Suspicion is not proof
The partner recruitment, compliance and fraud guide sets out the patterns a programme manager learns to recognise: cookie stuffing, click spam, self-referral, fake leads, orders placed to trigger commission and then cancelled, trademark abuse and code hijacking, each with the signals that give it away. It also sets the principle this tutorial puts into practice: investigate before accusing, ask the partner to explain, withhold commission on suspect conversions as the terms allow, remove partners whose explanation does not hold, and document the decision.
The signal itself rarely proves anything. A conversion rate (CVRShare of visits or clicks that convert.) far above similar partners can mean stuffed cookies, but it can also mean a partner has started sending people who were already on their way to buy, or that a tracking change is now double counting. A rise in cost per action (CPACost of each conversion.) can come from a legitimate new placement. The work of an investigation is to separate those explanations with evidence, so that the decision stands up if the partner or the network challenges it.
Fraud in this sense is close to what advertising calls invalid traffic (IVTAd impressions and clicks not caused by genuine human interest: bots, fraud and accidents.): activity no interested person caused. The difference in an affiliate programme is that the business usually has more evidence than an ad buyer does, because every payable conversion should end in an order or a lead it can inspect. The data quality guide covers the anomaly checks that usually raise the first question; this tutorial starts where an anomaly has already been noticed.
Before you start
- The programme terms, current version. You need what they say about prohibited methods, validation grounds, commission under review, the partner's duty to explain, and removal. If the terms are silent on holding commission, note it now; the case will be harder to decide and the gap should be fixed afterwards.
- Access to the platform's conversion records. Click times, conversion times, the partner's sub-identifiers, the referring pages the platform records, and the order or lead reference carried on each conversion.
- Access to the order or lead system. Order status, cancellations and returns, payment outcome, and for leads, whether sales reached the person and what happened next.
- A comparison group. Two or three partners of the same type, with similar audiences, whose figures you trust.
- A case record. One document or ticket per case, opened at the first signal, with dates, evidence, messages and the decision. The tutorial on reviewing a marketing error uses the same discipline of facts first, then a timeline.
The steps
Confirm the signal is not a tracking or reporting change
Before suspecting the partner, check your own side. Has the site, checkout, tag set-up or postback changed? Did a deduplication or validation rule change? Is the partner newly featured in an email or a placement the brand paid for? Recent days in any report also fill in late because of conversion lagThe time between an ad interaction and the conversion it leads to, which leaves the most recent reporting days incomplete., so compare like with like. The tracking guide explains how duplicates arise when both a pixel and a postback fire. If a change on your side explains the pattern, fix it, close the case and tell the partner nothing is wrong with their account.
Hold the affected commission under the terms
If the signal survives, hold commission on the conversions in question before the platform's approval date passes, using whatever the terms say about commission under review. Hold the suspect conversions, not every conversion the partner has, so that a genuine partner is not starved of cash during the review. Record the date of the hold, the conversions it covers and the clause it relies on.
Scope the conversions in question
Define exactly which conversions the case is about: the partner, the date range, and any sub-identifier, placement or code that the pattern clusters on. A narrow scope makes the evidence sharper and the decision fairer. If the pattern runs across several partners at once, look for a shared cause, such as a browser extensionA partner whose software sits in the shopper's browser and pops up at checkout with codes or rewards, often taking the last click on an order already decided. or a sub-networkAn affiliate partner that promotes the business's offers through many smaller publishers it aggregates, so less is visible about who actually promotes it., before treating each partner separately.
Build the evidence from timing, sources and devices
For each conversion in scope, pull the time from click to conversion, the page or source the click came from, and whatever the platform records about the device or browser. Look for the shapes the recruitment guide's table describes: conversions seconds after a click, clicks from pages with no visible link, the same devices or addresses repeating, clicks arriving in volumes no real placement could produce. Write down what you see in plain words, with counts and dates, rather than conclusions.
Check what happened to the orders or leads
Join the conversions to the order or lead system by reference. For sales: were they paid, cancelled, returned, or placed by the partner's own staff or address? A high return rateShare of orders or units sold that customers sent back. after commission is approved is one of the patterns the recruitment guide lists. For leads: could sales reach the person, were the details real, did any become customers? The guide to lead generation through affiliates explains why a lead that never answers is not the same as a lead that was simply slow to buy, and why the same buyer should not be paid twice.
Compare the partner with similar partners
Put the partner's figures beside the comparison group over the same period: conversion rate, time from click to conversion, cancellation and return rates, the share of new customers. The conversion rate calculator keeps the arithmetic consistent. Read small counts with care, as the guide to reading small numbers advises: a partner with a handful of conversions can look extreme by chance. A difference that is large, sustained and matched by odd outcomes is evidence; a difference on its own is a question.
Ask the partner to explain, with a deadline
Write to the partner. Describe the pattern, not the accusation: what you saw, over which dates, and why it does not look like the comparison group. Ask for an explanation and any records that support it, such as the pages, campaigns or traffic sources involved, and give a date for the reply. Keep the message factual, because it may be read later by the network or by advisers. A partner who has bought traffic from a third party may not know what that traffic was; their answer still matters.
Decide: clear, adjust, reject or remove
Weigh the partner's answer against the evidence. Use the five whys to push past the first explanation to the cause. Four outcomes cover most cases. Clear: the pattern has an honest explanation; release the held commission and say so. Adjust: some conversions are invalid, such as duplicates or prohibited methods, and the rest stand; reject only those. Reject: the conversions in scope are not genuine; reject them under the terms. Remove: the explanation does not hold or never came; reject the conversions and end the partnership. Base the decision on the terms and the evidence in the case record, not on how much money is involved.
Tell the partner and the network
Tell the partner the decision in writing, with the clause it relies on and what happens to the held commission. If the programme runs on a network, share the case with the network: the recruitment guide suggests reporting serious cases because the network may act across other advertisers too. Questions about money already paid, or anything that might go beyond the programme, go to the business's own advisers.
Record the case and close the gap that let it happen
Finish the case record: the signal, the scope, the evidence, the partner's answer, the decision and the dates. Then ask what would have caught it sooner or prevented it. That might be a validation rule, a check in the weekly review in the programme management hub, a clause in the terms, or a change to the commission structure that stopped rewarding the behaviour. A library of past cases makes the next investigation faster and the decisions more consistent.
Evidence by fraud pattern
The recruitment guide's table names the patterns and the signals that raise suspicion. The table below adds what an investigation looks for to confirm each one, and what would clear the partner.
| Pattern | Evidence that confirms | Evidence that clears |
|---|---|---|
| Cookie stuffing | Conversions seconds after clicks; clicks from pages with no visible link; clicks far exceeding the partner's real audience | Clicks traceable to visible links on real pages, with normal time to conversion |
| Click spam | Very many clicks with almost no engagement, conversions spread thinly at random | Clicks concentrated on placements the partner can show |
| Self-referral | Buyers sharing the partner's name, address, payment details or devices | Buyers unconnected to the partner |
| Fake leads | Details that do not reach a person; duplicates; leads that never answer across many attempts | Leads that answer, even if they do not buy |
| Order and return | Orders cancelled or returned soon after commission approval, clustered on one partner | Return rates in line with similar partners and products |
| Duplicate tracking | Two conversions for one order reference | One conversion per order once a tag issue is fixed (a tracking fault, not fraud) |
| Code hijacking | Orders using codes issued to someone else, through this partner's links | Codes the partner was issued |
Where the evidence is mixed, adjust rather than reject everything: pay for the conversions that hold up and reject the rest, and say which is which.
A tutoring service investigates a spike in sign-ups (illustrative)
Huxtavy Tutoring and a partner's sudden bookings
Huxtavy Tutoring, an invented online tutoring service, pays partners £15 for each free trial lesson booked. A study skills blog that usually brings about 40 bookings a month brings 410 in March. Huxtavy first checks its own side: no site, tag or rule changes, and the blog was not in any paid feature. It holds commission on the 410 bookings, worth £6,150, under the clause in its terms on commission under review.
The evidence is lopsided. 362 of the bookings came from 9 devices, most within 20 seconds of the click, and used pupil names that do not match the parents' email addresses. Only 11 of the 410 trial lessons were attended, against about two in three for its other study-skills partners. The comparison partners show no change in March.
Huxtavy writes to the blog, describes the pattern and asks for an explanation within 10 working days. No reply arrives. It rejects 399 bookings, pays for the 11 attended lessons, removes the partner under its terms, shares the case with its network and adds a check: any partner whose weekly bookings rise more than threefold is reviewed before approval.
Common mistakes
- Accusing before checking your own tracking. A tag change or a duplicated postback looks exactly like a partner inflating conversions, and the partner will rightly object.
- Holding the partner's whole balance. Hold the conversions in scope; starving a partner of all commission during a review harms partners whose case turns out to be clear.
- Letting the approval date pass. Commission approved by default before the review finishes is much harder to recover than commission never approved.
- Deciding on the size of the spike alone. A large number is a question; the outcomes of the orders or leads are the evidence.
- Vague messages to the partner. A message that says "suspicious activity" gives the partner nothing to answer and gives you nothing to rely on later.
- Rejecting everything when some conversions are genuine. Adjust where the evidence is mixed, and say which conversions stand.
- Closing the case without closing the gap. If the same pattern could happen again tomorrow, the investigation is not finished.
