Why Your Amazon Advertising Attribution Is Lying to You And How Agencies Read the Truth

A brand pulls its Sponsored Products report a few days into a new campaign, sees an alarming ACoS, and kills it before the attributed conversions have had time to arrive within the lookback window. This happens constantly, and it is not a data problem. It is a reading problem. Amazon gives you an enormous amount of advertising data, and much of it does not mean what a first glance suggests.

The reports are not wrong. They are provisional, time-shifted, last-touch, and scoped in ways that are easy to misread and expensive to act on. The single most important thing to understand is this: Amazon attribution can tell you which eligible interaction received credit under its methodology; it cannot, by itself, tell you whether that interaction caused the sale. Everything below follows from that distinction.

Amazon's current attribution documentation states that conversions may be reflected on the date of the shopper's ad interaction, which can differ from the actual conversion date. A sale showing under March 3rd means an ad interaction occurred on March 3rd; the purchase itself could have landed later, inside the lookback window that follows.

That mechanic breaks the instinct every DIY advertiser has, which is to judge a campaign by how the last few days look. The newest dates are the least mature data in the account, because the window on those interactions has not closed yet. A campaign can look inefficient in its first few days and improve as additional attributed conversions arrive within the applicable lookback window. Treating fresh data as if it were final is the most common and most costly attribution mistake on the platform.

The figures you pulled recently are not fully settled. Recent attribution data is incomplete until the applicable lookback window has closed, and Amazon notes that conversions can take time to appear in reporting. So the newest dates in a campaign should not be treated as equally mature as older dates.

The practical discipline is to use the applicable campaign lookback window before making final performance judgments, and to weight decisions toward date ranges that are already settled. Making bid changes, pausing keywords, or killing campaigns off data that has not finished maturing is acting on a number before it has finished being a number.

Not all your campaigns are measured on the same clock, and this quietly warps cross-campaign comparisons. Sponsored Products uses a 7-day click window for Seller Central accounts and 14 days for Vendor Central. Sponsored Brands and Sponsored Display generally use 14 days, and Amazon DSP can use flexible lookback windows. That means a Sponsored Brands campaign is capturing sales up to two weeks after the interaction while a seller's Sponsored Products campaign stops counting after one, so Sponsored Brands can look more efficient purely because it is allowed a longer lookback. Compare their ACoS side by side without accounting for the window, and you may be reading the window rather than the performance.

On January 1, 2026, Amazon introduced a shopping-signal enhanced last-touch attribution model for eligible Amazon Store view-attributed campaigns, including Sponsored Brands and Sponsored Display campaigns billed on a viewable impression (vCPM) basis and Amazon DSP campaigns serving ads in the Store. The new standard methodology applies a shorter view-attribution window and focuses credit on brand-discovery moments, such as exploratory browsing or general category searches. Click-based attribution remained unchanged, and eligible 14-day "all views" metrics remain available separately through Unified Reporting. See Amazon's view-attribution update.

The diagnostic that matters: Amazon itself says advertisers may see a change in conversion metrics from this update. So if view-attributed ROAS changes materially across the 2025 to 2026 boundary, check the attribution-methodology change before concluding that campaign performance itself changed. The number can move without the underlying sales moving.

Amazon's standard reporting uses last-touch attribution, with product relevance considered and clicks generally prioritized over views. A conversion receives credit for no more than one eligible ad interaction in standard attribution. Whichever eligible interaction is last takes the credit, with no splitting across the shopper's earlier touches.

This is where the 2026 information gain lives. Amazon now also provides multi-touch attribution metrics for eligible U.S. advertisers across Amazon DSP, Sponsored Display, Sponsored Brands, and Sponsored TV, displayed alongside the familiar last-touch metrics. Multi-touch divides conversion credit across contributing Amazon Ads touchpoints in proportion to their likely influence, which lets teams compare standard last-touch reporting against a model that recognizes the awareness ads that built interest earlier in the journey. For an article about reading past the surface number, this is the tool that does it. See Amazon's multi-touch attribution announcement.

Branded campaigns can receive credit for shoppers who already had strong purchase intent. That creates a cannibalization risk: attributed sales can include conversions that might have occurred organically. Standard attribution reporting cannot tell you the counterfactual, so a comfortable branded ACoS can be flattering the account rather than proving the ads created demand.

The opposite distortion is just as real. Advertising can influence shopping behavior beyond the conversions captured inside the standard attribution window or assigned to the final eligible touchpoint. A shopper may encounter an ad, leave, and return later through organic or direct visits that standard campaign reporting does not connect to the ad. For that reason, standard campaign reporting should not be treated as a complete causal estimate of advertising's incremental contribution.

Put those together and the honest framing is this: attributed revenue is a measurement output, not a direct estimate of incremental revenue. It can miss some upstream influence while also crediting conversions that may have happened without the ad. Reading it well means holding both possibilities at once.

The interpretive work is what separates reading the numbers from understanding them. A few of the moves that matter most:

Put spend in context with TACoS. Total Advertising Cost of Sales puts ad spend in the context of total Amazon revenue, organic included. A falling TACoS alongside stable or growing revenue can indicate that organic sales are becoming a larger part of the mix, but TACoS alone does not prove that advertising caused an organic-ranking lift.

Wait for the window to close. Let the lookback window mature and weight decisions toward date ranges that are already settled rather than optimizing on the newest, least mature dates.

Compare like windows to like windows. Normalize for the ad-type window differences rather than ranking campaigns whose clocks run at different speeds, and read the purchased-product report to separate the advertised ASIN from brand-halo credit.

Separate branded from non-branded. Because last-touch can inflate branded and defensive campaigns with existing demand, read those separately from prospecting campaigns rather than letting a branded ACoS flatter the whole account.

Read multi-touch alongside last-touch. Where available, compare the two views to see the value of ads that appeared earlier in the journey, which last-touch alone hides.

None of this is exotic. It is the difference between treating the console as a scoreboard and treating it as a set of signals that need interpretation, which is exactly the discipline behind how Amazon PPC management should be run.

Amazon's advertising data is not dishonest. It is provisional, time-shifted, last-touch, and scoped by rules that vary from one campaign type to the next, and every one of those properties makes it easy to misread at a glance. The brands that lose money to their own reports are the ones treating fresh, unsettled, last-touch numbers as final verdicts. The brands that win are the ones asking what the data is not showing before they touch a bid, and reading credit as credit rather than as proof of cause.

If your advertising decisions are being made off the raw console view, there is a good chance some of your best campaigns look worse than they are and some of your weakest look better. CMO's Amazon PPC management reads the attribution correctly and manages spend against total revenue rather than the surface number.