Article · 21 August 2026, corrected 8 September 2026 · 14 primary sources, 1 secondary · Verified against source on 7 September 2026

Paid, and still reversible

Your creator has been paid. The money can still come back. Here is why, and why it is so hard to see coming from inside any one platform.

There is a sentence in Awin's help documentation that rarely travels outside the network side, and it decides something important about their business.

"Commissions and amounts already paid to partners can't be reversed."

Now here is Mavely, since renamed Later Creator, describing something it calls Returns and Reversals:

Previously earned or paid commission that was later reduced or removed.

Same money. Same word, near enough. Opposite rule.

One of those companies treats paid as final. The other does not. And the reason both can be true at the same time is the thing I want to talk about, because it explains a great deal about why creators do not trust their earnings screens.

What these platforms are actually built on

Most creator commerce platforms are not affiliate networks. They are subnetworks sitting on top of affiliate networks.

That is not my label. It is Awin's, and it has a precise meaning in their own documentation: a subnetwork is "a collection of partners (subpartners) who are all working under one single partner name", where the subnetwork issues the tracking links, and "a subpartner can expect to sacrifice a percentage of their commission earnings from generating a sale to the subnetwork."

Now look at ShopMy, Skimlinks, Howl, LTK and Mavely. ShopMy's own brand documentation confirms it operates through Rakuten, CJ, impact.com, Pepperjam, Partnerize and Awin rather than running its own rails. The regional breakdown - Awin in Europe, Commission Factory in APAC - comes from Sacra, a third-party research platform, not from ShopMy. Sacra's reading is that the value is in using rails that "already know how to handle compliance, publisher terms, and payouts across many countries." That is an outside analyst's inference. It is a good one, and it is not the company's own words.

That is a sensible decision. Building settlement infrastructure across 30 countries is a decade of work and most of it is unglamorous. Standing on somebody else's is faster, cheaper and lets you spend your engineering on the thing creators actually see.

But it has consequences, and all 3 of them land on the same person.

The clocks stack

Every layer in the chain runs its own clock, and they run in sequence rather than in parallel.

The retailer's return window has to close. Then the network validates. Then the network's payment run comes round. Then the platform's payment run comes round.

Skimlinks says it takes roughly 90 days, and explains why: "Typically retailers will take 60 days to review commissions, with a further period before individual commissions clear and are paid out, so the normal period is around 90 days."

ShopMy tells creators their commissions "typically remain pending for 30 to 120 days, even if the retailer's return window has already ended."

That last clause is doing more work than it looks. Even if the retailer's return window has already ended. The risk the delay was originally protecting against has gone, and the delay is still there. At that point the wait has stopped being about returns and started being about settlement mechanics. Somebody in the chain is holding money that nobody can any longer lose.

The number the industry repeats is that publisher payments routinely take longer than 120 days. It comes from the APMA's guide to publisher payments, April 2025 - but from the sponsor's introduction rather than from the association's own research. It is written by Revving, a funding platform that lends against slow affiliate payments and therefore has a direct commercial interest in the number being high. It is quoted here because it is the number the industry repeats, and it is flagged here because of who wrote it. An APMA-authored figure would be better. I have not found one. It is the same 120 days ShopMy quotes its creators, 1 layer further out.

The cut stacks too

Every layer takes a slice of the same commission pound.

I could not find a single platform that publishes the stacked total, and I looked. A creator can usually discover what their platform takes. Finding out what the network underneath took first, before the platform's number was calculated, is a different exercise entirely.

I am not suggesting anybody is hiding it. I am saying that no layer has a reason to publish somebody else's number, so the total exists nowhere, and the person whose income it is has no way to work it out.

And then the rules disagree

The first 2 stack into something worse than inconvenience.

Awin says commission already paid to a partner cannot be reversed. Once the money has gone out, that transaction is closed. The finality is deliberate. It is what lets a publisher treat a payment as income rather than as a provisional figure.

Worth being exact about which stage that covers, because the 2 get conflated constantly. Approved and paid are different states with different rules, and Awin's own FAQ separates them: approved commission can be reversed by their support team, and on some plans that is a routine request rather than an exception. It is the payment that is final, not the approval. A sale can be approved and still go backwards afterwards, whatever sent the merchant back to it. Once the money has reached the partner, it cannot.

Mavely runs reversals against commissions it has already paid, and says they may reduce a future payout.

Mavely's adjustment mechanism cuts both ways, and it is worth saying so. Alongside Returns and Reversals it runs Historic Data Additions - "Eligible commission added after an earlier payout cycle was processed." The correction runs in both directions, which makes the point harder to dismiss rather than easier.

Both statements are accurate. They are describing different rails.

The network layer has decided that paid means finished. The layer sitting on top of it has decided that paid means paid for now. And the creator, who is standing at the end of the chain and can only see the layer they signed up to, carries a risk that the layer underneath has already extinguished.

That is not a bug anybody introduced. It is what happens when 2 systems with different rules get stacked and nobody reconciles the difference.

It is also completely invisible from either end on its own. From the network's side, everything is final and correct. From the creator's side, money that arrived has gone away again. You only see it if you follow the money through both layers.

The words do not survive the journey either

Here is the same problem in miniature, and it is my favourite thing I found.

"Lock" does 2 opposite jobs depending on whose dashboard you are looking at.

On ShopMy, Locked is where the money has arrived. Their own guide is explicit: "Once ShopMy receives those funds, your commission moves to Locked and is included in your next eligible payout." Locked is safe.

On Levanta, the locking window is the period the money is still at risk, and a conversion becomes final only once that window has expired: "Once a conversion passes its marketplace's locking window, it's considered final and won't be adjusted further."

Same root word. On one platform it marks the end of the risk. On the other it is the risk.

A creator working across 2 of them is working with 2 different definitions of the same word, usually without being told. So is an operations manager who has moved between platforms and carried the old meaning with them.

"Pending" is no better. On ShopMy the blocker is the retailer paying ShopMy. On Shopify Collabs it is the merchant's holding period, and the merchant sets it: "The holding period is set by the merchant, and can be between 1-90 days." The dispute right sits with them too: "Merchants can open a payment dispute for commission that's in Pending state as long as it's within the merchant-selected holding period." On Awin it is the advertiser not having validated yet. LTK calls the same state Open, and quotes around 120 to 150 days for it. Same word, 4 different parties responsible, and the creator has no way of knowing which one they are waiting on.

And 3 completely separate clocks all get called "the window": the cookie window, which is how long after a click a purchase still counts. The attribution window, which is the merchant's own crediting rule. On YouTube Shopping that window is fixed at 30 days by YouTube, while the pending period is the one the merchant sets, between 30 and 50 days. Two clocks that sound the same, owned by different parties. And the return or lock window, which is how long the sale can still reverse. Conflating them is the most common attribution mistake in the category, and it is easy to do because everybody calls all 3 the same thing.

Why any of this matters commercially

If you run one of these platforms, the creator experience of "my money went backwards" is not a support problem. It is a design consequence of a state being shown before anybody could stand behind it, on a rail whose rules differ from the rail underneath.

And there is a forecasting version of the same thing. If your payout timings are set by clocks belonging to a network you do not control, sitting behind a retailer you do not control, then your working capital assumptions rest on variables that are not in your model.

What I would ask

What I could not establish

The 120 day figure comes from the APMA's guide to publisher payments, April 2025 - but from the sponsor's introduction rather than from the association's own research. It is written by Revving, a funding platform that lends against slow affiliate payments and therefore has a direct commercial interest in the number being high. It is quoted here because it is the number the industry repeats, and it is flagged here because of who wrote it. An APMA-authored figure would be better. I have not found one.