The Amazon ads attribution window keeps filling in for 7 to 14 days, so judge bids on closed days only.

The Amazon ads attribution window means your most recent days of ad data are not finished: sales keep arriving for 7 days on Sponsored Products and 14 days on Sponsored Brands and Display. So yesterday's ACoS always looks worse than it will end up. Cut bids on it and you cut keywords that were working. Judge ACoS only on days whose window has closed, and use clicks and cost per click for anything newer.

Most guides stop at the window lengths. This one shows what the window does to the numbers you act on: how much a fresh ACoS overstates the real one, why your tool and the console disagree, and how to read Black Friday when half its sales haven't shown up yet.

What the Amazon ads attribution window is

When a shopper clicks your ad and buys within the window, Amazon credits that sale to the ad. It uses last-click attribution: the last ad clicked gets the sale. The windows differ by ad type and account:

• Sponsored Products, sellers: 7 days after the click (Intentwise, 2023; Seller Labs, updated May 2026).
• Sponsored Brands and Sponsored Display: 14 days, for sellers and vendors (Intentwise, 2023; Helium 10, 2022).
• Vendors on any sponsored ad: 14 days. Amazon's own vendor advertising guide says a sale counts when a shopper who clicked buys within 14 days.

Anything bought after the window is not counted at all. Amazon says that share is not small. It studied more than 2 million campaigns that ran in June and found that around a fifth of a brand's value on Amazon arrives after the attribution window closes (Digiday, 1 October 2026). Amazon's measurement team called the window "just a cutoff date for measurement".

Sales land on the click date, so recent days look worse

Here is the part that changes decisions. In campaign reports, sales are credited to the date of the click, not the date of the sale. A shopper who clicks on Monday and buys on Thursday adds a sale to Monday. Optmyzr notes that these restatements reach back through the whole window (April 2026).

Two smaller delays sit on top. Attributed sales can take up to 12 hours to appear, and cancelled orders and invalid clicks can take up to 72 hours to be removed (Helium 10; BidX, 2021). So spend for a day is close to final by the next morning. Sales for that day are not.

Take an example. A Sponsored Products campaign spends $100 a day at $1.00 a click, so 100 clicks. Once its 7-day window closes, each day ends up with $400 in sales: a 25% ACoS. Your target is 30%. Now assume this share of each day's final sales has shown up by the morning of:

• Day 1: 70%, so $280 in sales and a 35.7% ACoS.
• Day 2: 78%, $312 and 32.1%.
• Day 3: 85%, $340 and 29.4%.
• Day 4: 90%, $360 and 27.8%.
• Day 5: 94%, $376 and 26.6%.
• Day 6: 97%, $388 and 25.8%.
• Day 7: 100%, $400 and 25%.

Read on day 1, this campaign is 5.7 points over target, and any rule that cuts bids above 30% cuts it. Read on day 7, it is 5 points under target and deserves more budget. Same clicks, same shoppers. Only the reading date changed.

The rule: judge ACoS only on days whose attribution window has closed: 7 days back for Sponsored Products, 14 for Sponsored Brands and Display. For anything newer, judge clicks and cost per click, which are final within a day or so.

Why your tool, your console and your Sponsored Brands disagree

The window also explains most "the numbers don't match" questions.

• Sponsored Brands looks better than Sponsored Products. Sometimes it really is. But a Sponsored Brands click gets 14 days to earn a sale and a seller's Sponsored Products click gets 7. Compare the two side by side and the longer window gets credit for sales the shorter one would have missed. Weigh that before you shift budget, and see our guide to Sponsored Brands metrics for what else to compare.
• Your tool shows a lower ACoS than the console. Some tools use their own window. Seller Labs' Ad Genius, for example, reports on a 30-day window while Seller Central uses 7. A longer window catches more sales, so its ACoS is lower. Neither is wrong; just never set a target in one and check it in the other.
• Ad sales and Business Reports don't line up by day. Ad reports sit on the click date; your orders sit on the order date. A strong ad day can show up as a strong sales day later in the week.

If you need a longer view for one analysis, Amazon Marketing Cloud lets you set a lookback window of up to 28 days and pick other attribution models (Amazon, 2022). For day-to-day bids, one consistent window matters more than the longest one.

Black Friday on half-filled data

Black Friday is on 27 November 2026 and Cyber Monday on 30 November, as set out in our Q4 2026 Black Friday and Cyber Monday plan. That week is where the window does the most damage, because it is when sellers check their numbers most often.

On the Saturday morning, Black Friday's spend is nearly complete and its sales are not. Shoppers who clicked on Friday and bought on Cyber Monday are still to come. A Sponsored Products click from Cyber Monday can earn a sale until 7 December, so its numbers are only complete around 8 December. For Sponsored Brands and Display, that moves to about 15 December.

So during the deal days, steer by what is already final. Take sales per click from your last closed weeks and multiply by your target ACoS: that is the most a click can cost. In the example, $400 from 100 clicks is $4.00 a click, and $4.00 × 30% = $1.20. If deal-day cost per click runs above that, lower the bid now. If it runs below, leave the keyword alone, however high its fresh ACoS looks. Our guide to break-even ACoS shows how to set the target itself.

What to do this week

1. Shift your review range back. For Sponsored Products, end it 7 days before today; for Sponsored Brands and Display, 14 days. Change any automatic rule that reads "last 7 days" to read a closed range instead.
2. Write down which window each number uses: the console, your tool, and any report you send a client. Set targets in the same window you check them in.
3. Set a cost-per-click cap per campaign from your last closed weeks: sales per click × target ACoS. Use it for the days ACoS can't judge yet.
4. Don't cut a deal-day keyword on its fresh ACoS. Judge Black Friday after about 5 December and Cyber Monday after about 8 December on Sponsored Products.
5. After any bid change, wait one full window before you call it a win or a loss. Our post on what ACoS measures covers the rest of the reading.

If you want the bids handled: the Mirox 14-day free trial runs live on up to 5 of your products in 1 marketplace, on campaigns Mirox builds for them, so your existing campaigns stay untouched. Every bid change is saved with the numbers behind it and the options that were ruled out, so you can check what each decision was based on.

Common questions

What is the attribution window for Amazon ads?

For sellers, Sponsored Products credits sales made within 7 days of an ad click. Sponsored Brands and Sponsored Display use 14 days, and vendors get 14 days on Sponsored Products too.

Why does my Amazon ACoS change after a few days?

Amazon credits a sale to the day of the click, not the day of the purchase. Shoppers who buy a few days later add sales to days that already passed, so recent ACoS drops as the window fills in. Cancelled orders are removed within 72 hours, which can move it the other way.

Can I change the attribution window in Amazon Ads?

Not in the standard campaign reports. Amazon Marketing Cloud lets advertisers set a custom lookback window of up to 28 days and choose other attribution models, but that takes a separate setup.

How long should I wait before judging an Amazon PPC change?

At least one full window after the last day you want to judge: 7 days for Sponsored Products, 14 days for Sponsored Brands and Sponsored Display. Before that, judge clicks and cost per click, not ACoS.

The one-line version

A day's ACoS is a draft until its window closes, so judge bids on closed days and steer the fresh ones by cost per click.

Want to see the numbers behind a single bid? Read how transparency works, then see how ACoS and ROAS describe the same result in ROAS vs ACoS.