Why your Amazon ad sales and ACoS keep changing after the week ends

Amazon ad sales for a week are not final when the week ends. Amazon credits a sale to an ad if the purchase happens within a set number of days after the click, and books it on the day of the click. So a closed week keeps gaining sales, and its ACoS keeps falling, for up to two weeks. Smaller corrections can follow for weeks after that.

ACoS (advertising cost of sales) is ad spend divided by the sales Amazon credits to those ads. Most guides explain the windows. Few show how far one week’s numbers move, which ad types move most, or how to compare weeks when one is still young. That is what this post covers.

How Amazon credits a sale to an ad

When a shopper clicks your ad, Amazon keeps watching that shopper for a set number of days. If they buy within that window, the sale counts for the ad. The time allowed is called the attribution window.

Two details cause most of the confusion:

• The sale is booked on the click date. A shopper who clicks on Sunday and buys on Thursday adds a sale to Sunday. Sunday’s numbers change on Thursday.
• Each ad type has its own window. A longer window means more of a week’s sales are still to come.

The sponsored ad windows are the defaults Amazon lists in its reporting guide for the Ads API, which match the ad console. Amazon’s reporting guides also say sales are reported on the date of the ad click, not the purchase. Amazon’s own guide says Attribution uses a 14-day, last-touch model, so the most recent click gets the credit.

Reports pulled through the Amazon Ads API (the way software pulls ad data from Amazon) add a wrinkle. For Sponsored Products they can carry sales on 1, 7, 14 and 30-day windows side by side, and a tool or agency picks one. Standard Sponsored Brands and Sponsored Display reports carry 14-day sales. So “last week’s ad sales” can mean several different numbers, all correct.

Amazon keeps correcting sales after the window closes

The window is not the end of it. Amazon’s reporting FAQ says it restates sales 1, 7 and 28 days after each purchase. With a 14-day window, a day’s numbers can change up to 42 days later. Its reporting overview says sales can change for up to 60 days, and suggests pulling reports again each month. The Amazon pages we read do not say which way these changes run or how big they tend to be.

What about invalid clicks?

Amazon filters out clicks it judges to be invalid, such as bots. Its reporting overview says it checks every click so advertisers are not charged for invalid ones. Most are caught at once. Some take longer, and Amazon watches for them for up to 30 days. It suggests waiting at least 5 days for the most accurate click data.

Amazon also offers a gross and invalid traffic report for all three sponsored ad types. It shows total clicks and the invalid clicks removed. So its total click count can be higher than the clicks in a normal campaign report.

A worked example: one week pulled four times

Illustrative numbers. A brand selling candles and hand soap spends $8,000 on Amazon ads in the week of Monday, August 17 to Sunday, August 23, 2026. It pulls the same week from the ad console four times. Each pull uses the console windows: 7 days for Sponsored Products (SP), 14 days for Sponsored Brands (SB) and Sponsored Display (SD).

Spend is held fixed to keep the table readable. In real reports it can dip a little as invalid clicks are removed. Sales are what move most.

Nothing about the ads changed between August 24 and September 6. The week simply kept counting. On day 1 it holds 80% of the sales it shows on day 14, and its ACoS reads 8 points worse.

The totals hide the more useful story, which is by ad type.

Sponsored Products is nearly settled by day 7, because its window is 7 days. Sponsored Brands and Sponsored Display keep climbing to day 14. Pulled on day 1, Sponsored Brands looks like the worst search ad in the account. By day 14 it is the best.

Why the ad types move by different amounts

Three things set how far a week moves after it ends.

1. The length of the window. A 7-day window stops adding sales a week after the last click. A 14-day window keeps going for twice as long.
2. How quickly shoppers buy. Someone searching for “lavender hand soap” often buys the same day. Someone who saw a display ad while browsing may take days. Ads that reach shoppers earlier in their search tend to have more late sales.
3. Where the week’s clicks fall. Every pull of a week holds clicks of different ages. On day 1, Monday’s clicks are 7 days old, but Sunday’s are only 1 day old. A week with a heavy weekend reads younger than it looks.

Amazon’s later corrections, 7 and 28 days after each purchase, can still shift the numbers after that. So a day-14 reading is close to settled, not locked.

Where it goes wrong

These are the traps we see most often in real ad data.

Comparing a young week with a settled one

On Monday, September 14, the brand pulls last week, September 7 to 13. ACoS is 41.0%. It sets that next to the week of August 17, which now reads 33.1%. ACoS looks up 8 points in a month.

It is not. The August week read 41.5% when it was also one day old. At the same age, last week is slightly better. The 8 points are age, not performance. Every raw week-over-week read has this bias built in: the newest week tends to look worst.

Cutting a campaign on early numbers

In the example, Sponsored Brands read 46.7% ACoS on day 1. If your break-even ACoS is 40%, that looks like a campaign losing money. On day 14 it reads 31.8%, comfortably profitable. Cutting it on day 1 would have cut the best performer in the account.

What is safer to judge early: spend, clicks, cost per click and click-through rate. Amazon says small changes to clicks can happen for about three days, so give them five. What is not safe: sales, orders, conversion rate and ACoS.

Mixing 7-day and 14-day windows in one total

The console total adds Sponsored Products sales on a 7-day window to Sponsored Brands and Display sales on 14-day windows. That total is useful, but it is a blend of two definitions. Two traps follow from it:

• Switching tools changes the number. Move from the console to an API tool that shows Sponsored Products on a 30-day window, and ACoS “improves” the month you switched. Nothing happened except the definition.
• Adding Amazon Attribution sales. Off-Amazon traffic and a sponsored ad can both claim the same order. Adding the two counts some sales twice.

Label every ad sales figure with its ad types and its window. Keep the split by ad type next to any blended total.

The agency and the console disagree

Three sources report the Sponsored Products ACoS for August 17 to 23. Illustrative numbers.

All three are correct. They differ because each was pulled on a different day with a different window. Before anyone argues about which is right, write down the pull date, the window, the ad types and the exact dates behind each number. Most gaps close once those match. What is left is worth looking into.

Losing the early pulls

If you only keep the latest numbers, you can never see how far a week moved. You also cannot compare last week’s day-1 read with an older week’s day-1 read, because that older read is gone. A report pulled today shows today’s corrected numbers, not what the week looked like on day 1. Amazon’s version 3 reporting API also keeps campaign data for only 60 to 95 days, depending on ad type. A day-1 read nobody saved is lost.

How to compare ad weeks fairly

1. Pick one age and stick to it. For a weekly review, compare both weeks as they were read the same number of days after they ended. “Both weeks read 14 days after they ended” is the cleanest rule when Sponsored Brands or Display run.
2. For a fast read, use a window both weeks have finished. Through the API, Sponsored Products sales on a 1-day window count only purchases made within a day of the click. That window has closed for every day of last week, so weeks compare fairly on it. Amazon’s later corrections still apply. Standard Sponsored Brands and Display reports have no 1-day column, so show them separately and say so.
3. Label young weeks. Put “still counting” on any week younger than its longest window. Do not put a red arrow on its sales or ACoS.
4. Keep every pull. Save each week’s numbers each time you pull them, with the pull date. After a month you will know your own curve, such as “day 1 holds about 80% of final sales.” Then a young week can be read against that curve. Pull each closed month again a month later, as Amazon suggests.
5. Write the window on every number. Ad types, window, dates and pull date. Do this in every report, deck and agency export.
6. Judge profit on settled weeks. Set a week’s settled ACoS against your break-even ACoS. For TACoS (ad spend divided by all Amazon sales), remember that ad sales are an order-side number. The money Amazon pays you is in the settlement report, on its own clock.
7. Test a “real drop” on the settled part only. If clicks fell and 1-day sales fell too, the drop is likely real. If only the still-counting sales fell, it is age.

Synthesis labels every number with its period and caveats, such as “this month is still in flight,” and definitions are written once and shared by every answer. When two figures for the same metric and period disagree by more than 5%, the answer is flagged instead of one being quietly picked.