Amazon Makes Jan 31, 2027 Returns Permanent: Nov 1 Buys Get 91 Days, Apple 75

The Analysis

A purchasing director I have known since my Shenzhen days called me on 17 October with a spreadsheet open and a bad feeling about it. She buys home textiles out of Nantong and sells them on Amazon Germany and Amazon US. Her Q4 orders were already signed: forty-six containers, the last of them sailing on 8 November. What she had just read was Amazon's notice of 16 October, and what she wanted to know was whether she could still change anything. She could not. That is the part worth sitting with.

Here is what the notice says. Goods ordered between 1 November and 31 December 2026 can be returned until 31 January 2027. Something bought on 1 November therefore has 91 days of return runway against the standard 30. Apple-branded products have to be back by 15 January, which is 75 days from the same 1 November purchase and five times the 15 days those products normally carry. It applies to FBA and to merchant-fulfilled orders, and it applies whether you asked for it or not. And it is not a one-year gesture: Amazon has said this is now an annual rule.

So before anyone gets exercised about the principle, let us do what I always do. Open the cost structure and find out who is paying, how much, and when. Everything is negotiable, but only the parts you can still move, and the list of those parts is shorter than most sellers think.

Start with the calendar, because the calendar is where the money hides. A unit that leaves the mill in Nantong in the second week of October is on the water for roughly thirty days, clears, goes into a fulfilment centre, and is available to a shopper in early November. The shopper buys it on 1 November. Under the old rule I held 30 days of exposure on that unit: whatever comes back comes back by 1 December, while the season is still open, while the size still sells, while the Q4 reorder decision is still live. Under the new rule I hold that exposure to 31 January 2027. Sixty-one extra days. Sixty-one days is not a rounding error, it is the entire back half of the season plus the clearance period, and every unit that comes home inside it arrives after I have already made my January buying decisions.

And it is not one window, it is two. Apple-branded goods close on 15 January, 75 days from a 1 November order, five times the 15 days they normally get. Read that carefully, because the multiple is doing the work in the headline. Fifteen days is what you give a category that almost never comes back; Apple's standard window is short precisely because the return rate is low. Taking 15 days to 75 is a five-fold multiple, but the absolute addition is 60 days, within one day of the 61 added to everything else. If you sell accessories, the number to worry about is not the 5x, it is 60 extra days sitting on a product that gets bought as a present. Presents come back in January whether the window is 15 days or 75, and the ones that come back in the second half of January have been opened, used and reboxed by somebody who never intended to keep them.

Why 16 October, and why permanent. The date is the tell. Amazon gave sellers fifteen days between the announcement and the first day of the purchase window. Fifteen days is not enough time to reprice a Q4 assortment, not enough time to reopen a signed purchase order, and just about enough time to panic. Announcing it this late is the mechanism: whatever sellers think of the rule, the Q4 inventory is already bought and already on the water. As for annual, that is the more consequential half of the sentence. A one-off holiday extension is an expense you absorb. A rule that repeats every year is a parameter. Parameters go into the budget, into the safety-stock calculation, into the price, and into the conversation with the factory.

Why Amazon wants it at all is not mysterious, and knowing why helps you guess what comes next. A longer window converts a hesitant shopper, and the cost of that conversion lands on the seller's profit and loss rather than on the marketplace's. Every major marketplace has been lengthening windows for two years; if Amazon ran a 30-day window through December while everyone else ran into January, it would lose the gift purchase, which is the highest-value purchase of the year. The Apple carve-out tells you this is commercial rather than sentimental. Apple goods are the category with the shortest window and the lowest return rate, and Amazon still stretched it to 75 days, because an Apple product ordered on 1 November is a present, and a present needs an escape hatch that lasts past New Year.

So open the unit. I am going to put my own numbers on this and I want to be explicit that they are mine: assumptions, not Amazon data, not anybody's published figure. One SKU, 10,000 units shipped out of Shenzhen on 25 October, on sale from 1 November at 39.99 dollars. Landed cost I will set at 14.50 dollars a unit, split as 8.00 FOB, 3.50 for ocean freight, duty and drayage, and 3.00 for FBA fulfilment. Amazon commission at 15 per cent takes 6.00 off the top. Peak fulfilment fees, which run from 15 October to 14 January 2027, I will average at 0.55 dollars a unit. Assume the return rate under an ordinary 30-day window is 8.0 per cent, and assume the extended window adds 3.0 percentage points, with all of it landing between day 31 and day 91.

Work the 800 units that come back inside 30 days first. On each of them I have already burned the 3.00 fulfilment fee and I will burn another 2.20 on reverse handling, inspection and repacking, so 5.20 a unit is gone before I know whether the goods are sellable: 4,160 dollars. Of those 800 units, assume 70 per cent can go back on the shelf as new. That leaves 240 that cannot, and I will value them at 45 per cent of landed cost, 6 dollars 53, so the write-down is 240 times 7.97, call it 1,914 dollars. Add 224 dollars of carrying cost on the 560 units that sit for two months before they sell again. The first 800 units cost me 6,298 dollars, or 63 cents on every unit I sold.

Now the 300 incremental units that arrive between day 31 and day 91. Same 3.00 of sunk fulfilment, same 2.20 of reverse handling, so 1,560 dollars before anything else. What changes is what the goods are worth when they land. They arrive in January, after the season, so I will assume only 25 per cent can be resold as new, and that those 75 units wait eight months for the next season at 20 cents a month in carrying cost, 120 dollars, plus a 20 per cent markdown to move them, 600 dollars. The other 225 units are off-season stock nobody wants at full price; value them at 25 per cent of landed cost, 3 dollars 63, and the write-down is 225 times 10.87, about 2,447 dollars. The 300 units cost me 4,727 dollars.

Put those two numbers side by side. Three hundred extra units is 27 per cent more returned volume, and it costs 4,727 dollars against the 6,298 dollars for the first 800, which is 75 per cent as much money. My return provision goes from 63 cents a unit to 1 dollar 10. If my net margin on a 39.99 dollar item is 8 per cent, that is 3 dollars 20, and 47 cents has just taken 15 per cent of it. That single subtraction is the whole rule.

Here is the part the headlines are not carrying, and it is the part I would want a buyer to walk away with. Everyone is going to model this as a return-rate problem: how many more points does a 91-day window produce. Wrong variable. The incremental units are not the same population as the base units. Somebody who decides within 30 days that a duvet cover is the wrong colour sends back something I can put straight back on the shelf. Somebody who holds a purchase from 1 November to the last week of January has lived with it, used it, and in a large share of cases is returning a gift they never chose.

Those units come home opened, out of the original packaging, out of season, into a warehouse already full of January clearance. My 25 per cent resale assumption against 70 per cent in the first bucket is the entire argument in one line. This rule does not cost you its length in returns; it costs you the residual value of the units that come home late, and residual value is a purchasing problem long before it is an operations problem.

A second thing hides in the dates, smaller but real. The return window closes on 31 January 2027. Amazon's peak fulfilment fee window closes on 14 January 2027. Seventeen days apart. Anything returned in that gap comes back on standard rates, which sounds like relief until you notice that you paid peak rates to send it out and you are getting back a unit you can no longer sell at full price. The peak fee period and the return exposure cover almost the same stretch of calendar and then diverge exactly where the goods are worth least. If you are building a Q4 model, do not net those two against each other. They are not the same shape and they do not cancel.

Then there is the cash, which is what actually breaks a thin balance sheet. A 1 November order is delivered in early November, Amazon settles roughly fourteen days later, and that money lands in November or December as revenue. It is not revenue, it is a deposit that can be clawed back until 31 January. If 11 per cent of it comes back, and most of that in January, then the honest Q4 margin does not exist until the first week of February 2027, when the last refunds have cleared and the last returned unit has been inspected. Anyone closing the Q4 book on a 31 December snapshot and booking the full amount spends February explaining the gap. I have watched a business pay tax on revenue that was refunded six weeks later.

Who it lands on is worth a paragraph, because the answer is not uniform. High ticket pays more than low ticket: the write-down is a share of landed cost, and landed cost scales with the item. A 19.99 dollar case that cannot be resold costs me about 6 dollars. A 199 dollar pair of headphones that comes back opened in late January costs me the margin on 199 dollars, and no amount of repackaging turns an opened audio product back into a new one. Seasonal product pays more than evergreen product, because the late units miss the season entirely. Apparel and footwear pay on top of that, since sizing returns are already most of the volume and a 91-day window lets a size return sit in someone's hallway for two months before it moves.

And merchant-fulfilled sellers are worse off than FBA sellers on the operational side, though not on the money: if you ship yourself, you now carry a 91-day obligation you did not write and did not price, and you absorb the whole reverse leg yourself. Once the goods are back in the building, how fast they get inspected and whether they make the restock window is Nadia Pryce's territory and I am not going to stand in it. I stop at what the unit is worth on my books.

Now let me argue against myself, because a rule this size deserves it. My 3.0 percentage point assumption is the softest number in the calculation. If your real number is 1.0 point, the damage falls from 47 cents a unit to about 16 cents, which on an 8 per cent margin is 5 per cent of profit instead of 15. That is the difference between a rule you reprice for and a rule you shrug at, and you do not have to guess which one you are. Pull your own Q4 2025 return data and bucket it by the number of days between order date and return request.

Whatever share sat in the 31-to-91-day bucket last year is your starting point. If you never had a 91-day window, take the tail you do have and extend it honestly rather than generously. My second reservation sits on the resale assumption: 25 per cent may be pessimistic for a hard good in a plain carton and optimistic for apparel the customer has already washed. Move that one input and the answer shifts more than any other.

Which brings me to the alternatives, and the first one everyone proposes is the wrong one. The instinct in the room was to cut the November order and air-freight the shortfall in December. Run it. Take the first order from 10,000 units to 8,000 and air-freight 2,000 units in at 4 dollars 50 a unit more than ocean: 9,000 dollars of extra freight. What you save is the exposure on 2,000 fewer units inside the 91-day window, which at 3 per cent is 60 units, and at 15 dollars 76 of combined sunk cost and write-down each, about 946 dollars. You are paying 9,000 to avoid 946. Cutting the first order is not a return-window strategy, it is an air freight bill with a story attached to it. If the goods are going anyway, send them the cheap way and price the risk instead.

What I would do costs no freight at all, and there are three moves. Begin with price. Forty-seven cents on a 39.99 dollar item is 1.2 per cent of revenue, and if your Q4 pricing is locked in October, the provision goes into the 1 November price rather than being discovered in February. Next, change the purchase terms. On forty-six containers the factory can carry some of this: ask for a clause sharing the cost of returns above an agreed rate, or for payment on 15 per cent of order value to be deferred to 15 February 2027, by which date you actually know what the season was. A factory that wants the 2027 programme will talk about it. A factory that will not is telling you something about the relationship, and you should hear it. After that, change the mix: if a SKU is seasonal and its late returns are worth 25 cents on the dollar, it should not be 30 per cent of your Q4 buy.

On inventory, an annual rule changes how I set the numbers. If the return tail now runs to the end of January, a restock review on 1 February is being made blind, because up to 3 per cent of Q4 volume is still in transit back to you. Push the review to the second week of February and treat returned units as a separate supply line with its own lead time, not as a surprise that turns up in the warehouse. On stock-to-sales, I would set the Q4 figure one to two points lower than I would otherwise set it, because a slice of what you counted as sold is going to walk back in.

And on the annual point, take it into August: if I am signing the Q4 2027 programme with a Nantong mill next August, I am not signing the contract I signed in August 2026. The return tail is now a known, dated, annual cost of roughly 47 cents a unit on this SKU, and it belongs in the same negotiation as freight and duty. Factories understand freight clauses. They will understand this one too, once somebody puts a number on it.

The trap I would name last is one I have fallen into myself: treating this as a customer-experience story and leaving it with the marketplace team. It is not. It is a landed-cost line, and landed cost is a purchasing problem. Amazon decided on 16 October, applied it to everyone on 1 November, and will apply it again next year. The parts I can still move are what I pay the factory, what I charge the customer, and how much seasonal product I am willing to put into a window that stays open to the end of January. Everything else is somebody else's rule.

↑ Back to the key points

— By Grace Yang