Peak PO Discipline for Amazon 1P Vendors

Update, 25 October 2026: Amazon has extended the Seller Central Black Friday and Cyber Monday deal deadline from 20 October to 13 November. The dates below were correct when this piece was published; the extension applies to sellers, and Amazon still publishes no Vendor Central date.

A commercial agreement with Amazon does not obligate Amazon to issue you a purchase order, and it does not obligate you to accept one. I say that to vendors because I sat on Amazon's side of that table. The vendor manager across from you owns a profit and loss. They are measured on top line and on profitability at the same time, and they cannot grow the top line without your inventory. That is a position to negotiate from, not an invitation to start declining purchase orders.

It matters now because peak is when the gap between Amazon's forecast and Amazon's actual purchase orders turns expensive, and the expense is lopsided. Confirming units you cannot ship is charged. Shipping units nobody ordered is charged at the full value of the goods. Reducing a confirmation costs nothing inside a short, dated window and is charged after it closes. So purchase order discipline in the weeks before Black Friday is mostly about which error you choose, and about using the window Amazon prices at zero before it shuts.

What is Amazon's demand forecast actually telling you?

It tells you what Amazon expects shoppers to buy, and you should not assume it knows what you are planning to spend to make them buy it. The Amazon demand forecast in the Forecasting and Inventory Planning report is not a purchase commitment, and the purchase orders that arrive will differ from it.

The report has been showing 13 weeks of order history alongside a 26 week forward forecast, at mean, P70, P80 and P90 probability levels. The structure is the stable part; contents change, so read your own current report rather than any description of it, this one included. The Sell-In Forecast, still in beta, shows what Amazon intends to buy over roughly the next three months, and beta programmes change without notice, so treat it as a signal rather than a contract.

The demand forecast reflects expected consumer demand only. It does not account for the promotions, advertising or deal events Amazon does not know about yet, and you should not assume it counts the ones you have entered. Plan as if everything you are about to do to create December demand sits outside the number Amazon handed you, which is why a peak forecast and a peak plan diverge by design rather than by error.

On the probability levels, Amazon is reported in an agency's guide to prefer that vendors plan to P90, and Amazon does not publish that as a requirement. The higher level is the safer one for Amazon and the expensive one for you, because P90 on a slow-moving item becomes a storage and markdown problem in January. Pick the level per item, from the cost of a stockout against the cost of an overstock on that item, rather than setting one level for the account.

Amazon's forecast is also generated without regard for pallet or truckload economics. A former Amazon vendor manager put it plainly in an interview: if the internal forecast says one unit, a purchase order for one unit can follow. Take that as an illustration rather than a rule, and as the right expectation to hold when a peak purchase order arrives in a shape no freight planner would have chosen.

How far ahead you see purchase order volume depends on your supply chain setup more than on the forecast. A vendor on the Direct Import programme gets purchase orders issued much further out, because the order has to cover transit from an overseas factory to the Amazon fulfilment centre. Everyone else sees the P70 and P90 bands. What I told my vendors about those bands still holds: they are predictions to plan demand around, and the further out the prediction sits from today, the less likely it is to survive contact with the purchase order.

Deadlines in the next 30 days

Dates checked on 10 October 2026. The carrier surcharges and the USPS increase have started, and the October Prime event ran on 6 and 7 October.

The most useful thing anyone can tell a 1P vendor about peak deadlines is that Amazon does not publish yours. Vendor Central promotion and deal submission windows for Black Friday, Cyber Monday and the October Prime event are announced behind the login, and every public page quoting a 2026 vendor deadline traces back to a Seller Central notice that does not mention 1P at all. Ask your vendor manager or your Amazon Vendor Services brand specialist, or read the Vendor Central News feed inside your own account, and distrust any agency page that hands you a 1P date.

Across Vendor Central and Seller Central, the same event carries a published deadline on one side and a login-gated one on the other, and the side with the published date is not yours.

Amazon ran the 2026 October Prime event on 6 and 7 October. The Black Friday and Cyber Monday window is not yet announced; last year Amazon ran it as a single window from 20 November to 1 December, and its press release archive is where the 2026 dates will appear. The October event's Seller Central deal deadline was extended in-account from 8 to 22 September, and like every Seller Central date it says nothing about your Vendor Central window.

What does a missed peak PO cost?

Four charges do most of the damage. Three of them sit under Amazon's In Full Delivery consolidation, effective 21 July 2025; Not on Time is a separate chargeback. The rates below are what compliance specialists consistently report, and no public Amazon document states them, so read them against your own vendor agreement and your own chargeback report rather than against this table.

Over-shipping is the most expensive error available to you, at the full value of the goods, and it is exactly the error an optimistic peak plan produces. The on-time waiver is the rule specialists consistently describe rather than one Amazon states anywhere public, so confirm it against your own account before you rely on it, and do not assume it extends to Down Confirmed, because it does not. If you see 10% quoted as the Not Filled rate, that is the historical figure from before the July 2025 consolidation, still printed as current on a lot of agency pages.

How on-time is measured depends on who moves the freight. Prepaid vendors are measured against the carrier's requested delivery appointment; collect vendors are measured against the freight-ready date. Confusing the two is a quiet source of Amazon vendor chargebacks in a quarter where every truck is late.

Chargebacks add up across a year, and more so in Q4 when volume peaks and defect rates rise. You have about 30 days to file the first dispute, counted from the date on the chargeback record in Vendor Central, and a refusal opens one more 30-day window, giving two attempts.

Should you confirm every purchase order?

No, and confirming everything is not what a healthy vendor account looks like. Specialists put the PO confirmation rate benchmark at roughly 80% to 85%, on a single source and with nothing published by Amazon behind it, so treat a rate in that band as ordinary rather than as failure and check what your own vendor manager is actually measuring you against. Their number is the one that governs you.

• Confirm what you can ship. The forecast is not a commitment in either direction, and neither is the purchase order until you confirm it.
• Use the free window. Reducing a confirmation inside five business days of the ship window opening carries no charge. After that it is 3% of cost of goods with no waiver. That window is an option you hold.
• Never round up. Invoicing above the ordered quantity is charged at 100% of the excess, so the optimistic case costs more than the cautious one.

The confirmation rate is understood to be reported as an aggregate at the highest tier level, so line-item confirmations roll up and a single unavailable item barely moves it. Check your own report before you assume that holds for your account.

Inside Amazon the confirmation rate is a management number, not a report line. The category in-stock teams monitor it, and the vendor manager and in-stock teams together watch four figures closely: purchase order confirmation rate, rejections, out-of-stock percentage and Retail Fast Track. When any of them moves, the team is expected to explain why and say whether there is a plan to fix it. So a large peak purchase order that you down-confirm does get noticed. If the impact looks significant, the vendor manager may call to understand why and see whether it can be resolved. If it leaves them worried about out-of-stocks through the event, promotions on the item can be cancelled, because nobody funds a deal on an item they expect to run out.

Discretion runs wider than the confirmation screen. In an annual negotiation, a vendor facing aggressive terms demands can offer to de-list selection on less profitable items, which addresses Amazon's profitability problem directly and reframes the conversation as a trade between profitability and growth. The line is: you can't sell what you don't have. I used that counter often, against cost-decrease asks and against accrual increases. Every de-listed ASIN is a hit to the vendor manager's sales, and they want to sell your whole catalogue, so offering to remove an item is a fast test of how real the profitability problem on it is. It also sets the level at which you give anything back. Amazon manages profitability item by item, so if you concede money or terms, concede them at the ASIN, never across the catalogue.

What changes on Direct Fulfillment?

The rule tightens. Cancelling an order you have already accepted is one of the only Direct Fulfillment infractions carrying a direct financial fine, and a high pre-fulfilment cancellation rate can suspend the programme. Amazon publishes no hard thresholds here, so there is no number to aim at and the operating rule is to accept only what you can ship. Through Q4, operators extend planning lead times, deepen the safety buffer while carrier delays spike, acknowledge each order within about 24 hours, and confirm shipment only at true carrier handoff.

What to run weekly between now and Black Friday

Four checks, and one of them moves to daily. Operators who plan this well review forecasts weekly in steady state and daily through October to December, holding 30 to 60 days of supply normally and 60 to 90 days in Q4 with 15 to 45 days of safety stock, built to true lead times that often run 60 to 120 days. Those are common rules of thumb rather than benchmarks or an Amazon rule, so substitute your own lead times and velocity.

1. Move forecast review to daily from October. Weekly is a steady-state cadence and peak is not steady state.
2. Pair open purchase order quantity against sellable on-hand units. Open POs rising while sellable on-hand stays flat points at Amazon's receiving rather than at your shipping, and that changes who you escalate to.
3. Watch the trailing twelve-week fill rate. Above 95% is what waives the Not Filled charge, so it is one of two metrics with a price attached going into December, alongside the on-time rate behind the Not on Time waiver.
4. Diary the 30-day dispute window. Peak charges are reported to be assessed at each invoice's due date or 120 days after the ship window ends, whichever comes first, so they keep arriving through December and into the new year, and any 30-day window that runs over the holidays is money you never get to argue about.

Peak chargebacks start being decided in September. The purchase orders you confirm and fill over the next three months set what you are billed later, because you are measured against what you promised as well as what you shipped. The Amazon deductions that arrive after peak are the bill for choices already made, and the point of a margin recovery audit in the new year is to find out which of them you never agreed to in the first place.

If you want to know what last peak already cost you before this one adds to it, we run a free deduction scan for 1P vendors: send three or four standard Vendor Central exports, which you pull yourself so no login is shared, and get back a one-page readout of your estimated recoverable dollars, split by the streams your exports cover, and your top three findings. No commitment, and you keep the analysis either way.