Customer Cohort Analysis for Ecommerce: Find Out Which Customers Actually Stay
Your repeat purchase rate improved from 22% to 24%.
Good news?
Maybe.
But what if:
customers acquired this year are retaining worse,
while an older loyal customer base is temporarily making the store-wide number look healthy?
Averages can hide deterioration.
Cohort analysis fixes that by asking:
What happened to customers who started under the same conditions?
What is a cohort?
A group of customers sharing a meaningful starting point.
Most ecommerce cohort analysis begins with:
first purchase month.
January customers become one cohort.
February another.
March another.
Then you compare what percentage continues buying over time.
Shopify already supports this
Shopify’s current cohort report can group customers by first purchase and show:
- retention,
- sales,
- average order value,
- spend per customer,
- order counts,
- top marketing channels,
- subscription mix. (help.shopify.com)
That gives merchants much richer context than one repeat-customer percentage.
Read across the row
Suppose:
January cohort:
Month 1: 20% repeated
Month 3: 31%
Month 6: 36%
February:
Month 1: 18%
Month 3: 29%
March:
Month 1: 11%
March immediately looks concerning.
The question becomes:
What changed?
Read down the column
Compare month-one retention across cohorts:
January 20%
February 18%
March 11%
April 10%
Now you have a trend.
Something may have changed in:
- acquisition,
- product,
- experience,
- promotion.
Amplitude explains the same analytical principle well
Amplitude notes that a stable overall retention rate can hide dramatically different customer groups, while cohort analysis exposes drop-off cliffs and outlier groups. (amplitude.com)
Although its examples often come from software products, the concept translates perfectly into ecommerce.
Store averages can hide:
- great customers,
- weak customers,
- changing acquisition quality.
Acquisition cohorts tell you when
Behavioral cohorts tell you why.
Suppose March performed poorly.
Now split March by:
- first product,
- acquisition channel,
- discount,
- returns.
You may discover:
March was not the problem.
A new campaign was.
That is the power of combining time and behaviour.
Amplitude explicitly recommends this progression: acquisition cohorts identify when retention changed, while behavioral cohorts help explain why. (amplitude.com)
Use cohorts to evaluate promotions
Black Friday may generate huge new-customer volume.
But compare:
Black Friday 12-month contribution
with:
normal full-price customers.
If the promotion cohort retains badly, the first-order success may be less impressive than it looked.
Shopify specifically recommends using cohort analysis to evaluate whether customers acquired during promotions have lower future value than customers acquired under normal conditions. (shopify.com)
Use cohorts to evaluate first products
Create cohorts based on:
- first product purchased.
Now measure:
- order-two conversion,
- time to second purchase,
- customer value.
This identifies products that manufacture better customers.
That can change acquisition strategy.
Cohorts can reveal product-market changes
Imagine retention improves after a formulation update.
Or drops after shipping provider changes.
The timing can create a natural experiment.
Investigate what was different.
Avoid tiny cohorts
If you slice too aggressively:
March
California
Product A
Referral
Female
Full-price
you may end with 17 people.
Now the numbers are noise.
Amplitude specifically warns against cohorts so granular that sample size becomes unstable. (amplitude.com)
Use enough segmentation to answer the question.
No more.
A practical cohort dashboard
I would want:
Acquisition month
Customers
Order-two rate
Median days to order two
90-day revenue per customer
180-day revenue per customer
Average order value
Main acquisition channel
Then add slices for:
- first product,
- discount,
- subscription.
Cohort analysis is where acquisition meets retention
A paid advertising dashboard tells you:
this campaign acquired 5,000 customers at $45.
The cohort view tells you:
what those 5,000 customers became.
That is a more complete growth picture.
Cohort analysis also improves Revenue Recapture
Suppose one cohort has unusually strong historical value but has begun lapsing.
That may be a high-quality recovery opportunity.
Or one cohort has been weak since acquisition.
Maybe it does not deserve recovery investment.
History changes priority.
Do not build cohort charts just because they look intelligent
This is my favourite warning from Amplitude’s current cohort-analysis guidance: cohort analysis that ends in a dashboard rather than a hypothesis is simply more elaborate reporting. (amplitude.com)
Every cohort finding should end with:
What do we test?
Example:
Customers acquired through Product A retain better.
Hypothesis:
Move more acquisition toward Product A.
Test it.
A worked example
Three cohorts:
| Cohort | Customers | 90-day repeat | 180-day value/customer |
|---|---|---|---|
| January | 5,000 | 29% | $124 |
| February | 5,500 | 27% | $119 |
| March | 9,000 | 14% | $82 |
March acquired the most customers.
It may have created the least valuable customers.
Now investigate:
- offer,
- channel,
- first product.
The acquisition win may actually be a retention problem.
The real question cohort analysis answers
Not:
How are our customers doing?
But:
Which customers are behaving differently, when did that start, and what was different about how we acquired or served them?
That is an enormously valuable question.
Because once you see customer history by cohort, you stop treating your database as one population.
You begin to see how growth decisions compound over time.
Related reading
Ecommerce Customer Segmentation
Behavioral Customer Segmentation
How Repeat Purchase Changes Customer Acquisition Economics