What cohort analysis in marketing shows, how to do it, and where it goes wrong

Big-city marathons don't send 40,000 people off the line at once. Runners start in waves, sometimes half an hour apart, so checking who's ahead on the race clock mostly tells you who started first. That's why runners wear timing chips, which measure each person from the moment they cross their own start line.

Cohort analysis gives your customers the same treatment. It groups customers by when (or how) they started and compares them at the same point in their relationship with your brand. That matters because blended numbers can hide a real problem. Your overall repeat rate can hold steady while this year's new customers come back far less often than last year's, which changes what you can afford to pay per customer and where your marketing budget should go.

Key takeaways

• Cohort analysis groups customers by a shared starting point, like their first purchase month, so you compare them at the same stage instead of the same calendar date.
• "Cohort" can mean a group of customers or a generation of consumers in marketing, and the two types answer different questions.
• How you acquire customers shapes what they're worth, and one study found that a 35% acquisition discount cut long-term customer value roughly in half.
• Rows, columns, and diagonals in a cohort table each tell you something different, and diagonals reveal calendar events like a shipping delay.
• The most common mistakes are comparing cohorts of different ages, judging them on revenue instead of margin, and treating a pattern as the reason behind it.
• Customer cohorts only include buyers you can identify, so omnichannel brands miss retail and marketplace shoppers without broader measurement.

What is cohort analysis?

Cohort analysis is a form of data analysis that sorts people into groups based on a shared starting event, then follows each group over time. A cohort is a group of people who share a defining characteristic, such as the month they first bought from you, the campaign that brought them in, or their registration date on your app. Everyone in one cohort is measured from their own day zero, so a January customer at month three gets compared with a March customer at month three.

It helps to separate the cohort from the metric. The cohort is who you're tracking, and the metric is what you measure about them, whether that's retention rates, repeat purchases, revenue, or contribution margin. Pick the metric first, since it shapes what you'll learn about how customers behave.

Customer cohorts vs. generational cohorts

Some search results on this topic describe something completely different. There are two kinds of cohort analysis in marketing, and they share a name but not much else.

Schewe and Evans (2000) argue that the events people live through as young adults shape values that stick with them, which makes birth cohorts more useful segmentation criteria than age alone. That's useful for brand positioning and marketing strategies, but this guide covers customer cohort analysis, since that's what most teams mean when they use cohort analysis to measure performance.

Types of cohorts marketers use

Most guides stop at time-based cohorts and behavioral cohorts, but there are several ways to group customers depending on the question you're asking. These five cover most needs:

Acquisition cohorts are the easiest place to start with cohort analysis. Behavioral cohorts offer the most customer understanding, because they connect user behavior to behavioral patterns that predict long-term value. Comparing the conversion rate to a second order for loyalty program members against non-members, for example, can help you identify opportunities for your retention strategies.

What marketers use cohort analysis for

Most teams use cohort analysis to spot what averages hide, especially when facing marketing challenges like rising acquisition costs. Here's where it pays off.

Judging acquisition quality by channel

Two channels with the same first-order ROAS can bring in very different customers. Verhoef and Donkers (2005) found that retention and cross-buying at a financial services provider varied by acquisition channel. Your mix won't match theirs, so check how customers respond to your own marketing campaigns. Six months of channel cohort data gives you a much deeper understanding of acquisition quality than first-order numbers.

Testing whether promotions bring in lasting customers

Discounts grow your customer count fast, but those customers don't always stick. Lewis (2006) analyzed customer data from a newspaper and an online grocer and found that a 35% acquisition discount produced customers worth about half as much over the long run as those acquired without one.

Setting what you can afford to pay for a customer

Cohorts show how long each group takes to pay back its acquisition cost, which grounds your CAC target and customer lifetime value (LTV) estimates. Customer retention is usually the biggest lever. In an analysis of five public companies, Gupta, Lehmann, and Stuart (2004) found that a 1% improvement in retention raised customer value by about 5%, far more than the same improvement in acquisition cost. That's why teams that improve retention see it in their cohorts first.

Finding where customers drop off

Cohort analysis also shows when you lose people. If most churn happens between the first and second order, that's the point in the customer lifecycle where post-purchase emails and customer experience fixes will do the most to reduce churn and retain customers.

How to do cohort analysis, step by step

Here's what the cohort analysis work looks like for a hypothetical skincare brand selling through its own online store, Amazon, and Target. The steps hold whether it's your first marketing cohort analysis or your fiftieth.

1. Start with a business question. "Are paid social customers worth less after 180 days than paid search customers?" tells you what to group and measure.
2. Define cohorts with written rules. Decide who counts as a first-time buyer, how refunds are handled, and how you'll match one person across emails and devices.
3. Gather and check your cohort analysis data. Pull order records with customer IDs, first purchase dates, discounts, and costs. For our skincare brand, this is where Amazon and Target shoppers drop out of view.
4. Match time windows to how often people buy. Skincare might use 30, 60, and 90 days, while a mattress might need a year.
5. Pick a metric tied to profit. Repeat purchase rate (the conversion rate from first to second order) is a start, but contribution margin per customer is better.
6. Build the table and compare at equal ages. Put cohorts in rows and time since first order in columns.
7. Investigate before you act. When one cohort looks weak, list what changed for it (offers, products, pricing, and shipping) before blaming a channel.

You can create cohorts in a spreadsheet, a BI tool, or Google Analytics. The cohort exploration in Google Analytics 4 handles user engagement and user retention questions, like whether new users from a launch week return. It tracks user behavior, not profit, so use order data for financial questions.

How to read a cohort table

Cohort tables look intimidating, but there are only three directions to read them. This hypothetical example shows the share of each monthly cohort that bought again in a specific period after their first order.

When you analyze cohorts this way, each direction answers a different question:

• Rows show how a particular cohort ages.
• Columns compare different cohorts at the same age. Read alone, the Month 1 column makes March look weak.
• Diagonals show one calendar month across several cohorts. The bolded numbers all fall in April, which points to something that hit everyone that month, like a shipping delay, rather than a problem with March's customers.

Rentz, Reynolds, and Stout (1983) used this kind of thinking to show that snapshot data had overstated how much soft drink consumption would fall as people aged, because the habit traveled with each generation. It also keeps you from misreading shifts in cohort behavior and customer behavior.

Common cohort analysis mistakes

Building a cohort table from your data is easy, but interpreting cohort analysis results is where most marketing analysis goes sideways.

Comparing cohorts at different ages

A four-month-old cohort can't be fairly compared with one that's had a full year. Line them up at matching ages and label any projection for a younger group as a forecast.

Judging cohorts on revenue instead of margin

Higher revenue doesn't always mean a better cohort, as this simplified example shows.

Cohort B wins on revenue but leaves you with less. Define the outcome you care about, and keep groups consistent (don't mix first-time buyers with existing customers), before you declare a winner.

Reading too much into small cohorts

Groups shrink fast once you split them. If 10 of 50 customers return, that's 20%, and five more returns makes it 30%. That's five people, so check how many customers are in each group before reacting.

Mistaking survivors for growing loyalty

A cohort's retention curve often flattens or rises over time, which looks like growing loyalty. Fader and Hardie (2010) show that rising retention rates within a cohort usually come from sorting. The least committed customers leave first, so loyal customers make up more of what's left.

Treating a pattern as the reason

A cohort table shows that two different groups behave differently, but not why. Customers credited to your social media campaigns may have also seen other ads, gotten a different offer, or arrived during a seasonal rush.

Missing the customers you can't see

Customer cohorts only include buyers you can identify, and most retail and marketplace purchases don't carry a customer ID. A customer journey that starts with a paid social ad and ends at a Target register won't appear in any cohort. Channel cohorts also inherit your attribution's channel assignments, usually last click. Neither view captures halo effects—the revenue a campaign drives indirectly through another channel, like Amazon or retail—so cohorts can make some campaigns look weaker than they are.

Cohort analysis for marketers vs. data analysts

Both roles get value from cohort analysis, but they use it differently. Here's how the work usually splits.

For marketers, learning to read tables pays off faster than learning to build complex ones, and knowing a chart's limits leads to more informed decisions. Analysts do their best work when they analyze cohorts with a decision in mind, digging into specific cohorts tied to a budget call. Simple tables produce deep insights when both sides agree on the question first.

How cohort analysis fits with your other measurement

Cohort analysis is a powerful tool for understanding customer behavior, but it answers one kind of question. Here's how it lines up with other ways to measure marketing.

Cohort analysis shows who you're acquiring, but it can't show which marketing efforts drive those purchases, especially when much of your revenue happens off your site.

Where Prescient comes in

Cohort analysis shows what your customer base is worth after people arrive. Prescient's marketing mix model answers the question that comes before that: which marketing campaigns are driving revenue across your site, Amazon, and retail partners, including the halo effects your customer data can't see. Because our MMM updates daily, you can watch campaign performance shift without waiting on a quarterly readout.

Pair the two, and you can set acquisition targets from cohort payback, then use Media Forecaster to see what happens when you scale a channel or shift budget into it from another. Book a demo and we'll show you around the platform and all the features to help you understand what's truly driving growth for your brand.

FAQs

What is a cohort analysis?

A cohort analysis groups people by a shared starting point, such as the month of their first order, and tracks each group over time. That lets you compare groups at the same stage, which a single average across your customer base can't do.

What are the three main types of cohort studies?

In medical and scientific research, the three main types are prospective studies, which follow a group forward, retrospective studies, which look back through historical records, and ambidirectional studies, which do both. Most marketing cohort analysis is retrospective, since it relies on past order data.

What is cohort 1 and cohort 2?

Cohort 1 and cohort 2 are labels for two groups you're comparing, usually defined by when they started. In a marketing cohort analysis, cohort 1 might be customers acquired in January and cohort 2 those acquired in February. Comparing them at matching ages shows which group is performing better.

Is cohort analysis qualitative or quantitative?

Customer cohort analysis is quantitative, since it measures user behavior like repeat orders, revenue, and customer retention. Generational research can be both, often starting with interviews and then looking for shared characteristics in the responses.

What's the difference between cohort analysis and segmentation?

Segmentation groups customers by common characteristics like location, spend, or interests, usually as a snapshot. Cohort analysis adds time, tracking how each group changes from a shared starting event. A marketing cohort analysis often layers segments inside cohorts, like big and small first-order spenders acquired the same month.

How long should you track a customer cohort?

Track a cohort at least as long as a typical customer takes to buy again, ideally through your payback period. Consumable brands can learn a lot in 90 days, while big-ticket brands may need a year. Monthly cohorts work for most brands.