Cohort Analysis for Affiliate Channels: An Operator Deep Dive 2026

Operators that evaluate affiliate channels on last-30-day revenue make two systematic errors. First, they over-invest in channels with high front-loaded revenue (mass-market iGaming acquisition) that decay rapidly. Second, they under-invest in channels with slow ramp but long tail (forex IB networks, prop trading challenge cohorts) because the first 90 days look weak. Cohort analysis is the corrective: by grouping affiliates by acquisition period and tracking their behavior over time, operators see the true channel economics rather than blended-period averages. This deep dive defines cohorts for affiliate channels, walks the retention-curve math, and shows the vertical-specific decay patterns that change every deal-economics decision.

Definitions and inputs

A cohort is a group of entities (affiliates, players, or accounts) that share an acquisition event in a defined time window, typically a calendar month. The minimum inputs for an affiliate-channel cohort analysis are: an acquisition timestamp per entity, a recurring activity record (revenue, deposits, trading volume), and a stable identifier that ties the entity to its source affiliate or traffic source. Operators that lack reliable acquisition timestamps cannot run cohort analysis, full stop. Before any modeling, audit the data: are first-time depositor timestamps captured? Is affiliate attribution stamped at signup or at first conversion? Do reactivated dormant players get a new acquisition timestamp or retain the original?

• Cohort entity: usually the player (in iGaming, sportsbook, sweepstakes) or the trader (in forex, prop trading). For affiliate-channel rollups, also cohort by affiliate.
• Cohort period: month is the standard. Weekly cohorts add noise for affiliate channels. Quarterly cohorts mask intra-quarter shifts.
• Activity metric: net revenue (NGR for iGaming, net spread for forex, net challenge revenue for prop). Avoid gross deposits as they overstate value.
• Cohort age: 'M0' (acquisition month), 'M1' (one month later), M2, M3, and so on. Report through M24 at minimum for any cross-vertical analysis.
• Retention denominator: count of entities active in the cohort during the reporting month, divided by count of entities in the original cohort.
• Cohort slice: by affiliate, by sub-affiliate, by traffic type (paid, organic, content, paid-social), by geo, by device. Multiple slices needed.

Step-by-step calculation methodology

The standard calculation builds a cohort matrix where rows are acquisition months and columns are cohort ages. Each cell holds either revenue or retention rate for that cohort at that age. Below is a worked example for an iGaming operator across three monthly cohorts (Jan, Feb, Mar 2026), tracked through M5. Revenue is NGR per player. Player counts: Jan cohort 1,200 acquisitions; Feb cohort 1,100; Mar cohort 1,400.

Three observations from the table above. First, the Jan cohort produces $224 cumulative NGR per player through month 5 and is still earning $14 per player in M5, which projects to roughly $260 to $290 LTV through M12 if the decay continues at observed rate (typical 3 to 5 percent month-over-month tail). Second, the Feb cohort is outperforming the Jan cohort at every comparable age (M0 $82 vs $78, M1 $54 vs $52). Third, the Mar cohort is underperforming both, which is a leading indicator that something changed in March acquisition (a campaign mix change, a regulatory restriction, an affiliate-traffic-quality drop). Without the cohort matrix, blended monthly revenue would mask this signal: total revenue might still be growing if the Mar cohort is larger.

Cohort LTV is the sum of activity across all observed ages, optionally extrapolated. A conservative LTV estimate uses only observed values and treats unobserved months as zero. A modeled LTV estimate fits a decay function (exponential or power-law) to the observed values and projects forward. Operators should report both numbers: 'Jan cohort observed LTV at M5: $224. Projected LTV at M24 with 4 percent month-over-month tail decay: $312.'

Vertical variations: iGaming, forex, prop trading

Affiliate-channel cohorts behave very differently across verticals. The same retention-curve template will lead to wrong investment decisions if applied uniformly. The table below summarizes the three dominant vertical patterns. iGaming front-loads aggressively, forex IBs back-load with long tails, and prop trading produces a step-function pattern driven by challenge resets.

Practical consequence: a forex IB cohort that produces $40 NGR in M0 is not a weak cohort; it is a normal early-stage IB ramp. Cutting that affiliate at the 90-day review because the iGaming acquisition desk benchmarks $200+ in M0 is a strategic error. The platform's affiliate-channel attribution must support vertical-specific cohort dashboards or the analytics team will continue to make these errors silently.

Cohort-by-cohort worked example: a forex IB program over 12 months

The example below illustrates the back-loaded forex IB pattern. Three cohorts of newly recruited IBs (Q1, Q2, Q3 2025) are tracked through M12. Revenue is net spread share allocated to the IB-attributed sub-trader pool, in USD per IB. IB counts: Q1 cohort 18 IBs; Q2 cohort 22; Q3 cohort 24.

The Q1 cohort generated only $220 in the first three months but $4,320 cumulative through M12 (a 19x multiple on the M0 to M2 base). Any operator that judged this cohort on M0 to M2 performance would have cut budget at exactly the wrong moment. The Q2 cohort is on a similar trajectory ($3,220 by M9 vs Q1 at $2,860 by M9, slightly ahead). The Q3 cohort is too young to extrapolate confidently but tracks within the historical band. The decision question is no longer 'is the IB program profitable in month 1' but 'are recent cohorts trending up or down at comparable ages'. The answer changes the budgeting conversation entirely.

Common mistakes operators make

• Mistake 1: Comparing absolute revenue across cohorts of different sizes. Always normalize to revenue per acquired entity (per player, per IB, per trader).
• Mistake 2: Mixing reactivation revenue into the original cohort. If a player who churned in M3 returns in M8, that revenue belongs to a reactivation cohort, not the original M0.
• Mistake 3: Using gross deposits instead of net revenue. Deposits include bonus churn, wagering-requirement loops, and chargebacks. NGR or net spread reflects what actually reached the P&L.
• Mistake 4: Treating M0 partial-month acquisitions as full months. A player acquired on day 28 of the month has 2 days of M0. Either normalize by days-in-period or define M0 as the first full month after acquisition.
• Mistake 5: Ignoring vertical decay differences. Applying an iGaming cohort template to a forex IB program will lead to under-investment in IB channels because the early months look weak.
• Mistake 6: Not slicing by traffic type. A cohort blend of paid-social plus organic plus IB sub-network masks the fact that paid-social cohorts decay faster than IB sub-network cohorts.
• Mistake 7: Reporting only LTV without confidence intervals. Cohort LTV projected to M24 has wide uncertainty bands. Report median plus 25th and 75th percentile.

Benchmarks and what good looks like

The benchmarks below are typical operator ranges for mature affiliate channels. They are not absolute targets; cohort quality depends heavily on geo mix, traffic source composition, vertical, and product maturity. Use them as triangulation against your own cohort data.

Top-quartile cohort multiples (the final column) deserve special attention. A 3x to 5x gap between top-quartile and median forex IB cohorts means a small number of IB networks produce most of the program value. Cohort analysis identifies those networks early, so the operator can invest in support, training, and co-marketing rather than spreading account-management attention evenly across the IB roster.

Audit and implementation playbook

1. Audit acquisition timestamps: confirm every player/trader/IB has a stable acquisition timestamp that survives system changes. If not, this must be fixed before any cohort work.
2. Define cohort entity per vertical: player for iGaming and sportsbook, trader for forex retail, IB for forex IB programs, trader for prop trading. Document the decision.
3. Set the activity metric: NGR for iGaming, net spread for forex, net challenge revenue for prop. Avoid gross deposits and gross volume.
4. Decide cohort period: monthly is standard. Weekly only if acquisition volume is large enough (over 500 per week) to be stable.
5. Build the cohort matrix: rows are acquisition months, columns are M0 to M24, cells are revenue per entity. Repeat for retention rate.
6. Slice by affiliate, traffic type, and geo: aggregate first for stability, then drill into top-10 affiliates and top-3 traffic-type cohorts.
7. Fit decay curves: power-law (revenue declines proportionally to age) for iGaming and prop; logistic or back-loaded gamma for forex IB. Document the fit and back-test.
8. Project LTV through M24 with confidence intervals: report median plus 25th and 75th percentile. Avoid single-point estimates for board reporting.
9. Wire cohort dashboards into platform reporting: live dashboards beat static quarterly decks. Heads of affiliates and finance should see updated cohorts weekly.
10. Re-fit every quarter: cohort behavior drifts with regulatory changes, traffic mix shifts, and product changes. Static cohort assumptions decay in accuracy within 3 to 4 quarters.

Frequently asked questions

External references

• Harvard Business Review: cohort analysis methodology for customer economics.
• HubSpot Research: customer acquisition and retention benchmarks across B2B and B2C channels.
• Deloitte iGaming industry analysis: vertical-specific player lifecycle data.
• OpenView Partners: SaaS retention and LTV cohort studies usable as cross-vertical baselines.
• ESMA investor trading behavior reports: account lifecycle data for forex and CFD cohorts.
• H2 Gambling Capital: player-lifecycle industry data for iGaming cohort benchmarking.

Cohort analysis converts blended-period revenue into a structured view of channel health over time. For affiliate operators, it is the difference between cutting a forex IB program 90 days early and recognizing that it is on the same trajectory as last year's top-quartile cohort. The math is not difficult. The discipline is in defining the cohort cleanly, slicing the data correctly, and refreshing the model often enough that decisions are based on the current channel reality rather than a 14-month-old assumption.