Paying for players who never play.

Fictional scenarios and synthetic data. Reviewed 6 November 2026.

Meridian FX pays one affiliate on a cost-per-acquisition basis: a fixed sum for each referred client who makes a first deposit above a qualifying minimum. In October the affiliate sent 200 first-time depositors, the best month in the programme. The commission went out on the first of November. By the middle of November the marketing lead has noticed that almost none of the 200 have traded, and asks Ines at Northbank Pay whether the deposits looked normal.

What is actually being decided.

The decision is whether to keep paying this affiliate on these terms, whether to claw back October, and, underneath both, whether the 200 are clients at all. The operator carries the commission already paid and the cost of the next month if nothing changes. The affiliate carries a contract and a reputation; a wrongful clawback ends both. The PSP carries 200 deposits and, a disproportionate share of the disputes that follow. The 200, if they are people, carry the ordinary friction of a broker that has suddenly become suspicious of its newest clients.

CPA and revenue share pay for different things.

Write the commission as a sentence. "Pay 150 when a referred account makes a first deposit of at least 20." Every term is a choice. The event is the first deposit; the horizon is zero days after it. Under cost-per-acquisition the affiliate's interest ends at the moment the operator's interest begins. Under revenue share, where the affiliate takes a percentage of net revenue from the referred client over months or years, the two interests run together for as long as the client does. Hybrids split the difference and most programmes use them.

Affiliate traffic quality asks whether acquired customers deliver the relationship a commission was priced for. Poor performance alone is not fraud; that conclusion needs evidence of deception or misuse against the relevant terms. It need not involve a fabricated person. Incentivised sign-ups, where the affiliate's audience is paid or rewarded to open an account and deposit the minimum, are made up of real people who were never going to be clients. Fabricated sign-ups are the other end of the range, and OWASP's catalogue of automated threats lists account creation (OAT-019) as a threat because so many commissions, like so many bonuses, are paid per account. We do not describe how either kind of traffic is produced; the defender's task is to tell the two apart from the operator's side, after the fact.

Read the cohort, not the month.

The CPA contract defines the metrics by which the affiliate looks good: sign-ups, first-time depositors, conversion from one to the other. All three can be excellent in a month that delivers nothing. The measure that cannot be gamed at the moment of payment is what the cohort does afterwards, and that measure only exists if someone builds it.

A synthetic illustration. Take the 200 October first-time depositors as a cohort and compare with the broker's other affiliates' October cohorts, which is the baseline. At 30 days, 4 of the 200 have made a second deposit; the baseline is 38 in 100. Nine have placed a trade; the baseline is 71 in 100. The average first deposit is 20.00, the qualifying minimum, against a baseline of 85. One hundred and fifty of the 200 requested a withdrawal of their full balance within ten days, each to the card that deposited. The operator paid 150 in commission for each 20 that went out again by the same door.

Those are observations. The inference, that the traffic was incentivised or fabricated, is strong. The alternatives we keep alive are specific and should be tested: the affiliate markets to a price-sensitive audience that always deposits the minimum; the broker's own onboarding is losing clients at the first trade; a promotion in October required only a minimum deposit and attracted bargain hunters everywhere. The baseline answers the third. The first two are answered by looking at the affiliate's September cohort, which, if it behaved like the baseline, makes October the anomaly. The article on baselines and seasonality covers the discipline of choosing the comparison, and dispute lag and cohorts explains why the disputes from October have not arrived yet.

What the PSP sees that the operator does not.

The operator's affiliate platform knows the referral code and the sign-up date. The PSP's export knows the money. Ines filters Northbank's October deposits for Meridian FX by the merchant's sub-reference and finds, for the 200: cards from two issuers in one country, where the affiliate's stated market is three countries away; 140 of the 200 first deposit attempts returned a do-not-honour or insufficient-funds code before a second card succeeded; IPs concentrated in 198.51.100.0/24; and eleven disputes already, all raised within three weeks of the deposit, which is early. The operator has none of this unless it asks.

Each line is an observation. Together they shift our reading from "incentivised audience" towards "small number of actors, many cards", which matters because the two call for different conversations with the affiliate and different clawback evidence. The deposit-then-withdraw pattern itself, and what a closed-loop payout rule does to it, is the subject of deposits, withdrawals and cash-out.

A failure mode to watch.

Measuring affiliates on the metrics the contract defines and nothing else, and then, when a problem surfaces, cutting the affiliate rather than the cohort. The first means the problem is noticed by someone in finance months later. The second means the operator loses a channel on a suspicion, pays for it in the next quarter's acquisition, and never learns whether October was one bad batch bought from a sub-affiliate or the affiliate's standard product. The Volume 01 lesson time and outcomes is the frame: the outcome of an acquisition is only known at a horizon, and the commission was paid before the horizon.

A way to work it.

1. Write each affiliate's commission as an event with a horizon. Note which metrics the affiliate can move without sending a real client.
2. Build the monthly cohort for every affiliate: second-deposit rate, trade or play rate, average first deposit, full-withdrawal rate inside ten days, all at 30, 60 and 90 days. Compare with the portfolio baseline and the affiliate's own prior months.
3. For an outlying cohort, ask the PSP for the card references, issuer countries, first-attempt error codes, IP concentration and disputes to date.
4. Write the observations in one column and the inference in another. Name the innocent explanation and say which observation rules it out.
5. Act by cohort: withhold or claw back October, hold commission on November until the 30-day numbers exist, keep the affiliate live. Name an owner and a review date.
6. Share the cohort view with the affiliate. A good affiliate with a bad sub-source wants to know; a bad one will tell you by how they respond.

Take it to the desk.

Which of your affiliate metrics could an affiliate improve without sending you a single additional real player? Try it: Affiliate traffic, and the Volume 01 case Promo abuse for the retail cousin. Look up: affiliate fraud and cohort.

Edited 6 November 2026. Primary references are listed above; fictional scenarios are not claims about Antifrog customers or operating experience.

Fictional learning material. Any numerical examples are synthetic. Use only authorised, appropriately minimised records for workplace exercises; do not upload customer information to this learning site.