Retention Cohort Analysis for iGaming: A Practical Guide

Fix the definitions

Write down three things and keep them stable.

1. Anchor: first deposit date in one agreed time zone.
2. Activity: a real-money bet or deposit. A login includes players who look and leave. Decide how bonus-only play counts.
3. Window: daily buckets for the first two weeks and weekly buckets after that.

Then choose between N-day and unbounded retention. Amplitude explains that N-day retention counts users who return on one exact day. Unbounded retention counts users who return on that day or any later day, so it is equal to or higher. Amplitude recommends N-day for products used every day and unbounded for products where infrequent use is normal. Many sportsbook players follow events and do not play every week. An unbounded or weekly "active in week N" definition usually fits them better. Label the choice on every chart.

Mixpanel adds two rules. Retention counts users, not events. Incomplete time buckets are flagged because their data still changes.

Pull the data

You need one row per player.

• Player ID, the same one your CRM and PAM use
• FTD date
• Dates or weeks of qualifying activity
• Deposits and NGR by period, if you extend the table to revenue
• Channel, country, first product and bonus flag for slicing

Count players per cohort and age

For each cohort, count the players active in each week after the anchor and divide by the cohort size. Week 0 is the anchor week and equals 100% when activity is the first deposit.

A minimal SQL sketch for weekly FTD cohorts (adapt it to your warehouse dialect):

sql WITH ftd AS ( SELECT player_id, DATE_TRUNC('week', MIN(deposit_ts)) AS cohort_week FROM deposits GROUP BY player_id ), activity AS ( SELECT DISTINCT player_id, DATE_TRUNC('week', bet_ts) AS active_week FROM bets ) SELECT f.cohort_week, DATEDIFF('week', f.cohort_week, a.active_week) AS week_n, COUNT(DISTINCT a.player_id) AS active_players FROM ftd f JOIN activity a ON a.player_id = f.player_id GROUP BY 1, 2 ORDER BY 1, 2;

Divide active_players at each week_n by the week-0 count of the same cohort. A pivot table in Excel or Sheets gives the same result: cohort week in rows, weeks since FTD in columns, distinct players as the value.

Read the table

Swipe the table sideways →

Read it in two directions.

Across a row you see one cohort ageing. The 6 October cohort keeps 41% of players in week 1, then 29%, 21% and 14%. A steep first drop followed by a flatter tail is the common shape.

Down a column you compare cohorts at the same age. Week-1 retention is 41%, 40%, 43% and then 36%. The 27 October cohort is worse at every age you can observe. Something changed for that cohort, and a blended monthly number would not show it.

Then ask what differed in the week of 27 October. Check the acquisition mix (did a weak affiliate batch arrive?), the welcome offer, a payment or KYC issue and competitor events. The table tells you where to look, not the cause.

Cells marked "not yet" have not had time to mature. Reading a partial week as final understates retention. Mixpanel excludes incomplete buckets from its weighted average for the same reason.

Colour the table by column

Apply the colour scale by column, so a week-8 cell is compared with other week-8 cells. Mixpanel shades relative to each cohort row, which shows the shape of each cohort's decay. Column shading shows which cohorts are better or worse at the same age, and that is usually the question a CRM lead asks. Choose one and state which.

Recognise four curve patterns

Plot one line per cohort, with age on the x axis and percent retained on the y axis.

• Steep early drop, flat tail. This is normal. The question is how high the plateau sits.
• A curve that keeps falling with no plateau. The product or journey builds no habit.
• A bump at a specific age. It is often a scheduled event: an expiring second-deposit bonus, a weekly promotion, a league start. Check the CRM calendar.
• Newer cohorts below older ones. Acquisition quality fell or the experience got worse. Check the mix first.

Judge onboarding by the early columns (weeks 1 and 2) and habit by the later columns (week 8 onward). Bonuses can lift early retention that decays later, so look at both.

Extend the table to NGR

Retention shows who is still around. It does not show whether they are worth keeping. Two cohorts can have the same week-4 retention and very different value if one is bonus-driven with low net revenue.

Use cumulative NGR per acquired player by week since FTD. The reason to use NGR and not GGR is in NGR vs GGR.

Swipe the table sideways →

Look at whether each curve still climbs or has flattened. Affiliate A flattens by week 8. It brought players who stopped producing net revenue. That is a payback conversation with the channel, and it stays invisible if you compare only week-1 deposits.

Divide cumulative NGR by acquisition cost to get a payback curve per channel. The campaign economics calculator models cohort NGR, ROMI and payback over 12 months.

Read the table in reverse for churn

A churn cohort analysis is the same table read the other way. Look at the share lost at each age step and find which cohorts lose players earlier. Define churn first: an inactivity threshold of 30 days gives different answers from 90 days. The player churn guide covers the threshold.

Once you find a high-churn cohort, split it by first product, deposit size and bonus use. Check the time from registration to FTD and from FTD to second deposit. Look at support contacts, withdrawal delays and verification friction in the first fortnight. The result is a ranked list of where players leave. It feeds retention work and reactivation campaigns.

Judge CRM campaigns with cohorts

CRM teams ship campaigns constantly. The temptation is to compare the period after with the period before. That mixes the campaign effect with seasonality, acquisition changes and product releases. Use this process instead.

1. Fix the audience first. The audience is defined on a date with a rule and does not change afterwards. Otherwise players who return on their own count as campaign wins.
2. Hold out a control. A random share of the same audience receives nothing, or the previous standard treatment. Gamble Grip starts every engagement with one fixed cohort, one controlled reactivation campaign and one agreed measurement method.
3. Use the same retention definition and window for both groups. Report NGR, not only deposits or bonus redemptions.
4. Read the gap, not the level. Compare treatment minus control at each age, and check whether the gap persists or fades after the incentive ends.
5. Check for cannibalisation. If the campaign pulled forward deposits that would have happened anyway, the later columns of the treated group dip below the control.

When a clean control is impossible, use a baseline from comparable earlier cohorts and call the result an estimate.

The same logic applies to welcome-journey changes. Compare FTD cohorts from before and after, against a holdout where possible. Look at second-deposit rate, week-1 retention and week-8 NGR, not only first-deposit conversion.

Gamble Grip's published campaign results are figures from the treated group. In October 2026 its revenue-share campaigns in Türkiye brought $2,378,075 of deposits and $246,707 of client NGR. The control group or baseline shows how much of that the campaign caused. Results by market are on the campaign results page.