You judge an affiliate source early by looking at revenue per player in that source, not at registrations or first deposits, and by comparing it with what your better sources deliver at the same age. The usual approach is to wait out a hold of about two weeks before deciding. We try to read a source about a week after sign-ups. That is a goal we confirm on each operator's own data, not a guarantee, and the details depend on volume and data quality. Here is the method, and the limits.
What is a hold, and why does it slow you down?
A hold is the waiting window, usually two weeks, before the real value of a traffic source becomes clear. During it you keep paying for traffic you cannot yet judge. If the source is poor, the budget is spent by the time you know.
The cost is not only the payment to the affiliate. It also includes bonus money handed to players who never return, and the manager time spent on contacting them. Cutting a bad source a week earlier saves all three.
Why are registrations and first deposits not enough?
They measure volume. A first-time deposit, or FTD, is the very first deposit a player makes, and it tells you the player arrived, not that they will stay. Two sources can deliver the same number of registrations and the same number of FTDs, and one can bring players who return and the other players who vanish.
That is why I would compare sources on revenue per player and on the next steps in the player's life, such as the second deposit (see how to get a second deposit). Those are the signals that separate value from volume.
What do you compare?
Group players by source and by the week they signed up, so that each group has the same age. Then compare, across sources:
- Revenue per player so far.
- Share of players who made a second deposit.
- Share of players still active after the first week.
- Share of players whose behaviour matches abuse patterns, such as deposit-only play-through (see casino bonus abuse and how to spot it).
Comparing at equal age is the whole trick. A source that signed up yesterday will always look worse than one that signed up three weeks ago. If you compare unequal ages you will cut good sources by mistake.
You can start this tomorrow with a spreadsheet. Export sign-ups with their source tag, add each player's deposits so far, and calculate revenue per player by source for players of the same age. It is rough, and it will already show you which sources deserve a closer look. A forecast adds the step of estimating where each source will be when the hold ends, so you can act before it does.
How can an early verdict work?
By predicting rather than waiting. Sellrise gives a verdict per source about a week after sign-ups: scale, watch or cut. It uses the first days of behaviour to estimate where the source will be when the hold ends, and it shows the revenue per player for every source. The timing varies with volume and data quality, and in a pilot we check the verdict against your own day-14 numbers, which you already have.
That check is the part I care about. A claim about a week is cheap to make. A comparison between the early call and the number you see two weeks later, on your traffic, is evidence.
What can go wrong with an early verdict?
- Low volume. A source that brings a few dozen players per week gives a noisy signal. In that case wait, or accept a wider margin.
- Mixed quality inside a source. One affiliate may run several campaigns. Split them if the data lets you.
- Seasonality. A source that peaks at a weekend or a major event will look different from one running in a quiet week.
- Data quality. If source tags are missing or inconsistent, the verdict will be wrong however good the model is. Fix the tagging first.
I would also treat the verdict as input to a decision, not the decision. A person who knows the affiliate, the contract and the market should confirm a cut.
How does this change what you pay affiliates?
It gives you a basis to negotiate. If you can show an affiliate that their players bring a fraction of what your other sources do, the conversation about quality is based on numbers. If the source is only slightly weak, you can ask for changes in the traffic and watch the next cohort.
Be careful about promising pay changes on the strength of a model alone. The contract terms are yours, and the verdict is a forecast. It is safer to use it to decide where to move budget, and to use the matured hold numbers when the payment is at stake.
How does it relate to fraud?
Closely. A source with a high share of abusive accounts will look poor on revenue per player, and it will also show the patterns that signal abuse. Sellrise flags above-average win and withdrawal rates and deposit-only play-through, and brings them for review before a payout goes out. We do not claim to catch all fraud, and the review is yours.
Where do you start?
Take your last three months of sign-ups. Group them by source and by week. Check how many of the sources that you eventually cut already looked weak a week in. If many did, you have a case for reading sources earlier. The page on traffic quality describes what we offer, the module is priced from $500 as described in how casino CRM is priced, and a pilot tests it on your traffic.