Nimplo · Affiliate fraud review for online casinos

affiliate_id	first_dep	total_dep	cpa_paid
A-102	35	410	45
A-114	20	20	45
A-114	20	20	45
A-139	50	1,240	45
A-114	20	25	45
A-156	25	180	45

Example report

Example data

The 4 flagged sources cost 21,195, and their players deposited 2,070 after the first deposit. The 5 clean sources cost 16,235 and brought 28,690.

Example report, example data. 4 of 9 sources are flagged.
Source	Why flagged	Players	Paid	Later deposits
A-102CPA	Not flagged	64	2,880	6,410
A-114CPA	Deposit profile	212	9,540	1,120
INF-07RevShare	Not flagged	121	4,120	9,870
A-156Hybrid	Not flagged	47	2,640	3,050
A-207CPA	Account linkage	38	1,710	240
A-240RevShare	Not flagged	89	3,310	5,220
INF-12Flat fee	Known pattern	31	1,395	410
A-311CPA	Retention profile	190	8,550	300
A-388CPA	Not flagged	73	3,285	4,140

Sources that start with INF are influencer codes.

Later deposits are what a source's players deposited after their first deposit.

Schemes we trace

A CPA deal pays a source once a new player makes a qualifying deposit. A revenue share deal pays it a cut of what the player leaves behind. Each scheme below turns one of those payouts into profit with players who will never be worth it. Every scheme leaves its own trace, so the review runs a separate test for each. A source that evades one test still faces the others.

Deposits with no player behind them

A source collects CPA on first deposits that no real player stands behind.

How we trace it

Every source is compared with what normal traffic looks like in its market and payment method. Statistical tests and gradient-boosted models score it separately, and the thresholds are set to a fixed false-positive rate. A source is flagged only when they agree.

Deviation from expected, by sourceIllustrative

Each dot is a source. A dot far outside the shaded range is a source whose players do not behave as the model expects.

Accounts that are not independent

A source collects CPA on accounts that look separate and are not.

How we trace it

Accounts are matched into single entities with probabilistic record linkage, mapped as a graph per source and split into clusters with community detection. A trained classifier scores each cluster, and an analyst checks it before it reaches the report.

Accounts as a graphIllustrative

Most accounts stand apart. Linked accounts form a cluster.

And dozens more

These two are the schemes we can describe here. Sources use dozens of methods to collect CPA, revenue share or hybrid payouts they have not earned, and the review tests for each one your data can show. Some only appear across casinos, so we are building cross-casino matching for casinos that agree to it in writing.

Each dot is a deposit, grouped into rows by pattern. Highlighted rows are patterns normal traffic does not produce.Illustrative

Where your budget pays off

The same review ranks every source by what its players deposited after the first deposit, for each 1 you paid the source, whether as CPA, revenue share or both.

Put more budget behind the sources at the top. Renegotiate the terms with the ones at the bottom, or end the deal.

If you want ongoing monitoring, the report arrives as often as you choose, for example every week or every month, and shows which sources are worth keeping. The first review is one report.

Later deposits for each 1 paidExample data

INF-072.40
A-1022.23
A-2401.58
A-3881.26
A-1561.16
INF-120.29
A-2070.14
A-1140.12
A-3110.04

Ranked from the example report above. Sources to the right of the line brought more in later deposits than you paid.

How it works

What you send

Send as much as you can. Ideally that is every deposit from your affiliate and influencer sources in deposits.csv and the bets those players made in their first 30 days in bets.csv. The bets file is optional.

Replace player IDs and every other identifier with hashes first, and use the same hash in both files. No names, emails, phone numbers, card numbers or ID documents. The exact column list comes with the data processing agreement.

What you get back

A written report within two weeks of receiving usable files. For each suspect source it shows the players and deposits behind the flag, with what you paid the source on them. It also ranks every source by later deposits for each 1 you paid.

Terms

No fee if the review finds nothing. If it finds suspect sources, you pay a fee we agree in writing before the review starts.

We sign a data processing agreement before you send any data. We need no access to your systems, and we delete the files when the review ends unless you ask us to keep them.

Review my affiliates

We are computer engineers in Lisbon, Portugal. Tell us which markets you operate in, which deal types you use (CPA, revenue share, hybrid or flat fee) and roughly how many affiliate and influencer sources you pay.

hello@nimplo.com

Nimplo finds the affiliates and influencers who get paid for players that are not real.