Cohort Analysis Explained: How to Spot Quality Users, Not Just Cheap Installs

Early in my career I killed a channel that was actually my best one. It looked expensive on day one, so I cut it. A month later I realized the users it brought were the only ones still spending money, while the cheap channels I kept were gone within a week. That taught me to stop judging users on the day they showed up and start watching what they did over time. That is what cohort analysis is for.

If you have ever stared at a colorful triangle-shaped table and felt your eyes glaze over, this is for you. I want to walk you through what a cohort is, why grouping people by signup week changes everything, and how to use that view to decide where your budget goes.

What a cohort actually is

A cohort is just a group of people who started at the same time, usually built by signup or install week. Everyone who installed your app the week of June 1 is one cohort, everyone who installed the week of June 8 is another. You can group by month if volume is low or by day if you have a lot of traffic, but week is the sweet spot for most US apps and stores.The reason this matters is timing. If you look at all your users mashed together, you are mixing someone who signed up yesterday with someone who has been around for six months. Their behavior is completely different, and the average hides that. Cohorts keep each group separate so you compare apples to apples: how did week one look for every group, regardless of when they joined? It is like school classes; you would never average the class of 2024 and 2025 together, you look at each on its own.

Retention cohorts vs revenue cohorts

There are two questions cohorts answer, and they are not the same. The first is retention: out of the people who joined, how many are still active later? The second is revenue: how much have they spent by a given point? You want both, because a cohort can stick around without spending, or spend fast and then vanish.Retention cohorts are shown as a percentage. If 1,000 people installed and 300 opened the app again on day 7, your Day 7 retention is 30 percent. For an ecommerce store, retention might mean a repeat visit or a second order rather than an app open.Revenue cohorts are shown in dollars, usually cumulative per user. A cohort might be at 4 dollars per user by Day 7, 9 dollars by Day 30, and 18 dollars by Day 90. This is how you watch lifetime value build up over real time instead of guessing it. Line revenue cohorts up against what you paid to acquire each group and you can see exactly when a channel pays itself back.

How to read the cohort table (the triangle)

The triangle shape confuses everyone at first, so here is the trick. Each row is one cohort (say, users who joined the week of June 1). Each column is an age, not a calendar date: Day 0, Day 1, Day 7, Day 14, Day 30. So column three means the same thing for every row, the user's third measured point in their own life, no matter when they joined.The table looks like a staircase because newer cohorts have not lived long enough to fill in the later columns. The June 1 cohort can show Day 30 data, but the cohort that joined last week can only show Day 1 and Day 7. That empty corner is why it looks like a triangle.Read it two ways. Across a single row, you see how one cohort decays or grows over its lifetime. Down a single column, you compare the same life-stage across cohorts: is Day 7 retention getting better or worse as you change targeting, creative, or onboarding? Most tools color the cells so patterns jump out fast.

Good curves, bad curves, and what they tell you

A healthy retention curve drops at first and then flattens out. You always lose people in the first few days, that is normal. What you want is for the line to level off into a stable plateau, because that flat part is your core of users who genuinely stick. A product with a 25 percent Day 30 that holds steady beats one with a higher early number that keeps sliding toward zero.A bad curve keeps falling and never flattens. That usually means people are not finding real value, and no amount of ad spend fixes it, you are just pouring users into a leaky bucket. Here is what I look for:

• The plateau: does retention settle at a stable floor, or keep dropping to nothing?
• The smile: in some products, revenue per user curves back up as your best customers spend more over time. Great sign.
• Early cliffs: a brutal Day 1 drop often points to broken onboarding or a misleading ad promise, not a bad product.
• Cohort drift: if newer cohorts retain worse than older ones at the same age, something changed, often you scaled spend into lower-quality audiences. That is the warning light I watch most closely when pushing budget up.

Using cohorts to make budget decisions

Here is where it pays off. Say Channel A brings installs at 2 dollars each and Channel B at 4 dollars. On day one, A looks like the obvious winner. But pull the revenue cohorts: Channel A users are worth 3 dollars by Day 90, while Channel B users are worth 11 dollars. Channel A is losing money on every install, Channel B is a machine. Without cohorts you scale the wrong one, which is exactly the mistake I made years ago.The practical move is to tag each cohort by the channel or campaign that brought it, then compare retention and cumulative revenue side by side at the same ages. You want two things: which channels bring users who stay, and how long until each channel's revenue cohort crosses the cost you paid to acquire it. That payback point tells you how aggressively you can scale.One caution: give cohorts time to mature before you make big calls. A week-old cohort cannot tell you about Day 30 value. Use Day 1 and Day 7 retention to catch obvious disasters early, but wait for the curve to develop before you crown a winner. Cohort analysis rewards patience, and it is the clearest way I know to tell quality users apart from cheap ones.