Someone searches your product name, clicks the first result, and buys at full price. They never saw an affiliate. Yet a commission goes out, because a coupon site got its link into the funnel before the visitor reached checkout. That is coupon leakage: paying referral commission on demand you already had.
It is the least dramatic problem in affiliate marketing and one of the most expensive, because it is invisible in every aggregate you look at. Revenue is up, conversions are up, commissions are up, and a slice of your payout is buying nothing. It does not look like a fraud spike. It looks like a small permanent tax on your own brand traffic.
TL;DR
- Coupon leakage is commission paid on a sale the affiliate did not create, usually on brand traffic or a buyer already deciding.
- It is not always fraud; it is the predictable output of a last-click rule plus a coupon site sitting between intent and purchase.
- Four signatures find most of it: near-perfect conversion rates, brand-name clicks, one code used by everyone, conversions with no matching click.
- Fix the policy first: codes are yours, only codes issued to an affiliate pay, and bidding on your brand disqualifies a partner.
- Then make attribution provable: pass the referral code through checkout metadata and verify the webhook, so payout rests on a signed server-side event.
What coupon leakage actually means
Leakage is any commission that does not reflect incremental work. Coupon leakage is the version where the offer itself, a discount code, a deal listing, a cashback extension, gets credited for a purchase the visitor was already going to make.
The polite framing matters. A deal site with real traffic that surfaces your product to people who had never heard of you is doing the job you hired it for. That is not leakage. The difference is kind, not degree: the visitor was already in your funnel, had already chosen you, and the coupon site's contribution was being present when the last click got recorded. Some of that is deliberate. Much of it is the incentive you designed. If the last link wins and a coupon site can add a link at any point before the money moves, coupon sites will optimize for the only thing that pays: being last.
So leakage is a design problem before it is an enforcement problem. Who gets removed is downstream of what you count as proof that an affiliate caused a sale. If the standard of proof is "a coupon site appeared in the funnel," leakage is not a bug. It is the system.
The six ways money leaks
Almost every leak you find falls into one of six shapes. They differ in how brazen they are, not in what they cost.
| Leak pattern | What actually happens | The fix that works |
|---|---|---|
| Extension rewrite | A cashback or coupon extension overwrites the tracking parameter at checkout, on a sale it had no part in | Terms clause plus code-in-metadata attribution that ignores late client-side writes |
| Coupon page on brand traffic | A deals page ranks for "yourbrand coupon" and collects the last click from someone who already chose you | Check whether the click came from campaign traffic or from a brand search |
| A single shared code | One affiliate publishes a code everyone uses, returning customers included, and takes credit for all of it | Per-affiliate codes, and commission only on the code issued to that affiliate |
| Post-intent drop | The code appears in an exit popup or retargeting ad after the buyer already decided | Placement rules, plus a shorter attribution window for codes than for content |
| Your own promo codes | Launch and seasonal codes get reposted by affiliates and deal sites, discounting full-price buyers with no referral value added | Separate marketing codes from affiliate codes in terms and in tracking data |
| Self-referral and friends | An affiliate buys, or has people close to them buy, through their own link | Self-referral blocking on by default, plus a review before the first payment |
Two of these get called fraud and get people removed. The rest are gray, and gray costs more because nobody acts. A coupon site inside your terms, using a code you issued yourself, can redirect a meaningful share of brand demand into commission while breaking no single rule. The corrective is structural.
Why last-click makes leakage worse
Attribution models are proxies for causality, and last click is the crudest proxy that still ships. It gives full credit to whichever link was clicked most recently, which is fine when that link is a review or a newsletter that actually sent the buyer. It is broken when the last link is a code that appeared two seconds before payment.
Compare two affiliates. A comparison site spends real money on content and search, sends you visitors who had never heard of you, and converts at the rate cold traffic should. A brand-name deals page converts far higher because its traffic is not cold. Under a last-click rule both look identical: clicks in, conversions out, commission owed. The only thing separating them is data you may never record, such as whether the visit landed on a deep article or on your homepage after a search for your name.
Window length amplifies this. A 90-day window rewards honest content, and also gives a leaking code 90 days to overwrite an honest affiliate's cookie. How those overwrites resolve is covered in cookie windows and attribution models. For leakage the conclusion is narrow: the window should be long enough to reward real influence and short enough that nobody can squat on a decision made last week.

Four leak signatures in your own data
You do not need forensic tooling. Four patterns, read together, surface most of it in an afternoon.
| Signature | What you will see | What it usually means |
|---|---|---|
| Suspicious conversion rate | A very high click-to-conversion rate next to high-traffic affiliates converting far lower | The clicks are pre-qualified, often brand searches |
| Brand-keyword clicks | Clicks arriving on your homepage or pricing page with no campaign parameter | The affiliate is capturing people who already searched for you |
| Code reuse across the program | One code appearing on conversions attributed to different affiliates and audiences | Codes are being republished beyond their owner |
| Conversions with no click | A conversion credited from a code in checkout metadata with no recorded click in the window | Something is injecting the referral at the end of the funnel |
| One affiliate's refund spike | Refunds clustered on a single affiliate's orders | Incentive mismatch, or a mismatch between promise and product |
| Renewals that inherit oddly | Recurring commission flowing to an affiliate whose original conversion you cannot explain | One leaked first sale pays commission for years |
The first three are visible in the affiliate table, the last three need the conversions list. In Sled you can filter the Conversions page by status, provider, and order type, and the summary recalculates with the filters, so the filtered view and the filtered CSV export at /api/export/conversions match. If you would rather ask than click, the read-only MCP tools (list_conversions, list_affiliates, get_affiliate) answer "who has clicks but no sales" or "which conversions used this code" for any connected agent, on every plan including the trial.
Legitimate coupon affiliates versus leak sites
The hardest part is judgment, and the fastest way to lose a good affiliate is to treat a real publisher like a parasite. Ask what the site did before the last click.
A legitimate coupon or deal affiliate has an audience you can point to, a newsletter, a community, a channel, a search presence in your category. Its traffic converts the way discovery traffic converts. It adds context, a review, a comparison, an honest reason to buy. It does not claim every sale in your brand funnel, it can explain where a specific conversion came from, and it is fine with per-affiliate codes because its value is not derived from being last.
A leak site has traffic tilted toward your brand name, lists a code for every product in your category whether or not it knows them, publishes nothing but listings, and converts like a checkout page rather than a discovery page. Its answer to "how did you drive this sale" is a shrug and a link.
One filter costs nothing: ask whether the customer would still have bought without that link. If the honest answer for most conversions is yes, it is a leak however the traffic is dressed up. That is the incremental-revenue question at the center of affiliate program ROI, and it is worth answering before you pay.
Write the rules down: policy that closes the door
Most programs mention coupon sites and brand bidding in a single line, if at all, and then nothing that says what happens next. That gap is where leakage lives. Three additions are usually enough.
State code ownership explicitly. A clean phrasing: all discount codes are issued by you and remain yours; an affiliate may publish only codes issued to them; republishing another partner's code or a code they were never issued earns no commission. Without that line, a program paying "commission on sales made with your code" is technically paying for your own marketing code.
Define prohibited placement: brand bidding on your name and close variants, extension and toolbar injection, post-intent placement such as exit popups and retargeting aimed at visitors already in checkout, and aggregation of codes you did not distribute.
Say what counts as proof. Attribution comes from your provider's checkout data, not a screenshot, and commission is calculated on the pre-tax amount actually collected.
If you have no terms yet, start from the affiliate program terms template, which has a promotion clause built for this, and read 10 affiliate fraud tactics for the patterns worth naming individually. The point of naming them is the conversation: a policy you can quote turns an accusation into a compliance check, and lets you keep a useful coupon publisher by telling them the two things they may not do.
Make attribution provable, not assumed
Policy sets intent. Instrumentation decides what pays. The failure mode to design out is a referral code written into the browser, remembered in a cookie, and then trusted when the money arrives, because anything written client-side can be overwritten by the last script that runs. That is how extension rewriting works.
Sled keeps the browser out of the final decision. The tracking script at https://usesled.com/[slug]/t.js reads ?ref=CODE from the landing URL, stores it in a ta_ref cookie for your configured window (90 days by default, set per program in Settings with the commission rate and minimum payout), and pings the click endpoint once, deduplicated by hashed IP for 24 hours so refreshes do not inflate stats. That cookie is a convenience, not the source of truth. Your checkout reads the stored code and passes it as metadata.referral to Polar, Stripe, or Paddle, and attribution happens server-side when the signed webhook arrives. Nothing on the success page has to fire.
Two properties follow, and both fight leakage. Attribution mirrors what checkout recorded, so a code never attached to the order cannot claim it. And because conversions are keyed by provider order id, refunds reverse commission on the refunded amount, including refunds that arrive before the sale event, which are tombstoned so the later order is created already marked refunded. Recurring commission inherits from the subscription id of the original conversion, which is why a leaked first sale costs you monthly until you find it. Free trials and 100 percent discount orders charge and pay nothing, but when they carry a referral and a subscription id they are remembered, so the first real charge still credits the affiliate. Self-referral blocking is on by default, matching the order email against the affiliate email, and disposable-provider signups are blocked too. Wiring the referral field into checkout per provider is walked through in the Polar guide.
Your own discount codes are part of the leak
Programs audit affiliates and forget themselves. Launch codes, early-bird codes, and seasonal sale codes get reposted, and every order they touch is discounted whether or not an affiliate caused it. That is a marketing cost, not a commission cost, but it cuts twice: customers who would have paid full price learn to wait for a code, and any affiliate republishing your code can claim the last click on orders your own campaign generated.
Keep two categories of code. Promotional codes that you distribute and that never carry referral attribution, and affiliate codes issued per partner that are the only codes that pay. Say the difference in your terms and make sure your tracking data can tell them apart, so a republished promo code cannot generate commission just by sitting in checkout metadata. Q4 is when this gets loudest, and Black Friday affiliate campaigns assumes you drew that line before the traffic arrived.
The monthly 30-minute audit
Leakage is a maintenance task, cheap on a schedule and expensive after a bad quarter.
- Sort by conversion rate, not revenue. A publisher with a coupon page and a 20 percent conversion rate is a question, not a success story.
- Read one uncertain affiliate's newest conversions. Check amounts, order types, and whether the customer looks like a repeat buyer. Repeat buyers converting through a code are the clearest signature there is.
- Match every known code to its owner. A code you issued to one affiliate showing up across orders attributed to several means codes are circulating.
- Compare renewals against first purchases per affiliate. Healthy programs show renewals following their own sales. Renewals stacked under one affiliate often trace back to a single leaked conversion.
Then act before the next payout run. If the audit finds nothing, note that too: a program where you know why every conversion happened beats a higher number nobody can explain. First-time setup order lives in how to launch an affiliate program.

Warn, pause, remove
Once leakage is visible, the response should be boring and proportional. Three rungs are enough, and skipping to the third is how you lose an affiliate who would have fixed it in a day.
Warn in writing, naming the behavior, the clause, and examples from their own conversions. Most coupon publishers know which of their placements is gray, and a written warning with numbers attached resolves it more often than not. Pause second: hold pending commissions, stop crediting new conversions from that code, and state the conditions to resume. It is reversible, and it turns a dispute into a deadline. Remove third, and remove properly: new conversions stop counting, the same email cannot sign up again, and existing history stays intact for reporting, which is how Sled handles removal. Record the reason, so nobody has to reconstruct the evidence later.
The program you end up with is smaller and quieter than the one leakage built, and more of its commission pays for work that actually brought someone in.
FAQ
Is coupon leakage the same as affiliate fraud? No. Fraud means a rule was broken on purpose, such as an extension silently rewriting a referral parameter. Leakage is the larger category where nothing clearly prohibited happened but your rules credited a sale to something that did not cause it, because a last-click model plus a coupon page produces that outcome on its own.
Do per-affiliate codes actually stop code sharing? They stop the payouts, which is the part you control. You cannot govern what gets posted; you can govern the rule that commission is paid only on codes issued to that affiliate, and make it visible in your terms. When sharing a code stops earning for the person sharing it, the incentive to share it goes away.
How long should the cookie window be if leakage worries me? Long enough to reward a visitor who reads a review and subscribes weeks later, short enough that nobody can sit on a decision already made. Sled defaults to 90 days and the value is per program, so a product with a short consideration cycle can run 30. A shorter window shrinks the surface a late code can exploit, but it does not replace codes and placement rules.
What about a legitimate coupon site that drives real customers? Keep it, and separate the cases in writing. Name the placements that are allowed, issue codes only to them, and let them earn on the traffic they introduce. A leakage cleanup is not the end of coupon marketing, it is the end of paying for demand you already had.
Written by Sled, flat-fee affiliate software for Polar, Stripe, and Paddle. If you want the attribution plumbing handled before your next payout run, start with the Polar integration guide or start a free trial.