can assign each user to a monthly (or weekly) cohort based on their start date, and then check, for each later period, how many of that cohort were still active.
A retention curve plots the percentage of a cohort still active over time. Three shapes tell most of the story:
1. Falls fast, then flattens — the healthiest pattern. You lose some early, but the survivors stick around. The height of the flat part is your "retention floor".
2. Falls and keeps falling — a leaky bucket. Growth cannot outrun churn forever; this needs attention before you spend more on acquisition.
3. Falls then rises ("smile") — rare and excellent: dormant users coming back, usually because the product became more valuable over time.
Comparing recent cohorts to older ones also shows whether changes you made improved retention.
Traditionally, cohort analysis meant a data analyst writing gnarly self-joins and date math. With a natural-language tool like VedaX, you can ask for it directly. For example:
"For customers grouped by the month they first ordered, show what percentage placed another order in each of the following three months."
VedaX turns that into the right query against your connected data and returns the cohort table or chart. You can then pin it to a dashboard and watch retention move as you ship improvements — no SQL, no analyst queue.
Retention is only useful if it changes what you do. A few moves it commonly unlocks:
- Fix onboarding when the curve drops sharply in the first period — early churn is usually an activation problem.
- Find your "aha" moment by comparing what retained users did in week one versus churned users.
- Justify (or pause) acquisition spend — pouring money into the top of a leaky bucket just makes the leak more expensive.
Watch retention by cohort every month and you will catch product problems long before they show up in the revenue total.
They are two sides of the same coin. Retention is the percentage of a group still active after some time; churn is the percentage that left. If 70% of a cohort is retained at month three, churn for that period is 30%.
Enough users per cohort that the percentages are not noisy — often a few dozen at minimum. Even a modest history of signups and activity events is enough to start seeing meaningful retention patterns and compare cohorts over time.
Yes, if it has a start date per user and dated activity records. Connect the sheet to a tool like VedaX and ask for cohort retention in plain English; it generates the query and the cohort chart without manual SQL or pivot gymnastics.
See it in VedaX: AI business intelligence for small businesses