Cohort analysis: the real way to understand which customers come back (and which don't)

If you've ever looked at your business's "customer retention rate" as a single number (say, "we retain 40% of our customers") you've probably ended up with a figure that sounds informative but actually tells you almost nothing useful about what to do next. Cohort analysis is the tool that turns that flat number into something you can actually act on, and surprisingly few small businesses use it, not because it's complicated, but because nobody's ever explained it to them in plain terms.

What a cohort actually is, no jargon

A cohort, in this context, is simply a group of customers who share the moment they joined your business: everyone who signed up in January, everyone who made their first purchase during the summer sale campaign, everyone who subscribed in the same quarter. Instead of looking at "all my customers" as one single mass, cohort analysis splits them into these groups by their entry date, and then tracks each group separately over time to see what percentage of that specific group is still active (buying, using the service, renewing) one, three, six or twelve months after joining.

Why this tells you more than an overall retention rate

The problem with an overall retention rate is that it mixes very old customers with very recent ones in the same calculation, and that mix can hide completely opposite trends. Picture a business that's been running for three years: customers who joined in year one, when the product or service was still being ironed out, might show low retention. Customers who joined in the last six months, after major improvements to the product or customer service, might be retaining much better. If you only look at the average of everyone combined, that flat number won't tell you whether your business is getting better or worse at holding onto customers, because the two effects partly cancel each other out in the average.

Cohort analysis separates those two stories. If you compare the retention curve of the January cohort against the June cohort of the same year, and June retains better at the three-month mark than January did, you've got a clear, concrete signal that something you changed between January and June (a product change, a better onboarding process, a pricing tweak) is working. Without cohorts, that signal stays completely buried inside a general average.

A practical example for a small business

Picture an online store selling a monthly subscription product (say, a gourmet food box) with around 200 new customers a month. A typical cohort analysis organises the data into a table: each row is the month a group of customers joined, and each column shows what percentage of that group was still subscribed one, two, three months later. You might discover, for example, that customers who arrived through an aggressive discount campaign (50% off the first month) retain much worse from the second month onward than customers who arrived through a friend's recommendation, even though the discount campaign's initial volume looked like a success at the moment of acquisition. Without looking at this by cohort and acquisition channel at the same time, you'd keep pouring money into the campaign that brings in cheap customers who leave fast, thinking you're growing, when you're actually filling a bucket with a hole in it.

Another common pattern this analysis reveals is the "critical month": many subscription businesses find that once a customer makes it past their third or fourth active month, their odds of sticking around much longer jump noticeably. That turns that specific window into the moment worth concentrating retention effort on (a follow-up call, a special offer, staggered welcome content), instead of spreading the effort evenly across the entire customer lifetime.

How to build a simple cohort analysis without complicated tools

You don't need advanced analytics software to get started. With a spreadsheet and the basic data any business already has (each customer's signup or first-purchase date, and dates of subsequent purchases or interactions), you can build a cohort table by hand: rows for the entry month, columns for months elapsed since entry, and in each cell the percentage of that group still active at that point. Tools like Google Analytics 4 and most ecommerce platforms (Shopify, among others) already include pre-built cohort reports that run this calculation automatically, saving the manual work once you decide to track it regularly.

What to do with what you find

The value of cohort analysis isn't in looking at it once, it's in comparing it month over month to see whether the changes you make to your business (product, pricing, customer service, acquisition channel) improve or worsen retention for the more recent cohorts compared to earlier ones. It's essentially a way of measuring the real effect of your decisions on customer loyalty, with an added benefit: since each cohort can also be compared by the channel, campaign or offer it arrived through, it lets you identify which customer sources are profitable long-term and which only look that way in the short term.

A common mistake: comparing cohorts that aren't actually comparable

A typical mistake when starting out with cohort analysis is comparing groups that didn't start from the same baseline and drawing the wrong conclusion from that comparison. If the December cohort includes a flood of Black Friday shoppers who never intended to become repeat customers, and you compare it directly against a normal month's cohort, you'll conclude "retention has dropped" when what actually changed is the group's composition, not your ability to build loyalty. Whenever possible, it's worth also splitting cohorts by acquisition channel or by the type of offer they arrived through, not just by entry month, so you don't mix apples and oranges within the same analysis.

Frequently asked questions