Reviews
Part of 6 things to check in affiliate platforms
Affiliate platforms mistakes: week one, month three, year two, exit
Affiliate platforms mistakes sorted by when they surface: week one, month three, year two and exit — and the cheap written rule that prevents the late ones.
Affiliate platforms mistakes sort better by when they surface than by severity, because severity is only known afterward. Timing tells you how long you have to make the decision that prevents each one. A mistake that shows up in week one is a nuisance; the same class of mistake surfacing at exit can cost you a client. Nothing below requires anyone to be careless — every item is a reasonable choice that ages badly. Every setting named here lives in your own contract and your own account, and both are allowed to change; the answer that counts is the one written down. The same schedule holds whether the program recruits publishers, creators or influencer partners, because the tracking, consent and reporting machinery underneath is identical.
What to take away
- The mistakes that surface late are the ones decided earliest, usually at signature.
- A default you did not choose is still a decision you made.
- The cheapest correction in this category is a written rule agreed with another team before either side reports a number.
- Week one, month three and year two are not deadlines. They are the points at which the consequence of a decision becomes visible.
Week one: mistakes that announce themselves
These are the good kind. They hurt immediately, so they get fixed. Each one sits on the tracking chain that how partner programs actually work sets out, and the same errors turn up whichever platform runs the program — Awin, Impact, CJ, Rakuten Advertising, ShareASale, PartnerStack, Refersion. The names on the invoice change; the settings behind them are the same short list: an approval switch, a cookie window, an attribution rule, a commission rate.
Here are the five that surface in the first week, in the order they usually appear.
- The confirmation page is not tracked, so orders land as clicks and never as sales.
- The cookie window and the attribution window were left at whatever the platform set, and nobody chose them. The default a platform ships is not a recommendation.
- Partners are auto-approved, so coupon and cashback sites are in before anyone reads them.
- What the report shows as the commission rate is not the rate in the contract, because an override was applied at setup and never removed.
- Nobody reviews who still holds a login after onboarding week.
The fix for each is small enough to do in the same week. Place a test order end to end and confirm it lands as a sale rather than a click. Open the tracking settings, choose the cookie and attribution windows you actually mean, and write both down with the date. Turn off auto-approval and read the applications instead. Reconcile the rate the report shows against the rate in the signed contract. Pull the login list and remove the accounts that no longer do work.
Where does the value you are running on live? In two places, and they are allowed to disagree: the platform's tracking settings and the contract. Read both, choose the window you actually mean, and write the choice down with its date. California's CCPA and CPRA put the opt-out in the consumer's hands, so a window nobody chose is a liability before it is an accounting problem.
Week One Mistakes
- Launching before order paths tested
- Recruiting partners before rules exist
- Turning on every partner type at once
- Validation window from template, not returns
- Whole team can change commission rates
The fifth one is worth a minute. Access granted for speed is never reviewed later, and a commission rate is a payment instruction. Role-based access control is unglamorous and it is the difference between an audit trail that answers a question and one that records that somebody logged in.
Month three: mistakes that surface as a disagreement
By month three the program has produced enough data for two people to disagree about it.
Month Three Disagreements
What happens
- Same sale reported twice
- No deduplication rule
- Partner numbers too good
- Click after buying decision
- Partner chases pending money
- Pending and approved look same
- Client asks what program added
- Attributed reported as incremental
- Reports never match
- Day boundaries and currency differ
What it actually is
- Same sale reported twice
- One written rule
- Partner numbers too good
- Policy on extension partners
- Partner chases pending money
- Visible validation state
- Client asks what program added
- Say which in first report
- Reports never match
- Fix booking convention once
Preventing decision
- Same sale reported twice
- Partner numbers too good
- Partner chases pending money
- Client asks what program added
- Reports never match
Four of those disagreements arrive the same way:
- A routing figure read as growtha partner credited for a sale they did not influence.
- A click count set beside an order count as if they were one unit of measure.
- Two dashboards on different date ranges, one in local time and one in UTC.
- The deduplication rule never chosen, so credit on an order two partners claim goes to whichever rule the platform applies — last click, first click, last touch, or a split between them.
Each of these has a fix that lives in a report definition. Agree what a number is a count of before anyone quotes it. Pick one time zone for every export, say which one, and keep it. Write down the deduplication rule you intend, record which rule the platform applies, and put both in the partner terms, so nobody argues about credit later.
Nobody minds a routing figure described as a routing figure. What causes trouble is a figure allowed to be read as growth for two quarters and then explained. Every reporting stack carries a time-zone setting — Google Analytics 4 sets one at property level, the platform sets its own, and a spreadsheet inherits the laptop's — so two dashboards can disagree about the same day without either being wrong. Where the honest answer needs a test rather than a dashboard, run it: export the same day from the platform and from Google Analytics 4, put the two totals side by side, and settle the difference before anyone quotes a rate. That test belongs with the rest of your measurement tooling.
Year two: mistakes that surface as a slow cost
Nothing breaks. Things just get heavier.
The slow costs all have the same shape: something that was being maintained quietly stopped being maintained. Three arrive this way.
- The partner list grows and nobody prunes it, so a long tail of inactive partners sits in every report and slowly stops meaning anything.
- The rules were enforced once at the start and never since, so partners have quietly drifted back to whatever works for them.
- The person who set the program up has moved on, taking with them the reasoning behind three settings nobody now dares change.
That last one has a name: a genuine operating risk, not a personnel complaint. A program whose configuration lives in one person's memory has a bus factor of one.
The cure is a page of written decisions, not a training session.
Write down the rule, the reason and the date. The reason is the part that saves the next person, because a rule with no reason gets either obeyed forever or deleted on a whim.
At exit: mistakes with no remedy left
Three of these, and all three were decided at signature. Exit terms are the part of a vendor comparison worth doing while you still have a choice of vendors, because they are the terms nobody renegotiates later.
Exit Mistakes With No Remedy
- Tracking domain in somebody else's name
- Export nobody ever ran
- Notice terms nobody read
- Whose name is on the account
- Who keeps the partner list
The tracking domain in somebody else's name. Every link ever published keeps working, and it now works for them. This cannot be fixed afterward at any price, because the links are out in the world. The consent duties ride with the domain rather than with the contract: the FTC Act reaches a misleading claim about who runs the track, and the opt-out rights in the CCPA and CPRA follow the data, not the registrant field. Write down the registrant, the operator and who answers a complaint.
An export nobody ever ran. You find out what the file contains during the notice period, which is the worst possible moment to discover it holds monthly totals and not records. Run it during the trial and open it. What has to come out of it is the transaction line — the date, the order value, the status and the partner credited — plus each partner's contact details and payment history, in a format you can open in a spreadsheet without the platform.
Notice terms nobody read. Four things come out of the notice clause: how much notice you owe, whether tracking keeps running until it expires, whether partners are told and by whom, and who pays commission on sales that validate after the end date. Each has a default in the standard agreement you accepted when the account was opened — the terms of service you click through at signup, on whichever network runs the program — and none of those defaults was written for you or renegotiated after signature. The fix is before signature: read the clause and write the four answers into the deal, in days, next to the termination date.
For an agency, exit has a second half, because the relationship that ends may be with the client rather than the vendor. Whose name is on the account and who keeps the partner list are questions with a right answer at the start of a retainer and no answer at the end of one.
The mistake that is not a mistake
Running a small program on manual tracking and a spreadsheet, with a handful of direct partners, is often correct. So is refusing to admit a partner type that would technically increase revenue. So is declining to instrument an order path that carries almost no volume, as long as somebody wrote down that the decision was made.
The test is whether the exception has a name against it and a date. An exception with an owner is a decision. An exception nobody recorded becomes a mystery, and mysteries get resolved by whoever inherits them, usually in the direction of expensive caution. The partner side of the same relationship is treated under creator marketplaces.
Common questions
We inherited a program with several of these. Which do we fix first?
The domain registrant, because it is the only one that gets permanently worse. Then the export, then the deduplication rule.
Is it worth pruning inactive partners?
Yes, but archive rather than delete, and keep the history. A partner list that reflects who is actually working is worth having; a deleted record is a question you cannot answer later.
How often should the rules be re-enforced?
Enough that partners expect it. A quiet quarterly check that occasionally produces a message is more effective than an annual purge, and much less damaging to the relationships you want to keep.
Our client wants the program moved in house. What breaks?
Usually the contact list and the payment history, because those were the parts held in the platform rather than by you. Ask what the export contains before you agree to a date.
