Your overall retention number is one figure covering customers who joined years apart, so it cannot tell you whether you are getting better. Upload a CSV of your orders or customers and see how long each month’s arrivals stayed, whether the recent ones are holding up better, and the month things changed. It runs in your browser, so the file never leaves your machine.
Drop a CSV of orders or customers here, or
An event log (customer id, date, optional amount) or a customer table (id, signup date, last-active date). Up to 200,000 rows. Nothing is uploaded.
The file is read and the grid is built in this browser tab. It never reaches our server, so there is nothing for us to keep.
One retention figure for the whole business mixes customers who arrived under completely different products, prices and onboarding flows, and reports the average as though it described something.
One row per signup month, shaded so the groups that stuck around stand out. Months that have not happened yet are left blank rather than counted as people leaving — which is where most spreadsheet versions go wrong.
An order or activity log with one row per event, or a customer table with a signup date and a last-active date. It works out which you gave it, says so, and lets you correct it.
It names the signup month where retention shifts, and tells you plainly when the shift is too small to trust rather than letting you read a story into the colours.
Every signup month drawn as a line from its own day one. If the newer lines sit above the older ones, the customers you are winning now are better ones. That is easier to see than reading a grid row by row.
If your file has an amount column you get the same view in revenue. Losing a third of your customers but only 5% of your money is a very different business from the other way round.
Take the grid and a plain summary away as a CSV. Your file is read inside your browser tab and never reaches our server. Free, no sign-up.
An order or activity log needs a customer id and a date. A customer table needs a customer id, a signup date and a last-active or cancellation date. An amount column in either is optional and gets you the revenue view too.
The columns are picked out of your headers and the grid is built inside your browser. Nothing is uploaded and there is no account to make.
You get told which signup month things shifted in. That turns a chart into a question worth asking: what did you change then, and can you do more of it?
Four things to look at, and one trap. The trap is the empty corner.
Most people searching for cohort analysis in Excel are looking for a way to stop. The spreadsheet version is a helper column of months-since-signup, a COUNTIFS grid, and conditional formatting — perfectly good, and rebuilt from scratch every time the data refreshes.
Two things go wrong in that build often enough to be worth naming. Months that have not happened yet come back as zero, so the corner of the grid fills up with customers who never actually left, and the average row underneath inherits it. And the month you are currently in gets counted as though it were finished, so every group looks like it falls off a cliff at the end. Both are handled here, and you are told what was left out and why.
You still get a CSV at the end. It is the same grid, and it opens in Excel — the difference is that rebuilding it next month is a drag and a drop rather than a morning.
The file is read in your browser tab and never reaches our server.
Cohort analysis groups customers by when they joined and then follows each group forward in time separately. Instead of one retention number for the whole business, you get a curve for January signups, a curve for February signups, and so on. It matters because a blended number hides the thing you want to know: whether the customers you are winning now stay longer than the ones you won a year ago.
Start with a table of events that has a customer id, a date and ideally a value. Assign each customer to a cohort based on the month of their first event, then for every month afterwards count how many of that cohort were still active. Divide those counts by the cohort's starting size to get percentages, and lay them out as a grid with cohorts down the side and months since signup across the top. The diagonal shape falls out of the fact that recent cohorts have had fewer months to be observed.
It depends entirely on the model, so compare against your own category. B2B SaaS selling to larger companies should see 90%+ annual customer retention; SMB SaaS commonly sits at 70-80%. For consumer apps, month-one retention of 25-35% and month-twelve of 10-15% is respectable. The signal that matters in any of them is a retention curve that flattens rather than one that keeps sliding towards zero.
They are two views of the same number: retention plus churn equals 100% for the same period and the same group. If 92% of your customers are still here after a year, annual churn is 8%. Retention is the more useful frame for cohort work because it is what you plot as a curve, and because “still here” is easier to define consistently than “has left”.
Net revenue retention is the recurring revenue a cohort produces today as a percentage of what it produced at the start, including upgrades, downgrades and cancellations. Above 100% means the cohort is growing even though some of its customers have left. Good B2B SaaS is 110-120%; best-in-class is above 130%. Gross revenue retention is the same figure with expansion excluded, so it can never exceed 100% and shows you the leak on its own.
Cohorts group people by when something happened; segments group them by what they are. A cohort is “everyone who signed up in March”, a segment is “everyone on the enterprise plan”. Cohorts let you compare behaviour over time on a level footing, because every group is measured from its own day one. The two combine well: run the same cohort grid separately per plan, channel or region and the differences usually show up quickly.
Use monthly cohorts for subscriptions and anything with a long buying cycle, and weekly cohorts for high-frequency products or when you want to see the effect of a recent change. The test is whether a typical customer would be expected to come back within the period. If most people only act once a month, weekly cohorts just produce a sparse grid of small numbers that reads as noise.
Aim for at least a few hundred customers per cohort before you read much into the percentages. With 20 people in a cohort, one person leaving moves retention by five points, so the grid will look dramatic while telling you nothing. If the cohorts are thin, widen the period from weekly to monthly or quarterly rather than trying to interpret the wobble.
This page does one thing with one file, in your browser, and then stops. The obvious next question — why did March change, was it a channel, a price, a plan — needs a second file and a conversation, and a web form cannot have one.
For the follow-up questions, run the same analysis inside your AI assistant with meelu-analytics-mcp. It reads the files off your disk, joins your orders to your customers to your support data, and groups cohorts by any column rather than only by signup month. It is free and open source, and it runs locally for the same reason this page does. See the cohort retention use case.
A cohort grid tells you which arrivals held up. These answer the next three questions from the same customer file, and all of them run in your browser too.
All of them are pieces of Meelu, a desktop app where an AI marketing agent runs your marketing on your own machine. Join the waitlist.