First, rebuild cost of sale by partner type and customer type on validated revenue
Start with the figure finance is reacting to and take it apart. The reporting and incrementality guide defines cost of sale as the programme's total cost as a share of revenue, and insists on three things about how it is built: include every cost, use validated rather than tracked revenue, and calculate it by partner type, by product category and by customer type, because the averages hide where money is lost. Every cost means commission, bonuses, placement fees, the network's share or the in-house platform's licence, and a fair share of the people who run the programme. Validated revenue means orders that survived returns, cancellations and rejection, which the tutorial on closing the month on affiliate commissions produces as a by-product of its monthly routine. A programme whose total cost of sale looks acceptable can still hold a voucher segment paying more than its margin and a content segment paying far less, and the average of the two tells finance nothing.
Then put each segment beside its contribution margin. The commission structures guide gives the ceiling: contribution margin per order, less the platform's cost per order, less the profit the business requires. Any segment whose cost of sale sits above the margin left after product, fulfilment, payment and expected returns costs is losing money on each order, which the reporting guide says is acceptable only when the customers are new and their later purchases repay the loss inside the payback period that the LTV and CAC guide explains how to set. The CPA calculator and the break-even calculator turn each segment's order value, margin and cost into a ceiling you can hold the current commission rateThe share of a sale's value paid to the affiliate or partner credited with it. against; the glossary entry defines the rate itself.
Finally, rank. The Pareto principle asks for a list, a value measure and a ranking with a running total, and it warns that a ranking by tracked sales puts the wrong partners at the top. Rank partners and practices by the commission they absorbed over the period against the new customers and contribution they produced. Where the running total climbs steeply, the few partners and practices at the top of the list are where the diagnosis table below is read first. Two neighbouring pages are deliberately not repeated here: the diagnosis of customer acquisition cost rising covers the blended cost across every channel, and the tutorial on auditing an affiliate programme reviews an inherited programme end to end. This page is for the narrower question: the affiliate line alone, growing faster than the validated revenue behind it.
Diagnosis table
Diagnosis
| Symptom | Check | Likely cause | Fix |
|---|---|---|---|
| Commission on returning customers and on low-margin categories is as high as on new, full-margin orders | Cost of sale by customer type and product category against each segment's contribution margin | One rate for every order, so returning customers and low-margin products pay as much as new ones | A higher rate for new customers, a lower rate or none for returning ones, category rates for margin, and exclusions for products sold at or near cost |
| Voucher, cashback and extension partners earn a rising share of commission while their new-customer share stays low | The partner type of the last affiliate click on orders that analytics credits to search, email or a content partner | Voucher, cashback and extension partners taking last click on orders other channels started | Deduplication rules in the terms that say which channels override an affiliate click, within what period, and which partners are protected from a later checkout click |
| Commission is paid on orders the order system later shows as returned, cancelled or reversed | Approved conversions matched by order reference against returns inside the returns window | Commission approved before the returns window closes, so refunded orders are paid | A validation period that covers the returns window, conversions held pending until it closes, and a clawback clause for chargebacks |
| Partner ads appear on searches for the brand's own name and modifiers, and their orders arrive moments after a brand search | Search checks on the brand term list and the brand campaign's auction insights for partner domains | Partners bidding on the brand's own terms and collecting commission on demand the brand created | A written brand bidding rule with a term list, negative keyword requirements, a narrow allow-list and the enforcement ladder in the terms |
| Placement and tenancy fees are renewed on the strength of the platform's attributed sales during the placement | Total orders in the placement period against a matched period or a holdout, with the fee and the commission in the cost | Placement and tenancy fees judged on attributed sales rather than on what they added | Price every fee from contribution margin and the partner's verified audience, prefer a hybrid of fee plus commission, and rebook only on an incremental result decided in advance |
| Tracked, paid and incremental figures have never been shown side by side, so nobody can say what the programme adds | Whether any coupon, brand-term or partner-type test has been run in the last year | No incrementality test, so commission is paid on sales that would have happened anyway | A planned holdout on the partner type or practice that costs most and is least certain, with the result fed into the rate structure |
Fix, verify and prevent each cause
One rate for every order
The commission structures guide is blunt about this one: a single rate for every partner is simple and almost always wrong, and the most valuable change most programmes can make is to stop paying the same for every order. A returning customer would probably have bought without the partner, and a percentage rate on a low-margin product can wipe out the profit while the same rate on a high-margin accessory is cheap. The affiliate channel page for online sales through affiliates puts the consequence in one line: the rate structure is the programme's targeting, because partners follow the money, and a structure that pays most for new customers attracts partners who find them.
Fix: split the rate by customer type, paying more for new customers and less or nothing for returning ones, with the terms defining what new means; set category rates where margins differ and exclude heavily discounted items, gift cards and products sold at cost; tier on quality signals such as new-customer shareThe proportion of conversions or orders that come from first-time customers, flagged from the business's own records rather than a platform's report., order value and low return rates rather than on volume alone. The change needs the tracking to pass a reliable new-or-returning flag from the order system, so check that before announcing anything.
Verify: after the first full validation cycle under the new structure, rebuild cost of sale by customer type and category again; the returning-customer and low-margin segments should now sit under their ceilings, and the new-customer segment should be paying at or near its allowed cost per order from the CPA calculator.
Prevent: make the break-even floor the test of every segment at each quarterly review, as the commission guide's review questions ask, and change rates structurally and with notice rather than in a panic; the guide's rule is to raise rates for the segments you want before cutting the ones you do not.
Checkout partners taking the last click
The attribution and deduplication guide explains the default that makes this cause so common. The platform credits the last affiliate click within its window, which favours the partners at the end of the journey, voucher sitesAn affiliate partner that publishes discount codes and offers, often reached by shoppers who have already chosen and search for a code at checkout., cashback sites and browser extensions, while the content partner who introduced the customer goes unpaid and the paid search or email click that actually closed the sale is not visible to the platform at all. The recruitment, compliance and fraud guide adds the two mechanisms by which these partners reach the checkout: codes issued to one partner leaking onto voucher sites, and extensions that insert their own link at the last moment and overwrite the click of the partner who sent the customer. Because affiliate credit becomes a payment to a third party, every order credited this way is commission paid for nothing.
Fix: write deduplication rules into the terms and apply them, as the attribution guide sets out: which channels override an affiliate click, usually paid search, paid social and email, within what period before purchase, and which protected partners keep their credit when a voucher or extension partner clicks last. Tie each code to the partner it was issued to and withhold commission when a code appears through someone else, and restrict extensions to activations the customer chooses.
Verify: classify a month of tracked conversions by the channel of the last click analytics recorded, as the attribution guide's reconciliation does; the share deduplicated should match the rule, and the voucher and extension segment's commission should fall while its new-customer share stays where it was.
Prevent: keep the payment rule and the analysis separate, show the deduplication rule and the list of protected partners in the terms, and watch the trend in the gap between tracked and paid conversions each month, because the attribution guide lists a widening gap as one sign that voucher or extension partners are closing orders other channels started, alongside a tracking problem or fraud. Where the route is a leaked code rather than an extension, the diagnosis of coupon sites taking affiliate commission follows that one cause further.
Commission approved before returns are in
The closing-the-month tutorial lists tracked, pending, approved, rejected, invoiced and paid as different figures, and warns that commission approved by default because nobody acted before the platform's date is the most expensive mistake in the routine. The commission guide says the same from the other side: a validation periodHow long the business has to review an affiliate conversion before approving its commission, long enough to cover returns and cancellations. shorter than the returns window pays commission on orders that come back. The attribution guide adds that chargebacks can arrive later still and need a clawback clause.
Fix: set the validation period in the terms to cover the returns or cancellation window plus time to run checks, hold conversions whose window is still open as pending rather than approving them, validate every pending conversion against the order system by order reference, and add a clawback clause for reversals that arrive after approval. Where the platform supports status updates through the postback, send returns and cancellations as adjustments rather than waiting for someone to edit a spreadsheet before payment.
Verify: after one full validation period, match the month's approved conversions against the order system's returns for the same orders; the overlap should be close to nothing, and the estimate of commission still owed on pending conversions should land near what is finally approved, as the tutorial's last step checks.
Prevent: run the close on the same date every month with a named approver, keep the share of pending commission finally approved by partner type as the basis for the estimate, and never let the approval date pass unattended during a peak, when the clean-up tutorial for promotions asks for validation after the returns window.
Partners buying the brand's own demand
The guide to setting brand bidding rules describes the mechanism: a partner running paid search ads on the brand's own name, its misspellings or the brand plus a modifier such as discount code takes the click that the brand's own ad or organic listing would have taken, at a cost that now includes commission. Those searchers were already on their way to the business, so affiliate brand biddingAn affiliate running paid search ads on the brand's own name, variants or misspellings, taking clicks the brand would have had and earning commission on them. turns demand the brand created into a commission line, and the guide notes that a person searching the brand plus a modifier such as discount code is already at or near the checkout.
Fix: choose the default the guide recommends for most programmes, a ban with a narrow allow-list, and write the boundary out: the term list with misspellings, range names and modifiers, a requirement that partners using paid search add the list as negative keywords, whether the brand may appear in ad text or display URLs, and whether partner ads may link directly to the site. Handle breaches through the enforcement ladder in the terms, with the screenshot, query, date and market as evidence, and withhold commission on the conversions the breach produced rather than the partner's whole balance.
Verify: repeat the search checks across the term list in each market and read the brand campaign's auction insights for partner domains; the ads should be gone and the partner's overlap should fall out of the report.
Prevent: agree one term list with the brand's own search team so the affiliate terms and the search account describe the same searches, record every exception in writing with a review date, keep the monitoring log on a fixed rhythm, and test the assumption occasionally by restricting brand-term activity for a period and measuring whether total brand-search orders change, as the reporting guide describes.
Placement fees judged on attributed sales
The guide to buying affiliate placements and tenancies draws the line that this cause crosses: a commission pays for an outcome, a placement fee pays for exposure, and a fee is paid whatever happens, which makes it media. The platform will report the sales tracked through the partner during the placement, but some of those customers would have bought through the partner anyway and some through another route, so a tenancy renewed on the attributed figure is renewed on a number that proves nothing. The closing-the-month tutorial adds a quieter version of the same leak: placement fees mixed into the commission line, so finance sees commission rising when part of the rise is media.
Fix: price every offer from the business's own contribution margin and the partner's verified audience, not from the partner's projection; prefer a hybrid of a lower fee plus commission so the partner keeps an incentive; put the position, period, fee, evidence of delivery and disclosure in writing; and decide in advance what incremental result would justify rebooking. Code placement fees separately from commission so the two lines can be read apart.
Verify: compare total orders and their contribution margin in the placement period with a matched period, or with a holdout region the partner can exclude, including the fee and the commission in the cost; if the position only pays at the most optimistic order count, it was too expensive.
Prevent: keep a record of every placement against the partner with what it produced, book one period before any long tenancy, and never mix a placement with a promotion without noting it, because the promotion moves orders too.
No test of what the programme adds
The reporting and incrementality guide separates attribution, which decides how to split credit, from incrementalityConversions caused by marketing., which asks whether there was anything to credit, and it says the doubt in affiliate programmes concentrates in a few places: coupons, brand terms and the partner types closest to the checkout. A programme that has never tested those places pays commission on every sale its rules credit, and the dashboard cannot tell the business which of them would have arrived anyway.
Fix: run the test suited to the costliest doubt. For coupons and vouchers, stop partners publishing codes in some regions for a period and compare total orders and full-price orders with a control. For brand terms, restrict partner activity and measure total brand-search traffic and orders. For a partner type, pause it in some regions or for a share of traffic and compare total sales, which the guide calls the strongest test and the most disruptive, so agree it with partners and keep it short. Then feed the result into the structure: a type that adds little loses its rate on returning customers, and the money moves to the partners whose sales held up in the test.
Verify: cost of sale by partner type, rebuilt after the rate change, should fall in the tested segment without a matching fall in total orders; if orders drop with it, the segment was adding more than the test suggested and the rate was cut too far.
Prevent: present tracked, paid and incremental figures side by side every time the programme is reported, as the guide asks, and test the partner type or practice that costs most and is least certain once or twice a year, so that no rate survives for long without evidence behind it.
Prevention: a quarterly structure review against the break-even floor and a yearly test of the costliest doubt
Every cause above grew because the programme's figures were read at the level of the dashboard, where they are true and useless. The routine that catches the next rise early runs at three speeds, and all of it starts from the same rebuilt cost of sale by partner type and customer type on validated revenue, with every cost in it. The rebuilt table is not a one-off diagnosis; it is the monthly report, produced from the same data the close already uses, and read beside the segment ceilings from the CPA and break-even calculators each time.
Monthly, the close does most of the work. Pending conversions are validated against the order system by reference before the approval date, conversions inside the returns window stay pending, the estimate of commission still owed is written with its evidence, and the invoice is matched line by line to the approvals with placement fees and the platform's fee on their own lines. The same pass reads the trend in the gap between tracked and paid conversions and the voucher, cashback and extension segment's share of commission against its new-customer share, which is where last-click leakage shows first, and checks that codes are appearing only where they were issued. When finance asks why the platform's total sits so far above what analytics credits to the channel, the diagnosis of affiliate sales not matching analytics explains that gap line by line from the same export.
Quarterly, the structure is reviewed against the floor. The commission guide's review questions are the agenda: is each segment above break-even once fees are included, which partners bring new customers and are they paid enough to keep doing so, which partners earn heavily on returning customers or on codes customers would have found anyway, do the windows and the validation period still match purchase behaviour and returns, and are any partner-specific deals out of date. The reporting guide's tiering happens in the same meeting, on quality signals as well as volume, and the programme management hub places the review in the operating rhythm that keeps rates, terms and partners from drifting apart. Changes go out with notice as the terms require and a written reason, raising rates for the segments the business wants before lowering any other. Before a change the review decides reaches partners, the affiliate commission change checklist checks the rebuilt economics, the notice, the platform settings and the partners in one pass.
Yearly, the costliest doubt is tested. The reporting guide asks for an incrementality test once or twice a year on the partner type or practice that costs most and is least certain, and for the tracked, paid and incremental figures to be presented together so the programme is judged on all three. A brand bidding rule, a deduplication rule and a placement policy are assumptions until a test has looked at them, and the yearly test is how one of those assumptions is retired or confirmed each year. Keep the terms under version control and the platform's settings matched to the document partners signed, as the recruitment guide asks, so that a window, a rate or an exclusion in the terms is the window, rate or exclusion the platform applies. When the review finds several of these habits missing at once, the tutorial on auditing an affiliate programme is the better starting point than a single diagnosis, and when the rise is not confined to the affiliate line, the diagnosis of customer acquisition cost rising reads the blended figure across every channel. The affiliate troubleshooting hub lists the neighbouring symptoms, which are usually the same programme seen from another angle.
A bathroom retailer finds where its commission went (illustrative)
Ondravik Bathrooms
Ondravik Bathrooms, an invented online bathroom fittings retailer, has paid every partner a flat 8% of order value since its programme opened. Over the past 12 months commission has grown by 63% while validated revenue from the channel has grown by 11%, and finance has asked why. The programme manager rebuilds cost of sale for the year by partner type and customer type on validated revenue, with the network's fee, two bonuses and a seasonal placement fee included.
The table explains most of the gap at once. Returning customers now make up 57% of commissioned orders, and 81% of those orders were closed through two voucher sites whose new-customer share is 9%; the retailer has been paying the full rate on customers who had already bought from it, to partners that analytics shows were rarely the first touch. The returns window is 60 days, but the validation period in the terms is 30, so £14,200 of commission in the year was approved on orders that came back after approval. And a £6,500 fee for the top slot in a kitchen-and-bath review site's spring guide was renewed in the autumn on the strength of 410 tracked orders during the placement, with nobody asking how many of them the site would have sent anyway.
Ondravik changes four things. It introduces a 10% rate on new customers and a 3% rate on returning orders, with new defined in the terms as no order in the previous 24 months and the flag passed from the order system. It lengthens the validation period to 70 days and holds conversions pending until it closes. It writes a deduplication rule into the terms under which a voucher or extension click within 30 minutes of purchase does not overwrite an earlier content partner's click, and gives partners the notice the terms require. And before renewing the tenancy, it asks the voucher sites to stop publishing its codes in three of its eight delivery regions for five weeks: orders using codes fall by 68% in those regions, total orders fall by 4%, and full-price orders rise. The retailer declines the autumn tenancy, keeps the new-customer rate for the review site that sent the holdout regions' best full-price orders, and in the next quarter commission falls by 29% while validated revenue is flat.
Before rates change, the walk-through of the break-even calculator for an affiliate programme sets each segment's commission ceiling and the return floor its revenue over commission and fees must clear.
Common mistakes
- Reading the platform's dashboard as the programme's cost. It reports tracked sales under its own rule; the cost finance sees is commission on validated orders plus fees.
- Judging the programme on its average cost of sale. The average hides a voucher segment above its margin and a content segment below it.
- Cutting the rate for everyone. A flat cut loses the partners who bring new customers first and keeps the cheapest sales.
- Approving commission because the platform's date arrived. Commission approved by default is paid on returns, duplicates and fraud.
- A validation period shorter than the returns window. The programme pays on orders that come back.
- Deduplicating in secret. Rules applied without being in the terms lose the best partners; rules written in advance are accepted.
- Renewing a tenancy on attributed sales. A fee is media and is judged on what the position added, decided before it ran.
- Never testing. Without a holdout on the costliest doubt, nobody knows how much of the programme's revenue would have arrived anyway.
