Cohort analysis: growing ecommerce revenue

Cohort analysis: what most shops get wrong

Most shop owners segment their customers by demographics. Age. Gender. Postcode. It is convenient because the data is easy to get. But it is useless when it comes to buying behaviour.

A 35-year-old man in Munich does not buy differently from a 62-year-old woman in Leipzig if both are looking for trainers. What counts is not who the customer is, but what they do.

Cohort analysis groups customers by shared actions, not shared traits. Everyone who bought for the first time in January 2024 forms a cohort. Those who came back within 30 days of their first purchase form a subgroup. Those who had not returned after 90 days form another. These groups reveal patterns that demographic data never would.

Why classic customer segmentation fails

A sporting goods shop with around 120,000 sessions a month segmented its customers by age: 18–30, 31–45, 46+. The result: the conversion rate differed between the groups by 0.3 percentage points at most.

Statistically irrelevant. The buying decision did not depend on age but on whether the visitor had been on the site before, in other words whether they were a new customer or a repeat buyer.

The gap between new customers and repeat buyers: 4.2 percentage points of conversion rate. Fourteen times the gap of the age segmentation. The consequence: the marketing budget moved from age-based campaigns to cohort-based retention flows. The effect after 6 months: +18% revenue from the existing-customer segment at the same marketing spend.

The problem with classic segmentation is that it rests on assumptions. “Young customers buy more online.” Maybe, but not in this shop. “Regular customers buy more often.” Yes, but when exactly? After 14 days? After 60? The answer is in the cohorts, not in the assumptions.

The three cohorts every ecommerce shop should track

1. First-purchase month cohorts

The basic question: how many customers who first bought in month X come back in month X+1? In month X+3? After a year? This curve shows the real quality of your customer acquisition, not the cost per click.

A furniture shop ran this analysis. January first-time buyers: 12% returned in February, 8% in March, 5% after 6 months. April first-time buyers: 18% in May, 14% in June, 11% after 6 months. The difference?

In January an aggressive discount campaign was running on Google Ads. Many first-time buyers only bought because of the discount. In April there was no discount, so customers bought because they needed something. Retention was higher and customer lifetime value was 34% better.

The consequence for the budget: January CPA (cost per acquisition) was 18 euros, April CPA 24 euros. At first glance January looked more efficient. But 12-month retention of January customers was so low that lifetime value per customer was 45 euros, against 89 euros for April customers. The channel that looked more expensive was the profitable one.

2. Repurchase interval cohorts

Not all customers buy at the same rhythm. A cosmetics shop identified three groups: those who buy every 30 days (subscription-like), those who buy every 60–90 days (need-driven) and those who buy irregularly (impulse). Each group needs a different approach.

The 30-day group was offered a subscription: 15% off for monthly delivery. Assumption: higher retention, lower marketing effort. Result: 23% of the group took the subscription and retention rose to 94% (from 71% before). But average order value fell by 8%, because subscribers bought fewer extra items on impulse.

The 60–90-day group got reminder emails not after 30 days but after 75, right in the middle of their interval. The assumption: remind too early and you annoy people, too late and they have forgotten the shop.

Result: the reminder open rate rose from 12% to 31%. The click rate from 2.4% to 8.7%. The conversion rate of returning visitors from 4.1% to 7.3%.

The impulse group, the irregular buyers, got no reminders but new-arrival alerts. When new products arrived in their favourite categories, they received a notification. Result: +14% revenue from this group, measured over 4 months. The trick was not frequency but the relevance of the timing.

3. Checkout abandonment cohorts

Someone who abandons the cart is not lost. But not every abandonment is the same. A fashion shop segmented abandonments by stage: drop-off on the shipping page, on the payment page and on the confirmation page.

The results were revealing: 64% of abandonments happened on the shipping page, mostly because of unexpectedly high shipping costs or long delivery times. 21% on the payment page: payment method problems or technical errors. 15% on the confirmation page: often doubt or distraction.

Each cohort got a different recovery email. The shipping cohort got a shipping code (+12% recovery rate compared with a generic email). The payment cohort got an email listing alternative payment methods (+8% recovery rate). The confirmation cohort got an email with social proof: customer reviews and photos (+6% recovery rate).

The total effect: the recovery rate rose from 4.2% to 9.8%. With 3,200 abandonments a month, that means 179 extra orders. At an average order value of 87 euros: 15,573 euros of extra revenue per month, just from cohort-specific emails instead of one generic one.

How to set up cohort analysis properly

Many shops have the data but not the structure. Google Analytics 4 offers cohort reports. They are a start, but not enough for operational decisions. What you need:

1. A single customer ID system. Without an ID that follows the customer across sessions, you cannot build cohorts. That means mandatory login or at least a stable cookie system that links customer ID and purchase history. A shop that relies only on session cookies loses 40–60% of its cohort assignments.

2. Clear time windows. “Repeat buyers” is not a cohort. “Customers who returned between day 30 and day 60 after their first purchase” is a cohort. The more specific the window, the more meaningful the analysis. We typically work with 7-day, 30-day, 90-day and 365-day windows.

3. One metric that counts. Not every metric matters for every cohort. For first-purchase cohorts, 30-day retention counts. For repurchase cohorts, the average interval counts. For abandonment cohorts, the recovery rate after 24 hours counts. Pick one primary metric per cohort, not five.

An electronics shop tried to judge all cohorts by the same metric: conversion rate. The result was confusing. The checkout abandonment cohort had a “conversion rate” of 4.2%, but that was the recovery rate, not checkout completion.

The numbers were not comparable and the analysis was worthless. After switching to cohort-specific metrics, the reporting became usable for operations within 2 weeks.

The biggest mistakes in cohort analysis

Mistake 1: Cohorts that are too small

A shop with 5,000 sessions a month defined 14 cohorts. Each cohort had 40 customers on average. The result: statistically significant conclusions were impossible. The swings from month to month were bigger than the differences between cohorts.

The rule: at least 200 customers per cohort for a statistically robust analysis. For smaller shops: fewer cohorts with wider time windows. Instead of “customers who bought in January”, use “customers who bought in Q1”. The granularity has to match the traffic.

Mistake 2: Confusing correlation with causation

A cohort of customers acquired via Instagram had 40% higher retention than Google Ads customers. The conclusion: Instagram is the better channel. The reality: 70% of the Instagram customers already followed the brand, so they were pre-qualified. Google Ads customers were cold traffic. The cohort said something about the audience, not about the channel.

Before you move budget, ask: what really sets this cohort apart? The channel? The campaign? The product they bought? Or something that happened before the first click?

Mistake 3: Analysing cohorts but not acting

The most common form of cohort analysis is the dashboard nobody looks at. One shop produced monthly cohort reports, 12 slides of PowerPoint. The reports were shown in meetings but never turned into action. After 8 months the reporting was dropped: “It doesn’t achieve anything.”

Cohort analysis is not a reporting tool. It is a decision tool. Every analysis has to lead to a concrete action: moving budget, changing an email flow, changing a landing page, adjusting pricing. If a cohort analysis does not change a decision, it is a waste of time.

Cohort analysis and A/B testing: getting the balance right

Cohort analysis shows what happens. A/B testing shows why it happens, and whether a change makes it better. Each needs the other.

A real example: the cohort analysis showed that customers who bought on mobile had 30% lower retention than desktop buyers. The assumption: the mobile UX is poor. The A/B test: an optimised mobile checkout flow against the existing one.

Result: retention in the mobile cohort rose by 14%, but not to desktop level. The remaining gap was not about UX but about the product: mobile customers more often bought low-priced items that were used up faster. The lower retention was partly a product effect, not a UX effect.

The cohort had found the problem. The A/B test had narrowed down the cause. Without both, the team would have spent months on mobile UX work that explained only half the story.

Frequently asked questions about cohort analysis

What is cohort analysis in ecommerce?

Cohort analysis groups customers by shared actions, usually the date of their first purchase, and tracks their behaviour over time. Unlike classic segmentation (age, gender, region), it is based on behavioural data and shows patterns such as retention, repurchase rates and lifetime value per group.

Which cohorts should an ecommerce shop track?

The three most important cohorts are: first-purchase month cohorts (show acquisition quality), repurchase interval cohorts (show buying rhythm and retention potential) and checkout abandonment cohorts (show recovery opportunities). Other useful cohorts: channel cohorts (where did the customer come from?), product cohorts (what did they buy first?) and device cohorts (mobile vs. desktop).

How big does a shop need to be for useful cohort analysis?

At least 200 customers per cohort for statistically robust results. A shop with 1,000 purchases a month can usefully analyse 2–3 cohorts. At 5,000 purchases a month, 5–7 cohorts are realistic. Below that, the granularity is too low for operational decisions, and classic A/B testing helps more.

Which tool is suitable for cohort analysis?

Google Analytics 4 offers basic cohort reports. The Academy track Google Analytics 4 for CRO ↗ gives you a full introduction to the tool. For deeper analysis, use Mixpanel, Amplitude or custom dashboards built on your shop database. The choice depends on whether you need real-time data (Mixpanel/Amplitude) or monthly reviews are enough (GA4 + export).

How does cohort analysis differ from classic segmentation?

Classic segmentation groups by static traits: age, gender, region. Cohort analysis groups by dynamic actions: when was the first purchase? How quickly did the repeat purchase follow? Which channel led to the purchase? Cohorts show behaviour patterns over time. Segmentation shows snapshots.

How often should you analyse cohorts?

Monthly for operational cohorts (first purchase, repurchase). Weekly for tactical cohorts (checkout abandonment, campaign performance). The analysis should not take longer than the decision that follows from it. If a report takes you 2 weeks, the process is too heavy.

Can B2B shops use cohort analysis too?

Yes, with adjusted time windows. B2B customers buy less often and the intervals are longer. Instead of 30-day retention, you measure 90-day or 180-day retention. The logic stays the same: group by shared actions, track behaviour over time, find patterns.

What is the difference between cohort analysis and calculating CLV?

Customer lifetime value (CLV) is a metric: the average revenue per customer over the whole relationship. Cohort analysis is a method for understanding CLV.

Instead of calculating one global CLV, cohort analysis shows that January first-time buyers have a CLV of 45 euros and April first-time buyers one of 89 euros. The global CLV of 67 euros is useless for operational decisions. The cohort-specific values are not.