Amazon just rewrote the rules of attribution, and most advertisers didn’t even realize it.
On January 1, 2026, the platform quietly retired the “blind” 14-day view-through window that has shaped campaign reporting for years. In its place, Amazon introduced a machine learning–driven attribution model designed to answer a much harder question: Did this ad actually influence the purchase?
The change applies across Amazon Store ads, including Amazon DSP, Sponsored Brands, and Sponsored Display, shifting attribution away from broad impression credit and toward signals that indicate real shopper intent. Offsite DSP delivery remains the exception, continuing to operate under the traditional 14-day click / 14-day view lookback window.
If your dashboards suddenly look different, it’s not because shoppers changed their behavior overnight. It’s because Amazon is getting more selective about what it counts as influence.
And just to make the measurement landscape even more interesting, Amazon launched a Multi-Touch Attribution beta at roughly the same time.
Both updates rely on machine learning, but they’re solving very different problems, and understanding that distinction is critical if you want to interpret your performance data correctly.
Two Attribution Changes. Two Very Different Jobs.
The current confusion isn’t surprising. Amazon rolled out two attribution updates at nearly the same time, and both are powered by machine learning.
But they’re tackling entirely different issues. One improves the accuracy of last-touch attribution, while the other introduces a new way to distribute credit across multiple touchpoints with Multi-Touch Attribution. We’re breaking them down below, plus what advertisers need to do right now to stay ahead of the changes.
The New Last-Touch Attribution Model (Quality Control)
The change that took effect on January 1, 2026 is Amazon’s new shopping-signal enhanced last-touch attribution model. Click attribution remains unchanged, but this update specifically targets view-through attribution.
Historically, Amazon used a 14-day click / 14-day view attribution window. Under the previous model, if a shopper simply saw your ad and purchased within two weeks, that conversion could be credited to DSP, even if the ad view had little real influence on the purchase.
The new model replaces that broad time-based window with a machine learning evaluation of whether the ad view actually contributed to the sale.
Instead of asking: “Did the shopper see the ad within 14 days?”
The system now asks: “Did the shopper see the ad at a moment when they were discovering this brand or category?”
The algorithm evaluates signals like:
- exploratory browsing behavior
- general category searches
- historical shopping patterns
- signals of product discovery vs. intent to purchase
If the system determines a shopper was already likely to purchase, the ad view may no longer receive credit.
The result? Fewer purchases qualify as attributed conversions, which means DSP revenue attribution appears lower and ROAS may appear lower, but, actual sales remain unchanged.
This is why many advertisers are seeing double-digit declines in attributed DSP revenue in 2026 reporting.
Let’s be clear: It’s not a performance drop; it’s a stricter judge of influence.
The Metric That Preserves Your Historical Benchmark
Because the new model removes many legacy view-through conversions, Amazon also introduced a new metric to maintain historical comparisons:
Purchases (All Views)
This metric preserves the original 14-day view methodology, counting all ad views within that window regardless of ML filtering.
It’s available in:
- Amazon Ads Unified Reporting
- the Reporting (beta) tab in the DSP console
Using this metric allows advertisers to create true apples-to-apples year-over-year comparisons with pre-2026 performance.
Without it, DSP performance in 2026 will almost always appear artificially lower than in 2025.
The Other Attribution Update: Multi-Touch Attribution (Beta)
At roughly the same time Amazon tightened its standard attribution model, it also launched something new: Multi-touch Attribution (MTA), currently in beta for US advertisers.
While the new last-touch model makes attribution stricter, the multi-touch model does something entirely different: it distributes credit across the full advertising journey.
Instead of assigning 100% of the conversion to the final ad interaction, the model distributes credit across the entire advertising journey.
For example, a purchase might be credited as:
- 0.3 conversions to a DSP display ad
- 0.7 conversions to a Sponsored Products click
This helps quantify the role of upper-funnel media that builds awareness before a shopper is ready to purchase.
Metrics like Orders (multi-touch) and ROAS (multi-touch) are now appearing alongside standard reporting metrics in the Amazon Ads console and DSP Manager.
What Advertisers Should Do Right Now
Panicking? Here are three practical steps advertisers should take right now to stay ahead of the changes.
1. Establish a True Year-Over-Year Benchmark
Pull Purchases (All Views) from the Amazon Ads reporting console and use it as your baseline for comparing 2026 to historical performance. This restores continuity with the legacy 14-day view methodology.
2. Prepare Your Teams for Attribution Questions
The reporting change is large enough that brand-side teams will start noticing sudden declines in DSP revenue attribution. Without context, those numbers look like a performance problem. In most cases, they’re simply a measurement change. The best place to be is explaining the shift before someone else flags it.
3. Start Watching Multi-Touch Metrics
As Amazon rolls out Multi-touch Attribution beta metrics more broadly, advertisers should begin comparing:
- Standard ROAS
- Multi-touch ROAS
If multi-touch ROAS is significantly higher, it’s a strong signal that upper-funnel DSP media is contributing more value than last-touch reporting shows.
The Bigger Picture
Amazon is moving away from fixed attribution windows and toward machine-learning valuation of ad influence.
In the short term, that means:
- stricter conversion credit
- lower reported DSP numbers
- more complex measurement models
But it also means the industry is moving toward a clearer answer to the question advertisers actually care about: Did this ad truly influence the purchase?
Shopper behavior didn’t suddenly change on January 1; the measurement did. Advertisers who recognize that shift will interpret their data more accurately, defend DSP performance more confidently, and make smarter decisions about how Amazon media actually drives growth.