If you judge your store on blended monthly revenue and a single repeat rate, you are averaging away the answer. Cohort analysis groups customers by when (or how) they first bought, then follows each group over time. That is the only view that shows whether each new wave of customers is better or worse than the last.
This guide shows you how to build the table, read it, tie it to customer acquisition cost (CAC), and avoid the mistakes that make cohort charts lie. Where a figure is ours, it is labelled as an example. Where it is someone else’s, it has a source and a date.
Key Takeaways
- A cohort is a group of customers sharing a starting point, usually the calendar month of their first purchase. Storepush’s August 2026 guide describes the March cohort as everyone who bought for the first time in March, tracked as m1, m2, m3 and so on.
- Read the table by customer age, not calendar date. Month 0 is always 100%, so the signal sits in months 1 to 6.
- Compare cumulative revenue per customer against CAC by cohort. Xero’s AU guide (11 May 2026) gives a healthy CLV as at least three times CAC.
- Treat small and recent cohorts as hints. A cohort of 80 customers swings wildly, and a cohort eight weeks old has not had time to show month 3.
- Cohorts show what changed, not why. Pair them with a test plan, or you will explain every dip with a story.
Table of Contents
- What a cohort is, and why averages hide the problem
- The cohort types worth building
- How to build a retention cohort table
- Customer retention vs revenue retention
- Tying cohorts to CAC and LTV
- Reading the table: five patterns and what they mean
- Mistakes that make cohort charts lie
- From findings to actions
- Want a second pair of eyes on your cohorts?
- FAQ
- Sources
What a cohort is, and why averages hide the problem
A customer cohort is the group you get after choosing a shared parameter, such as the month of first purchase or a signup to SMS, and then watching that group’s behaviour across later periods. Klaviyo’s glossary frames it the same way: group by shared traits, track over time, adjust acquisition and retention.
Why not just use segments? Segments are snapshots of who your customers are right now. Cohorts start from a fixed point and follow the group. According to Storepush (13 August 2026), only cohorts show whether a retention problem is improving or worsening with each new wave you acquire.
Here is the mechanism. A blended repeat rate mixes customers acquired last week with customers acquired two years ago. It also mixes the people who came from a 30% off Meta ad with the people who came from a product review on YouTube. Those groups behave differently. Stripe’s Australian guide (30 June 2025) makes the same point: top-line numbers such as total users or average revenue miss which groups are doing well and which are dropping off.
Sit with that. Your average can stay flat while every new cohort gets worse, because a loyal old base props it up. The average is the last place you will see the rot.
The cohort types worth building
Most stores start with acquisition month. Stop there and you leave the useful cuts on the table. Peel Insights lists acquisition period (usually by month), number of orders, discount codes used and customer tags. Kissmetrics adds first product, acquisition channel, and whether the first purchase was full price or promotional.
Rank them by how directly they change a decision:
- Acquisition channel. Tells you where to put the next dollar. Meta prospecting, Google Shopping and email signups rarely retain alike.
- First product or category. Storepush reports that retention by first-purchase category often shows a 3 to 5x spread inside a single store. That spread should steer acquisition spend toward the hero products that bring back buyers.
- Full price vs promotional first order. Shows whether your discount is buying customers or renting them.
- Acquisition month. The baseline. Use it to spot trend and seasonality.
One rule: change one cut at a time. Mixing channel and category in the same table turns 12 rows into 60 rows of noise.
How to build a retention cohort table
The layout is a staircase. Each row is a cohort. Each column is months since the first order. Each cell is the share of that cohort who ordered in that month, as Soran’s guide describes it. Lining customers up by age means you compare month 1 with month 1 and month 6 with month 6, rather than comparing a January that includes Christmas with a June that does not.
The maths is simple. Shopify’s cohort retention guide (2026) gives retention as active customers in the period divided by the original cohort size, times 100.
Step by step
- Pull the data. You need customer ID, first order date, and every order date and value. Export it from your store platform or warehouse.
- Assign cohorts. Bucket each customer by the month of their first order.
- Count activity by age. For each cohort, count how many customers ordered in month 1, month 2, month 3 and so on after the first order.
- Convert to percentages. Divide by the cohort size. Month 0 will read 100%, because everyone in the cohort bought once.
- Add a second table for money. Cumulative revenue per customer, by cohort and month. This is the table your CAC conversation needs.
A worked example, with made-up numbers so you can check the arithmetic:
| Cohort | Customers | Month 0 | Month 1 | Month 2 | Month 3 | Month 6 |
|---|---|---|---|---|---|---|
| March | 600 | 100% | 9% | 7% | 5% | 4% |
| April | 540 | 100% | 11% | 8% | 6% | 5% |
| May | 720 | 100% | 6% | 4% | 3% | not yet |
May has the biggest intake and the weakest month 1. That is a prompt to ask what you changed in May: new channel, new offer, new creative. It is not yet a verdict.
If you run Shopify, the free report is a fast start. A local Shopify guide from Ecommerce Circle walks through the path (Analytics, then Reports, then the customer cohort report). Run it for the last 12 months and start with retention rate before moving to amount spent per customer.
Customer retention vs revenue retention
These are two different tables, and people blur them. Per Soran, customer retention counts people who place at least one order in a period. Revenue retention compares the cohort’s spend in that period with its first-month spend.
Why both? A cohort can keep 8% of its buyers and still grow revenue retention if those buyers order bigger baskets. The reverse also happens. If you lean on discounts to win repeat orders, customer retention looks fine and revenue retention sags. If you are working on basket size, our guide to increasing average order value covers six levers to test.
Tying cohorts to CAC and LTV
Retention is a means. The end is whether a cohort pays back what you spent to get it.
The building blocks, with sources:
- Cohort CAC. Reach Capital explains the calculation: spend to acquire the cohort divided by cohort size. Enalitica calls the blended version nCPA: a month’s total marketing spend divided by that month’s first-time customers.
- CLV. Xero’s Australian ecommerce guide (11 May 2026) calculates it as AOV times average number of purchases times average customer lifespan in years.
- Payback month. Per Enalitica, the month when cumulative profit per customer first exceeds nCPA.
- The benchmark. Xero says a healthy business typically has CLV at least three times CAC. Stewart and Smith Advisory (17 December 2025) also targets 3:1 or higher, and calls cohort analysis the tool for checking whether your churn and CLV assumptions are actually showing up.
Example, illustrative only. Your March cohort cost $30,000 in ad spend and brought in 600 customers. Cohort CAC is $50. If cumulative contribution margin per customer reaches $50 in month 4, month 4 is your payback month. If it plateaus at $110 by month 12, the cohort sits at 2.2:1. That is below the 3:1 benchmark, and now you know where to look.
Note the word contribution margin. Revenue per customer against CAC flatters you, because it ignores product cost, shipping and fees. Use margin whenever you can.
For the reporting side, a clear ecommerce reporting workflow helps you pick which KPIs sit beside the cohort table each week, so the cohort view does not live in a spreadsheet nobody opens.
Reading the table: five patterns and what they mean
A steep month 1 drop, then flat. Most of your buyers are one-and-done, and the few who return stay. Fix the first-30-days experience: delivery, onboarding emails, a reason to reorder.
A sharp drop at a specific month. Shopify (2026) suggests that if a cohort falls hard at month three, a win-back campaign at about the two-and-a-half-month mark gets ahead of it. The mechanism is timing: you reach the customer while the product is running low or the memory is fresh, not after they have drifted.
Newer cohorts worse than older ones. Stewart and Smith frame it this way: if the 2024 cohort churns faster than the 2023 cohort, you have a problem with CAC quality, product-market fit or pricing. For us, it is usually the first one. Scaling spend pushes ads into colder audiences, and colder audiences retain worse.
One channel with a fat month 0 and a thin tail. That channel is cheap to buy and expensive to keep. Judge it on payback month, not on first-order ROAS.
A promo-driven cohort that spikes then craters. Discount buyers came for the discount. Compare their month 3 against full-price buyers before you run the offer again. (Offer design ≠ discount.)
Mistakes that make cohort charts lie
The usual ones, and the fix for each:
- Reading recent cohorts as finished. A cohort acquired eight weeks ago has had no chance to show month 3. Grey out cells that have not matured, or you will read incomplete data as churn. Kongo Tech calls this censored data.
- Trusting small cohorts. A cohort of 80 customers can swing hard month to month. Under roughly 100 customers, call it directional. We use that as a rule of thumb, not a law of statistics.
- Blaming a cause you have not tested. Cohort analysis shows what changed, not why. A weak March cohort could be a bad promotion, a stockout or a seasonal fluke. The table cannot tell you which. Run the test.
- Crediting a channel from the table alone. Channel cohorts show who came through the door, not which ad caused the order. If you need causation, you need an incrementality test.
- Ignoring seasonality. A November cohort full of gift buyers will not reorder like a February cohort. Compare November to last November.
Why do we put this so bluntly? Because a wrong read costs real money. Teams cut a channel that was fine, or double a budget on a cohort that was still maturing.
From findings to actions
Use cohort findings as a loop. Klaviyo notes that once you change marketing to match cohort behaviour, you change the behaviour, which changes the next cohort analysis. Rerun it every month.
Where the actions usually land:
- Acquisition. Shift budget toward the channels and first products with the best payback month. Our Google Ads and Meta and TikTok ads teams do this by cohort, not by platform-reported ROAS.
- Early lifecycle. The first 90 days carry the weight. A consultancy case study from Customer Science reports that small first-90-day improvements produced outsized CLV gains, linked to onboarding emails, replenishment reminders and timely cross-sells. Good onboarding is a process with steps and metrics, and a customer onboarding process is worth mapping before you automate it. Then build the flows through email and SMS marketing.
- On-site conversion. If a cohort is thin because the first-order experience is poor, the fix sits upstream of retention. See what to fix first in ecommerce conversion rate optimisation.
- Budget. Fund the cohorts that clear 3:1 and put a test budget behind the ones that do not. Avoid blanket cuts.
Our view: cohorts are where creative, media and CRO finally meet
Most agencies hand you a ROAS number from the ad platform and call it a day. That number is a first-order number. It cannot tell you whether the customer came back in month 3, and month 3 is where the money is made or lost.
We run strategy, creative production, paid media and conversion as one senior team. That matters here for a practical reason. When a cohort table shows that customers from one creative angle retain better, the same people who read the table brief the next round of creative and sign off the edit. There is no hand-off where the insight dies.
Our rule is simple: judge acquisition on payback month and cohort LTV:CAC, not on the cheapest first order. We aim to deliver at least 3 times your investment in us, and cohort economics are how we keep ourselves honest about that.
If your table shows newer cohorts decaying faster than older ones, do not buy more of the same. Change the creative promise, the offer or the audience, then watch the next cohort.
Harry Sekhon
Want a second pair of eyes on your cohorts?
If you have a cohort table and no clear next move, we can read it with you and turn it into a test plan across creative, paid media and conversion. Start with our marketing services for DTC and consumer brands.
If the table points to a conversion problem, our conversion optimisation team works on the first-order experience. If it points to a creative problem, ad creative production is where we start.
FAQ
What is cohort analysis in ecommerce?
It groups customers who share a starting point, most often the month of their first purchase, and tracks their behaviour over later months. You see repeat purchase rate, spend per customer and drop-off points for each group. That shows whether newer customers are better or worse than older ones.
How many customers does a cohort need?
There is no official threshold. As a rule of thumb, treat any cohort under about 100 customers as directional, because small groups produce jagged curves that reflect sample size, not behaviour. If monthly cohorts are too small, group by quarter.
What is a good LTV to CAC ratio?
Xero’s Australian guide (11 May 2026) and Stewart and Smith Advisory (17 December 2025) both point to 3:1 or higher. Calculate it per cohort, using contribution margin where you can, and check the payback month alongside it.
How often should I run a cohort analysis?
Monthly is enough for most stores, since the table updates by cohort month. Check it sooner after a big change in channel mix, offer or creative. Remember that fresh cohorts will not have matured, so judge them against older cohorts at the same age.
Does cohort analysis tell me why customers leave?
No. It shows where and when the drop happens. The reason comes from testing, customer research or surveys. Use the table to pick the question, then run the test.
Sources
- What is a Cohort Analysis? (Klaviyo)
- Cohort analysis for businesses (Stripe AU, 30 June 2025)
- Ecommerce Cohort Analysis: Read, Run, and Act on It (Storepush, 13 August 2026)
- Cohort Analysis, Ecommerce Analytics Explained (Peel Insights)
- E-commerce Cohort Analysis: Behavior Over Time (Kissmetrics)
- Cohort analysis for ecommerce: a step-by-step guide (Soran)
- Cohort Retention Analysis: How to Reduce Churn (Shopify, 2026)
- Ecommerce metrics to track for your online store (Xero AU, 11 May 2026)
- The CFO’s Lens: Driving Value Through Unit Economics (Stewart and Smith Advisory, 17 December 2025)
- Customer Lifetime Value, Without the Myths (Enalitica)







