Your monthly revenue can be flat for a year while your business quietly falls apart underneath it. New customers arrive at roughly the same rate they leave, the top-line number doesn’t move, and nothing looks wrong until acquisition slows for one month and the floor gives way.
Cohort analysis is the technique that exposes this. Instead of asking “how many customers did we have in June?”, it asks “of the customers who first ordered in January, how many were still ordering in June?” It then asks the same question of the February group, the March group, and so on. The answer is a table, and that table has a shape that tells you precisely where your retention breaks.
This is a mechanics post. By the end you’ll have built a cohort table from your own order export and know how to read it.
The Bottom Line
- A cohort is a group of customers defined by when they first bought. Cohort analysis tracks each group forward in time separately.
- The technique finds the specific period repeat buying drops off, which an aggregate retention rate structurally cannot show.
- Increasing retention by 5% has been associated with profit increases of 25% to 95%, and acquiring a customer costs roughly 5 to 25 times more than keeping one.
- You need two columns to start: customer identifier and order date. Everything else is optional.
Save 80% of delivery management time
We handle everything:
- Dedicated operations manager
- Real-time tracking dashboard
- Automated customer notifications
- Urgent issue resolution
What a cohort actually is
A cohort is a set of customers who share a starting point in time. For most local businesses the useful definition is the month of first order: everyone whose first-ever purchase landed in March is the March cohort, permanently, regardless of what they do afterwards.
That permanence is what makes the method work. A customer never changes cohort, so when you track the March group across April, May and June, you’re watching the same people. Any decline is real behaviour rather than a change in who’s being counted.
The alternative, an overall retention rate, mixes everyone together and hides the thing you need. As one widely cited illustration puts it, a stable retention rate can conceal new customers leaving quickly while long-standing customers keep the average afloat (Stripe). The average is fine. The business isn’t.
Cohort work sits downstream of basic measurement. If you aren’t yet logging order data consistently, the five daily numbers in the data analytics starter system come first. The reorder gap in particular is the raw material for everything below.
Why retention is worth this much effort
The economics are lopsided enough to justify an afternoon with a spreadsheet. Acquiring a new customer costs roughly 5 to 25 times more than retaining an existing one, and the probability of selling to a customer you already have runs around 60% to 70% against 5% to 20% for a new prospect (GrowSurf).
The profit link is the figure most often quoted: a 5% improvement in retention has been associated with profit increases between 25% and 95%, a finding from Bain and Company’s work on loyalty economics that has been repeated across sectors since. Existing customers also spend more per order, around 67% more on average than new ones.
If the underlying concept is new to you, it’s worth reading a plain explanation of what customer retention means in retail before building the table. Cohort analysis is a measurement technique, and it only pays off once you know which retention behaviour you’re trying to change.
For a business running deliveries, there’s a compounding effect the general statistics miss. A repeat customer on an established route costs less to serve than a new customer at a new address, because drop density is what makes a route efficient. Retention improves your margin twice: once on acquisition cost, once on cost per drop. That second effect is why the business spending habits that hold a company back so often include discounting that buys one-off orders.
How to build a cohort table from your order history
You need an export with two columns: a customer identifier and an order date. Most point-of-sale and e-commerce systems produce this in one click. Then:
1. Find each customer’s first order date. Group your orders by customer and take the minimum date. This assigns every customer to a cohort. Label it by year and month, like 2026-01.
2. Calculate a period number for every order. For each order, work out how many months have passed since that customer’s first order. A customer’s very first order is period 0, an order the following month is period 1, and so on.
3. Count unique customers per cohort per period. A pivot table does this: cohort month as rows, period number as columns, count of distinct customers as the value.
4. Convert to percentages. Divide each cell by the cohort’s period 0 count. Period 0 is 100% by definition.
That’s the table. In a spreadsheet the whole thing is one date-minimum formula, one month-difference formula, and a pivot.
How to read a cohort table
Here’s a worked example for a business with a monthly reorder rhythm. Each row is a cohort; each column is the share of that cohort still ordering that many months later.
| Cohort | Month 0 | Month 1 | Month 2 | Month 3 | Month 4 | Month 5 |
|---|---|---|---|---|---|---|
| January | 100% | 62% | 48% | 44% | 42% | 41% |
| February | 100% | 60% | 47% | 43% | 42% | — |
| March | 100% | 61% | 46% | 43% | — | — |
| April | 100% | 45% | 31% | — | — | — |
| May | 100% | 43% | — | — | — | — |
Three things to read, in this order.
Read down a column first. Column “Month 1” runs 62%, 60%, 61%, then 45%, 43%. Something changed for customers acquired in April. Reading down a column compares cohorts at the same age, which isolates changes in what you did: a price rise, a new delivery fee, a service problem, a promotion that attracted the wrong buyers.
Then read across a row. January goes 100 → 62 → 48 → 44 → 42 → 41. The steep fall is between month 0 and month 2, after which the curve flattens. That flattening is your loyal core, and the point where it flattens tells you how long your retention window actually is. Every intervention you run should land before it.
Then look for where the curve stops falling. If a row never flattens, you don’t have a retention problem so much as a product-market problem. If it flattens high, your existing customers are solid and your issue is acquisition volume.
In the example, the finding is narrow and specific: something introduced around April is costing roughly 16 percentage points of month-one retention. That’s a question with a findable answer: check what changed in April.
Choosing the right cohort size
Monthly cohorts suit most local businesses. The rule is to match the cohort period to your natural purchase rhythm.
- Weekly cohorts suit high-frequency operations such as prepared food, daily or twice-weekly delivery rounds. Weekly grouping is the standard choice for fast-moving delivery services, where a month is long enough to hide the entire retention story.
- Monthly cohorts suit most florists, bakeries, caterers and wholesalers, where customers order every few weeks.
- Quarterly cohorts suit low-frequency, high-value relationships such as event catering, seasonal wholesale accounts.
Pick the period your typical customer orders within, not the period your accountant reports in. If your median reorder gap is nine days, monthly cohorts will make almost everyone look retained.
What to do once you’ve found the drop-off
A cohort table gives you a month and a magnitude. Turning that into a change takes one more step: work out what the customers in the failing cohort experienced differently.
Start with the obvious operational candidates: a delivery fee introduced, a window narrowed, a substitution policy changed, a supplier switch that affected quality. Then check acquisition source, because cohorts that lapse fast are frequently cohorts acquired through discounting. Meal-kit delivery businesses have used exactly this analysis to identify which subscriber segments were worth acquiring and redirect marketing spend accordingly, cutting acquisition cost by concentrating on the cohorts that stayed.
If you want to go further than a spreadsheet at this point, the category of software built for this is worth a look. FullStory’s rundown of behaviour analytics tools covers what the paid options add, which is mostly session-level detail rather than better cohort maths.
The behavioural detail underneath the drop-off is where the fix usually lives: which delivery windows lapsing customers chose, how their order timing shifted before they stopped, where they abandoned the ordering page. Behavioral analytics for local delivery covers those signals and what each one indicates.
It’s also worth segmenting cohorts by customer type before concluding anything. Retention curves differ sharply by generation. Gen Z shopping habits include a strong deal-seeking reflex that produces fast-lapsing cohorts, while millennial spending habits skew toward loyalty where values and experience align. A cohort that looks broken may simply be a cohort of bargain hunters.
Frequently asked questions
What is cohort analysis in simple terms?
It’s grouping customers by when they first bought from you and then tracking each group separately over time. Instead of one blended retention number, you get one retention curve per group, which shows when people drop off and whether that’s getting better or worse for newer customers.
What data do I need for a cohort analysis?
A customer identifier and an order date for every order. That’s the minimum and it’s enough for a full retention table. Order value lets you extend the same table into revenue retention, which is useful but not necessary for the first pass.
How many months of data do I need?
Six months gives you a readable table; twelve is better because it lets you distinguish seasonality from decline. You can build the table with three months, but with only two or three periods per cohort you can’t see where the curve flattens, which is the most useful part.
What’s a good retention rate?
It varies enormously by sector, so benchmarks are of limited use. Reported e-commerce averages cluster somewhere around 30%, while established service relationships run far higher. Your previous cohorts are the comparison that matters. A cohort table is valuable because it compares you to yourself.
Is cohort analysis the same as customer segmentation?
No. Segmentation groups customers by attributes such as location or order size, and a customer can move between segments. Cohorts are defined by a fixed point in time and never change, which is what makes them reliable for measuring behaviour over time.
