Cohort Analysis in Mobile Analytics: What It Is, Metrics, and Stages of Conducting

What Is Cohort Analysis?

Cohort Analysis is an analytics method that divides users into groups (cohorts) based on when they performed their first target event, such as installing the app or registering. Unlike aggregated metrics like DAU or MAU, cohorts allow you to track the behavior of the same group over weeks and months, revealing retention and churn trends.

The main advantage of this method is eliminating the illusion of growth, where older users mask the problems of new ones. According to Amplitude Blog (2024), companies that adopted cohort analysis reduce user churn by 20–30% in the first 90 days.

Apply cohort analysis to evaluate UA campaign effectiveness, compare behavior across different app versions, and calculate the return on investment in user acquisition.

Key Metrics of Cohort Analysis

Retention Rate

Retention Rate is the percentage of a cohort's users who returned to the app N days after installation. This is the primary product quality metric: if users don't return, the issues lie in onboarding, value proposition, or app stability. According to Statista (2025), the average Day-1 Retention for mobile apps is 25–30%, Day-7 — 10–15%, Day-30 — 5–7%.

Compare your cohort's Retention against category benchmarks and analyze which product changes increased returns.

LTV — Lifetime Value

LTV (Lifetime Value) is the total revenue from all users in a cohort over the observation period. This metric is critical for evaluating advertising campaign ROI and making decisions about scaling UA budgets. According to the Adjust Benchmarks report (2024), the median LTV per user in gaming is $3.50, in e-commerce — $1.80, and in utilities — $0.40.

Calculate cohort LTV on Day-30, Day-90, and Day-180 to understand the long-term value of acquired users.

Churn Rate

Churn Rate is the percentage of a cohort's users who stopped using the app. High early Churn signals onboarding problems or expectation mismatch. According to Localytics (2024), 25% of users abandon the app after the first open, and by day 90, fewer than 5% remain.

Monitor Churn Rate weekly — a 5% increase in a cohort requires immediate analysis of churn causes through surveys or session recordings.

Types of Cohorts by Time Period

Daily Cohorts

Daily Cohorts are users who installed the app on the same day. This is the most granular slice, allowing you to spot anomalies: a sudden Retention drop on a specific date may indicate a technical failure or a problematic creative campaign. Daily cohorts are used for operational monitoring and A/B tests when changes are made daily.

Weekly Cohorts

Weekly Cohorts are users who installed the app in the same calendar week. These provide more stable data, smoothing out daily fluctuations. They are suitable for regular reporting and comparing advertising channel performance — the weekly step neutralizes the day-of-week effect, where conversion may differ on weekends.

Monthly Cohorts

Monthly Cohorts are users who installed the app in the same month. They are optimal for analyzing long-term trends, seasonality, and calculating LTV over a 6–12 month horizon. They are used at the product management level for strategic decisions about product development.

How to Conduct Cohort Analysis

Cohort Analysis consists of four stages: data collection, grouping into cohorts, calculating metrics by period, and visualization. First, define the target event — typically install, signup, or first_purchase. All users who performed it in one period belong to the same cohort.

Second, set the time interval: day, week, or month. For each subsequent period (Day-1, Day-2, Day-3...), calculate the percentage of returning users. The results are compiled into a cohort table where each row is a cohort, each column is a period, and each cell is a metric value.

Third, calibrate the data: exclude cohorts with fewer than 100 users and account for seasonality. According to Mixpanel (2024), cohorts with fewer than 50 users yield statistically insignificant results with an error margin exceeding 20%. Fourth, visualize the data as a cohort table or heatmap — rows (cohorts) and columns (periods) make retention trends clear for stakeholders.

For correct interpretation, always compare cohorts of the same size: it is unacceptable to compare a cohort of 50,000 users with a cohort of 500. Normalize metrics as a percentage of the cohort's initial size, not as absolute numbers — this allows comparing groups of unequal size. Also consider the novelty effect: users who installed the app after a major marketing event may behave differently from organic installs.

Cohort Analysis Tools

Cohort Analysis Tools are divided into those built into analytics platforms and custom SQL solutions. Amplitude and Mixpanel offer ready-made cohort reports with automatic Retention and LTV calculation, requiring no query writing. Firebase Analytics provides basic cohorts by install date broken down by days, weeks, and months.

For custom cohorts, use SQL queries in BigQuery or Redshift — this gives you full control over metrics and periods. According to Amplitude (2024), teams using custom SQL cohorts discover 35% more behavioral insights than with standard reports. BI tools (Tableau, Looker, Google Data Studio) allow building interactive cohort dashboards for regular monitoring.

The choice of tool depends on the team's analytics maturity. At the start, Firebase Analytics or built-in Amplitude cohorts are sufficient — they cover 80% of cohort analysis needs. As the team grows, switch to SQL cohorts in BigQuery: this provides flexibility in defining non-standard metrics (e.g., cohorts by first purchase date rather than install date) and allows building predictive LTV models using ML. For visualization, use Looker Studio, which connects directly to BigQuery at no additional cost.

Cohort Analysis SQL Example

To calculate daily cohorts yourself, use an SQL query that groups sessions by install date. Below is a BigQuery query that returns the number of active users by day within each cohort.

SELECT
  install_date AS cohort_date,
  DATE_DIFF(session_date, install_date, DAY) AS day_number,
  COUNT(DISTINCT user_id) AS active_users
FROM user_sessions
WHERE install_date BETWEEN '2025-01-01' AND '2025-03-31'
GROUP BY cohort_date, day_number
ORDER BY cohort_date, day_number;

After running the query, build a summary table in Google Sheets or a BI tool. Each row is a cohort date, each column is the day number (Day-1, Day-2, etc.). The percentage of active users is calculated as the ratio of active_users on day N to active_users on Day-0 (install day).

Common Cohort Analysis Mistakes

The first mistake is working with small cohorts. If a cohort contains fewer than 100 users, each churn or return of one person significantly distorts the percentage. The minimum cohort size for reliable conclusions is 500 users.

The second mistake is ignoring seasonality. A cohort that installed the app before New Year will behave differently from a cohort in mid-January. Always compare cohorts from the same season or use data normalization. According to Mixpanel (2024), seasonal fluctuations can reach 30–40% for Day-1 Retention.

The third mistake is confusing cohort and cross-sectional analysis. Cohort analysis tracks one group over time, not comparing different groups at a single point in time. Use repeated measurements on the same cohort, otherwise you will get false conclusions about Retention growth.