Amazon Ads pushes year-long attribution window to reclaim credit

Full analysis

Amazon Ads says roughly 20% of the sales its ads generate never gets counted, because the standard window that credits a sale to an ad runs only 7 to 30 days. So Amazon is building tools to stretch that window to 12 months, link ad views to off-Amazon and in-store sales, and push budgets toward its DSP. Amazon's Lily Tong put it plainly: "roughly 20% of the sales value for a brand shows up over time and isn't captured in that immediate attribution window."

What's actually being decided here isn't Amazon's. It's whether advertisers and their agencies let a seller grade its own homework on a longer timeline. That choice is easy to undo at the config level and very hard to undo at the planning level. Once a 12-month Amazon metric anchors a CPG media plan, pulling it back out means re-litigating two years of budget decisions. No hard deadline, but the closed beta sets the clock: early adopters bank benchmark data before the metric generalizes.

The Market Analyst

Amazon is doing what every walled garden does when its core ad product matures: move up the stack from selling inventory to defining the yardstick. Search budgets are under pressure from AI shopping and social platforms grabbing attribution credit, so Amazon wants its DSP to be the planning anchor. The buying tool is already built; the scoreboard is what's left to own. The asset nobody else has is the cart. The Trade Desk can buy inventory everywhere and still never see the purchase. If Amazon's omnichannel metric becomes the default for retail categories, Criteo and Walmart Connect either build comparable closed-loop attribution or lose the strategic narrative. In plain terms: Amazon is trying to own the scoreboard, and owning the scoreboard is worth more than winning any single game.

The Skeptic

Twenty percent of your own value is uncredited, and you found it in a study only you can run. That's the oldest move in walled-garden math: widen the window until the number flatters you, then call it research. The problem is right there in what the agencies at Power Digital and Moburst are already flagging. Amazon hands over 90 days of historical sales. A marketing mix model, the statistical tool that ties spend to revenue across channels, needs about 18 months. You cannot validate a 12-month claim with a 90-day export. And streaming TV uplift that doesn't line up with actual weekly product sales isn't measurement, it's a number with no check against reality. Longer windows always find more credit. That's arithmetic, not insight.

The Operator

The config change is trivial. The reconciliation fight is not. Full-funnel DSP tools that optimize for a 12-month signal will pull budget from channels that only report on a 30-day window. So your short-term performance dashboards go sideways while Amazon's long-game number climbs. Guess which one finance trusts less. The real break shows up around 90 days, when the team sits down to reconcile Amazon's "long-term sales lift" against the MMM output and the two numbers refuse to agree. Now one media dollar generates two ROI stories, and nobody can say which is true. The teams that audit their attribution setup now at least know where the seams are. The teams that adopt first and reconcile later own the mess.

The Customer / End User

Nobody at a CPG or DTC brand woke up asking Amazon to extend their attribution window. They woke up asking why their blended ROI keeps drifting. There's a real pain here: deferred sales are genuine, new-to-brand shoppers do buy again, and a 7-day window does undercount that. So the brand that's been starving its upper-funnel spend because it only sees 30-day returns has a legitimate reason to listen. But the brand is being asked to trust Amazon's count of value Amazon can't let anyone else verify. That's the trade. More credit for your ad dollar, less ability to check the math. For a category manager under pressure to defend spend, the extra credit is tempting and the unverifiability is the part that bites in the next budget review.

The CFO

Follow the incentive. Amazon's long-term metric is designed to move more budget onto Amazon's DSP, and every point of attribution Amazon reclaims is a point it takes from search, social, or a rival retail network. That's not neutral. The cost nobody's pricing is the planning risk: if you re-anchor a media plan on a 12-month number and that number later fails third-party reconciliation, you've committed real spend on a signal you can't defend to your own board. The deferred value might be real. I'd believe it when someone outside Amazon can replicate it. Until then, I treat the 20% as a sales argument with a study attached, and I don't let it set my allocation.

The tensions

Two disagreements matter. First, the Market Analyst and the Skeptic split on whether this is strength or sleight. The Analyst says Amazon owns the one signal nobody can fake, the cart, so the metric has real teeth. The Skeptic says a metric only Amazon can compute is exactly the problem, teeth or not. Both are right, which is the trap: the data advantage is genuine AND unverifiable at the same time.

Second, the Customer and the CFO part ways on the deferred value itself. The Customer grants that 7-day windows undercount real repeat buyers and wants the credit. The CFO grants the same and still won't let it drive allocation until an outside party confirms the number. The gap between "this is probably true" and "I'll bet budget on it" is the whole decision.

What it hinges on

Three things. Whether the deferred value is real (probably yes, partially). Whether anyone outside Amazon can verify Amazon's version of it (today, no). And whether agencies cave before independent reconciliation exists (the pressure is on, and the closed beta rewards caving early).

The council leans skeptical on adoption and bullish on Amazon's position. The data moat is real. The measurement claim is unproven. Before committing: run Amazon's long-term metric in parallel against your own MMM for at least two quarters, treat the 20% as a hypothesis not a given, and refuse to re-anchor allocation until the streaming uplift reconciles against syndicated weekly sales. Let Amazon's number inform the plan. Don't let it own the plan.

Prediction: Through the end of 2027, no independent measurement provider (Nielsen, Comscore, VideoAmp, or an MRC accreditation) will validate Amazon's 12-month long-term sales metric, and it will remain a closed-loop number only Amazon can compute, settled by whether any third party publishes a reconciliation by the 2028 upfront planning cycle.

Confidence: Medium. The data that proves the claim is the data Amazon won't share.

Why: The whole value of Amazon's metric to Amazon is that it links ad exposure to Amazon's own cart data, which no outside party can see, so opening it to independent validation would hand rivals the one signal that makes Amazon's DSP special. Agencies already flag that Amazon supplies 90 days of history while an MMM needs about 18 months, and that streaming uplift doesn't reconcile with weekly sales, which means the reconciliation infrastructure doesn't exist and Amazon has no incentive to build it. The opposite outcome, Amazon submitting the metric for third-party accreditation, would require Amazon to trade its data moat for credibility it can win more cheaply by just shipping the tools and letting adoption create the standard. Walled gardens standardize by distribution, not by accreditation.

Revisit by 2028-02-01: We're right if no third-party measurement firm has published an independent validation or reconciliation of Amazon's 12-month long-term sales metric by then. We're wrong if Nielsen, Comscore, VideoAmp, or the MRC issues one.