Cohort Analysis & Retention

The mobile app that fixed churn before investors noticed

In 2021, a language-learning app was growing at 12% month-over-month in total active users. The team was confident. Investors were interested.

Then a new analyst ran a cohort analysis.

The picture was alarming. Of every 100 users who installed the app in Month 1 of the year, only 11 were still using it by Month 6. The aggregate growth looked healthy because acquisition was growing faster than the users were leaving. The growth masked a leaky bucket.

More alarming: cohorts acquired through paid ads retained at 8% by Month 6. Cohorts acquired through organic search retained at 22%. Cohorts acquired through referrals retained at 34%.

The team had been scaling their ad spend. The channel scaling fastest was producing the worst long-term users.

They cut paid acquisition by 60%, reallocated to referral incentives and onboarding improvements. Six months later, Month-6 retention across all cohorts was 28% — and the underlying economics had transformed.

The aggregate dashboard would never have revealed this. Cohort analysis did.

(Illustrative scenario based on patterns common in marketing analytics. Specific figures are representative of real-world outcomes — not a verified account of a specific named company.)

What cohort analysis is

A cohort is a group of users who share a common characteristic — most commonly, when they first became a customer.

Cohort analysis tracks what each cohort does over time, rather than looking at all users aggregated together.

Why aggregates lie:

If you have 1,000 users this month and 1,100 next month, you might assume 10% growth. But those 1,100 might be: 900 of the original users who stayed, plus 200 new ones — meaning 100 of your original users left. The aggregate showed growth; the reality was net negative for the original cohort.

Cohort analysis makes churn and retention visible. Aggregate metrics hide it.

A retention cohort table:

Reading this table: the April cohort is retaining better than January at every comparable month (78% in Month 1 vs 68%). Something changed between January and April — either product improvements, acquisition channel mix, or onboarding quality — that made later cohorts retain better.

Retention curves: understanding natural churn patterns

Retention curves are visual representations of cohort retention over time. They reveal predictable patterns:

The initial drop is the most important part of the curve. If Day-7 retention is 20%, you're losing 80% of new users in the first week. This is a product or onboarding problem, not a marketing problem. More acquisition spend will not fix it.

Flattening the curve: A healthy retention curve flattens after the initial drop — the remaining users are habitual and stay. A retention curve that continues declining toward zero means there's no sustainable user base.

Improving Day-7 retention is often the highest-leverage marketing action. Doubling Day-7 retention from 20% to 40% means the same acquisition investment generates twice as many 30-day users, twice as many 90-day users, and dramatically higher LTV.

Acquisition cohorts: which channels produce the best customers?

Not all acquisition channels produce equivalent customers. Cohort analysis by acquisition source reveals the truth:

This table shows that social ads have the lowest CAC after organic search — but also the worst retention. The LTV:CAC tells the real story: social ads are barely profitable; organic search and content marketing are outstanding.

CAC alone is a misleading optimisation target. Optimise for LTV:CAC, which requires knowing the LTV of customers acquired through each channel.

The RFM model: segmenting existing customers

Beyond acquisition cohorts, marketers segment existing customers to understand who to invest in:

RFM analysis:

• Recency: How recently did they last purchase or engage?
• Frequency: How often do they purchase or engage?
• Monetary value: How much do they spend?

Customers with high recency, high frequency, and high monetary value are your best customers. They're also the most valuable for:

• Lookalike audience creation (who resembles your best customers?)
• Referral programme targeting (happy high-value customers are your best advocates)
• Upsell and cross-sell campaigns

Customers with low recency but historically high frequency and monetary value are lapsed high-value customers — the highest-priority re-engagement target. They were excellent customers who stopped. Understanding why they stopped is often the most valuable research you can conduct.

There Are No Dumb Questions

"How do I run a cohort analysis if I don't have the data?"

You need two things: a customer list with acquisition dates, and some measure of activity or purchase over time. Most email platforms, e-commerce platforms (Shopify, WooCommerce), and subscription systems (Stripe) have this data — you may just need to export and analyse it. GA4 has a built-in Cohort Exploration report that shows retention by first session date. If you're starting from scratch: start tracking acquisition source at signup now, and in 6 months you'll have your first meaningful cohort comparison.

"What retention rate is 'good'?"

It varies dramatically by category. Mobile games: 5–20% Day-30 retention is commonly cited (AppsFlyer/Adjust mobile gaming benchmarks — 10–25% represents stronger-performing titles). Productivity apps: 20–40%. SaaS: monthly churn for healthy businesses is typically 1–3% (97–99% monthly retention). Churn above 3–5% monthly is considered concerning; above 5% is critical. Healthy SMB SaaS commonly targets under 2–3% monthly churn; enterprise SaaS typically targets under 1–2%. (ChartMogul SaaS Benchmarks — verify current data by ARR tier) Consumer subscription (streaming, food): 70–85%. The most important benchmark is your own cohort trend over time: are later cohorts retaining better than earlier ones? Improving your own baseline matters more than benchmarking to industry averages.

Back to the language-learning app

The aggregate retention metric — overall active users, growing 12% month-over-month — was telling a true story about a false reality. The business looked healthy because new users were arriving faster than existing ones were leaving. Cohort analysis revealed what the aggregate was hiding: paid ad cohorts retained at 8% by Month 6, while referral cohorts retained at 34%. The channel scaling fastest was producing the worst long-term customers. Cutting paid acquisition by 60% and reallocating to referral incentives and onboarding improvements felt counterintuitive against a backdrop of growth — but six months later, Month-6 retention across all cohorts had improved from 11% to 28%. The company hadn't fixed its acquisition. It had fixed who it was acquiring. Aggregate metrics show you the water level; cohort analysis shows you whether the bucket has a hole.

Key takeaways

• Aggregate metrics hide churn. A growing total user count can mask accelerating churn if acquisition grows faster than retention declines. Cohort analysis makes the reality visible.
• Retention is often more valuable than acquisition. Improving Day-7 or Month-1 retention by 20% has the same economic effect as reducing CAC by 20% — but often costs less to achieve.
• Acquisition channel affects long-term customer quality. Cheap acquisition channels that produce churned customers are often more expensive in LTV:CAC terms than premium channels producing loyal ones. Always measure LTV by source.
• RFM segmentation identifies who to invest in. High-recency, high-frequency, high-value customers are your lookalike seeds, referral advocates, and upsell targets. Lapsed high-value customers are your highest-priority win-backs.
• The retention curve reveals the real problem. A retention curve that keeps falling toward zero indicates a fundamental product-market fit issue. More acquisition cannot fix churn — only product and onboarding improvements can.