In July 2025, Airtel offered all 360 million of its customers a free year of Perplexity Pro, a plan otherwise priced at ₹17,000. Twelve months later, those free subscriptions are approaching expiry, one sign-up date at a time.
Somewhere at Perplexity, a growth team is staring at one question: who will pay?
The answer won't come from “men aged 25 to 34” or “users in metro cities”. It will come from something far simpler: when did these people join, and what have they done since?
That group, people who share a starting point and are tracked over time, is called a cohort. And it is one of the most misused words in marketing.
What Do Most Marketers Get Wrong About Cohorts?
Sit in any campaign review and you will hear it: “Let’s target the female cohort this week.” It is one of my biggest pet peeves, because that isn’t a cohort. It’s a segment.
A segment is a filter. A cohort is a group you track over time.
So, What Exactly Is a Cohort?
A cohort is a group of people who share a defined starting point or experience and whose behaviour is tracked over time. A segment groups people using selected characteristics or conditions.
Anand Jain, co-founder of CleverTap, explained the difference with a comparison everyone understands: generations.
“If you’re a boomer, if you’re a Gen Z, you’re part of that group. You can’t change yourself.”
A generation is defined by when you were born. You may move cities, change jobs or start earning more, but you never stop being Gen Z. Because the group stays the same, you can follow it over time and see how shared experiences shape its behaviour.
Customer cohorts work the same way. Someone who made their first purchase in January 2026 belongs to the January 2026 cohort for good. They may spend more, switch cities or stop buying, but they never move into February’s cohort. Because the members never change, you can make fair comparisons: are January’s customers still buying after three months, and are they buying more often than February’s?
A segment can’t answer those questions. “Women in Mumbai who ordered above ₹500” gains and loses members every month, as people meet or stop meeting those conditions.
That is why every cohort can be used as a segment, but not every segment is a cohort. For more on how cohort analysis evolved from early medical research to modern analytics, read our guide to
How Do You Build a Cohort That Actually Tells You Something?
Start with how the customer arrived or what they did, not simply who they are. Four rules will help you build useful cohorts.
Define the starting point. Use the first purchase month, acquisition source, campaign or another clearly defined event. Anand’s example was everyone acquired during an Asia Cup campaign.
Freeze the definition. Once January ends, the January acquisition cohort is fixed. Customers acquired in February belong to a different cohort.
Keep cohorts separate. Track January’s and February’s new customers independently so you can compare their behaviour over time.
Match the rhythm of your category. Monthly cohorts work well for many D2C brands. A food delivery app may benefit from weekly cohorts, while a business with longer purchase cycles may need quarterly views.
Consider a skincare brand that acquires 8,000 customers during its Diwali sale and 3,000 in an ordinary January. Look at all 11,000 together and the repeat-purchase rate might appear healthy. Split them into two cohorts, however, and a different story could emerge. The Diwali buyers may have come for the discount and never returned, while the smaller January group keeps reordering.
Same brand. Same products. Two very different customer behaviours. Location, category and spend still matter, but they become additional ways to analyse a cohort. You could examine January customers from Pune who bought sunscreen first and compare them with customers who started with moisturiser.
What Does a Cohort Table Actually Show You?
Ganesh Balakrishnan, co-founder of Flatheads, explained on our podcast that you read this table in two directions. Along a row, you follow one group over time: of everyone who joined in January, how many came back in February, March and April? Down a column, you compare different cohorts at the same stage of their lifecycle. Did March’s newcomers repeat faster in their first month than January’s customers did in theirs?
Ganesh also explained how cohort analysis fits into a broader understanding of customers. He starts with behavioural splits, such as full-price buyers, sale hunters and early adopters, moves to RFM analysis, and then builds acquisition and category cohorts. RFM stands for recency, frequency and monetary value: when customers last purchased, how often they buy and how much they spend.
“By segment you create cohorts and then start improving those, and this actually dictates your top of the funnel.”
Ganesh Balakrishnan, co-founder, Flatheads
What Does a Healthy Cohort Look Like Over Time?
A healthy retention curve does not have to follow one fixed shape. Some cohorts decline and then stabilise. Others continue to lose customers, while some rebound as people discover lasting value in a product.
Anand describes one possible pattern as the retention smile curve. A new cohort starts with high activity because people are exploring. Usage then dips as less interested users drop off. If a product becomes part of people’s routines, engagement can pick up again.
Think of a food delivery app. A new user orders three times in the first fortnight while trying different restaurants. In week four, they go quiet. By week eight, those who have found a favourite biryani place may be ordering every Friday. That return is the upturn of the smile.
Read retention over three, six and twelve months depending on the category’s purchase cycle. Also distinguish between customers who return to an app and customers who actually buy. Those are different signals, and both matter.
Types of Cohort Analysis
Cohort analysis can be used in different ways depending on the business question you want to answer. Two of the most common approaches are acquisition cohort analysis and behavioural cohort analysis. Revenue and value-based analysis can add another useful layer.
What Should You Do When a Cohort Goes Quiet?
Act before the renewal or repeat-purchase moment, not when it arrives. Go back to Perplexity. The most useful signals for predicting who might pay when a free subscription ends are likely to emerge before the expiry reminder. Did people build a habit in months two to eleven? Did they return regularly? Did they discover features that made the product valuable enough to keep paying for?
Anand told me about a unicorn founder whose growth team has just two jobs. One group brings dormant users back into the app. The other helps users do something meaningful once they are there. Neither job is simply about offering discounts.
Look at cohorts of users who return and browse, not only cohorts of buyers. Identify when browsing turns into buying, and investigate what separates people who eventually purchase from those who remain inactive. A customer who opens your app every week but has not bought in two months is not necessarily lost. Their browsing may signal continued interest, but you need to understand what is stopping them before deciding how to re-engage them.
How Do You Prove Your Retention Effort Is Working?
Hold a group back and compare. When I ran retention, we kept 10% of the customer base aside as a universal holdout and refreshed it every three months. Customers we communicated with had a 90-day repeat-purchase rate of 10%. The holdout came in at 5%, a five-percentage-point difference.
The comparison gave us a way to evaluate the potential incremental value of our retention programme. To establish that the programme caused the difference, the groups need to be comparable, with consistent eligibility rules and measurement windows and an appropriately designed holdout.
Will AI Make Cohorts Unnecessary?
Beerud Sheth, founder of Gupshup, believes AI will take marketers from a handful of segments to thousands, even millions.
“If AI can run and orchestrate these campaigns, it can have a thousand or a million segments.”
Beerud Sheth, founder, Gupshup
In conversational commerce, he argues, customers can tell brands what they want in their own words. This opens the door to individualised interactions, where a segment could be as small as one customer.
When every customer receives a different message, marketers still need to know whether the overall customer experience is improving. Are people acquired this quarter staying longer than those acquired last quarter? Are customers from a new campaign returning more often? Are AI-driven recommendations producing incremental purchases, or changing what customers would have bought anyway?
Personalisation decides what to say. Cohort analysis helps you understand whether your customer relationships are getting stronger.
The more personalised marketing becomes, the more important it is to measure its results rigorously.
Where Should You Start This Week?
Start small. Pick last month’s new customers and freeze that list. Track how many return, how many buy again and how their behaviour changes over the following months. Then compare them with the previous month’s cohort.
Once that becomes a habit, add acquisition-source and category cohorts. Introduce a holdout group when you want to measure the incremental impact of retention campaigns. You do not need a complicated dashboard to begin. You need a clearly defined group, a consistent measurement window and the discipline to compare like with like.
And the next time someone says “cohort” in a meeting, ask one question:
Who joined, and when?
If there is no clear answer, you are probably talking about a segment.
Want help setting up cohort analysis for your brand? Write to us at alibha@daiom.in.
