Amazon DSP Attribution: Windows, Models and 2026 Changes

Amazon DSP attribution is the set of rules Amazon uses to decide which DSP ad impression or click gets credit for a shopper action such as a detail page view, an add to cart or a purchase. By default, Amazon DSP uses a 14 day, last touch model that credits both clicks and viewable impressions, prefers clicks over views, prefers the promoted product over other products from the same brand, and credits each conversion to only one ad interaction across every campaign and every advertiser.

That one paragraph hides a lot of detail that changes how you read your ROAS. Since January 2026, some view conversions follow a stricter model. Off Amazon conversions can be modeled. Multi touch metrics now sit next to last touch metrics in the DSP console. And sales can appear in your reports on a different date from when they actually happened.

This guide walks through each rule with worked examples and formulas, so you can explain every number in your DSP report to a finance team or a client. If you are new to the platform itself, start with our primer on how the Amazon demand side platform works.

Key Takeaways

• Amazon DSP credits conversions that happen within 14 days of an eligible click or viewable impression.
  
  

• When a shopper touches several ads, Amazon ranks them in this order: promoted product clicks, brand halo clicks, promoted product views, brand halo views. The most recent interaction in the winning tier gets the credit.
  
  

• A conversion is credited once, across all Amazon Ads campaigns and advertisers. DSP competes with your Sponsored Ads for the same sale.
  
  

• Sales are reported on the date of the ad interaction, not the purchase date, so the last 14 days of any report are still filling in.
  
  

• From January 1, 2026, view attribution for DSP ads on Amazon Store inventory uses shopping signals and a shorter window. The legacy 14 day view numbers live on in Purchases (all views) metrics.
  
  

• For the full picture, pair last touch reports with multi touch metrics, conversion path reporting, Amazon Marketing Cloud (AMC) and incrementality tests.
  
  

What Is Amazon DSP Attribution?

Amazon defines attribution as the measurement of conversion events that occur after shoppers interact with ads. For Amazon DSP, those conversion events fall into two families:

1. On Amazon conversions: purchases, units, sales, new to brand purchases, detail page views, add to cart, add to list, branded searches, Subscribe and Save subscriptions and product review page views.
   
   

2. Off Amazon conversions: actions on your own website or app, such as a sign up, a lead form or a checkout, measured through Amazon's tags or server side event import.

Attribution matters more on DSP than on search ads for one simple reason. Most DSP impressions are never clicked. A streaming TV ad, an online video ad or a display banner on a news site does its job by being seen. If you judge DSP only by clicks, almost every upper funnel campaign looks like a failure. If you accept every view as the cause of a sale, almost every remarketing campaign looks like a genius move. Good DSP attribution analysis lives between those two extremes

How Amazon DSP Attribution Works: The Three Rules

Every DSP conversion passes three tests before it is credited to a campaign. Amazon describes them in its ad campaign attribution documentation.

Rule 1: Product relevance (promoted vs brand halo)

When you build a DSP order, you add the products you want to track. Those tracked ASINs are your promoted products. DSP managing can also claim credit for brand halo products, which Amazon defines as products within the same brand and subcategory as a promoted ASIN.

Example: You promote a pair of over ear headphones. A shopper sees your ad, then buys your wireless earbuds from the same brand. If the earbuds sit in the same subcategory (Headphones), that purchase can be credited to your campaign as a brand halo conversion. A charging cable from the same brand in a different subcategory would not count as halo for that campaign.

In Demand Side Platform reporting, halo metrics are labeled "brand halo" (often abbreviated BH), and Total metrics combine promoted and brand halo results.

Practical tip: Track every ASIN you genuinely want the order to drive. If a product is left off the tracked list, its sales can only show up as halo, and only when it shares a subcategory with a tracked product. Missing ASINs are one of the most common reasons DSP looks weaker than it is.

Rule 2: Ad interaction type (clicks and views)

Unlike click only ad types, all Amazon DSP campaigns attribute conversions to both clicks and views. A view counts only when the impression met Amazon's viewability standard: at least 50% of the ad's pixels on screen for 1 second or more for display ads, or 2 seconds or more for video ads. Where viewability cannot be measured directly, Amazon infers it from historical data and placement characteristics.

Standard DSP metrics such as Purchases and ROAS include both click and view conversions. Click only breakouts, such as "ROAS from clicks" or "Purchases from clicks", are labeled separately.

Rule 3: The lookback window

The lookback window is the maximum number of days a conversion is still considered related to an ad interaction. For Amazon DSP, the standard window is 14 days for clicks and views. The 2026 change for some view conversions is covered below.

The Amazon DSP Attribution Window, With Examples

Amazon's own illustration makes it clear: with a 14 day lookback, a purchase on October 17 can be credited to ad interactions between October 3 and October 17. An impression on October 2 is out of range.

Sales Show up on the Ad Interaction Date

This is the rule that confuses most teams. Amazon reports attributed conversions on the date of the ad interaction, not the date of the purchase. Conversions can also take up to 12 hours to appear, and the numbers for any report date stay incomplete until that date's lookback window closes.

Example: A shopper sees your DSP video ad on September 1 and buys on September 12. The sale appears in your September 1 row. If you pulled your report on September 5, that sale was not in it yet. Pull the same report on September 20 and September 1 suddenly looks better.

On top of that, payment failures and orders cancelled within 72 hours are deducted from attribution metrics, and those deductions can take up to 7 days to process. So a report date is truly settled only about three weeks after it happened.

Formula: Project final ROAS from immature data

Because recent days are always underreported, early ROAS readings are biased downward. A simple fix is to build your own "attribution maturity curve" from past campaigns.

Completeness factor at day n = Attributed sales for a report date as of day n ÷ Attributed sales for the same date once settled (day 21)

Projected final ROAS = Current ROAS ÷ Completeness factor

Example: Looking back at your last three months of remarketing data, a report date typically holds 55% of its final sales after 3 days, 80% after 7 days and 100% after 21 days. Today your 3 day old data shows ROAS of 2.2.

Projected final ROAS = 2.2 ÷ 0.55 = 4.0

The curve is different for every account. Prospecting and streaming TV usually fill in more slowly than remarketing, because shoppers exposed earlier in the funnel take longer to buy. Calculate one curve per tactic, and you will stop pausing good ad groups on day three.

How Ads Compete for Credit: The Attribution Priority Ladder

Amazon DSP does not live in its own Amazon attribution bubble. DSP, Sponsored Brands, Sponsored Display, Streaming TV and (for vendors) Sponsored Products all compete for the same conversion. According to Amazon's documentation, when a shopper interacted with more than one eligible ad, Amazon first considers product relevance, then uses a last touch model that prioritizes the last click over the last view.

A conversion is credited to no more than one ad interaction across all campaigns and all advertisers. The same sale cannot be counted as both promoted and halo, or as both a view and a click.

Worked Example: How Amazon DSP Attributes a Purchase

A shopper interacts with a headphone brand's ads over two weeks:

Many people expect the October 12 remarketing view to win, because it came last. It does not. There is no promoted click, so the ladder moves to tier 2, and the brand halo click on October 8 takes the credit. The consideration campaign gets the sale, the remarketing campaign gets nothing, and the streaming TV ad, which may have started the whole journey, gets nothing either.

This is exactly why single touch reports undervalue awareness tactics, and why the multi touch tools covered later exist.

Note for sellers running DSP

Sellers who also run Sponsored Products should enter their merchant token in their Amazon DSP account configuration. Amazon uses it to make sure the same advertiser is not credited twice for one conversion, once as a merchant and once as a brand.

The 2026 Amazon DSP Attribution Change: Shopping Signal View Attribution

If your DSP ROAS dropped at the start of 2026 without an obvious cause, this is the most likely reason.

Starting January 1, 2026, Amazon introduced a shopping signal enhanced last touch model for view based attribution on Amazon Store inventory. It applies to Amazon DSP ads served on Store inventory, along with vCPM Sponsored Brands and Sponsored Display. Instead of crediting any viewable impression inside 14 days, the model credits views that reached shoppers during discovery moments, such as exploratory browsing or general category searches, and it uses a shorter window. Click attribution did not change, and neither did other inventory types.

What that means in practice:

• Standard DSP purchase, sales and ROAS metrics now use the new model for affected views.

• Purchases (all views) metrics keep the old rules, crediting all eligible viewable impressions within 14 days. Use them for year over year comparisons.

• Actual sales did not change. Only the share of them credited to views did.

Formula: Measure your view attribution gap

View attribution gap % = (Purchases (all views) minus Standard purchases) ÷ Purchases (all views) × 100

Example: Last month a remarketing ad group shows 1,000 Purchases (all views) and 780 standard Purchases.

(1,000 minus 780) ÷ 1,000 × 100 = 22%

So 22% of the view conversions the old model would have credited no longer qualify. When you compare 2026 against 2025, compare Purchases (all views) in 2026 against standard Purchases in 2025. Otherwise you are comparing two different rulebooks and blaming the campaign for a measurement change.

A large gap is also a useful signal on its own. Ad groups that lose the most credit under the new model were leaning heavily on passive views of shoppers who were already going to buy. Those are exactly the ad groups to put through an incrementality test.

Amazon DSP Attribution Metrics and the Formulas Behind Them

Your DSP reports contain dozens of conversion metrics. These are the ones that matter most for attribution analysis.

A few details from Amazon's attribution documentation that change how you read these numbers:

• Attributed sales use the unit price shown in the cart. Many cart level discounts, such as coupons, lightning deals and some tentpole event pricing, are not subtracted, and tax is excluded. During Prime Day or Black Friday, attributed sales can therefore run above your real net revenue.
  
  

• Depending on the campaign type, attributed sales can include units sold by Amazon or by another seller, and even used condition items. If a reseller holds the Buy Box on your ASIN, your DSP report may be crediting sales you did not earn.
  
  

• In reports grouped by advertised ASIN, the ASIN is the one in the ad, but the attributed sales can include other ASINs.
  
  

For a deeper look at the efficiency metrics themselves, see our guide on how to calculate Amazon ROAS.

Off Amazon Conversion Attribution in Amazon DSP

Amazon DSP is one of the few Amazon ad products that can send traffic to a brand's own website or app and measure what happens there. Brands selling direct to consumer, service businesses and app publishers rely on this.

How off Amazon conversions are tracked

• Amazon Ad Tag: a JavaScript tag placed on your site that fires on events such as page views, sign ups, add to cart and checkout. It replaced the older DSP pixels.
  
  

• Server side event import: Amazon's Conversions API lets you send conversion events from your server, which is more resilient to browser restrictions and ad blockers.
  
  

• Offline events: in store or CRM conversions can be uploaded and matched in a privacy safe way.
  
  

Once events are flowing, you pick which ones matter in the DSP order, and they appear as off Amazon conversions in campaign reporting.

Modeled conversions

Not every off Amazon conversion can be linked to an ad interaction. Some impressions serve on anonymous supply, and some browsers break the link between an ad view and a later site visit. Since August 2024, Amazon DSP has used modeled attribution for off Amazon conversions for US advertisers. Directly measured and modeled conversions are reported together as one combined result in the campaign manager, the reporting center and the API.

Two practical consequences:

1. You cannot split modeled from measured conversions in the standard metric, so compare trends in DSP against your own site analytics rather than expecting a line by line match.
   
   

2. Modeled conversion rates settle over time. Give a new campaign one to two weeks before judging its off Amazon performance.

Formula: Reconcile DSP with your own analytics

Capture ratio = DSP attributed off Amazon conversions ÷ Conversions your own analytics credits to the DSP campaign (by UTM or landing page)

Example: DSP reports 420 sign ups in a month. Your analytics tool credits 300 sign ups to the DSP landing page by last click. The capture ratio is 1.4. That gap is expected, because DSP counts views and modeled conversions while most site analytics count only clicks. Watch the ratio over time. If it jumps from 1.4 to 3.0 without a change in strategy, check for a broken tag, a duplicate event or a change in supply.

Amazon Attribution vs Amazon DSP Attribution

Search results for "Amazon DSP attribution" often mix in Amazon Attribution, which is a different product. The difference matters when you are setting up measurement.

In short: use DSP attribution to evaluate DSP. Use Amazon Attribution to evaluate traffic you send to Amazon from other channels. Our Amazon Attribution guide covers the second one in detail.

Beyond Last Touch: Advanced Amazon DSP Measurement

Last touch is a convention, not a measure of cause. Amazon's own science team makes the point well: in its paper on Amazon Ads multi touch attribution, it cites a large study of roughly 2,000 Meta advertising campaigns in which purely observational models misestimated ad effects by 488% to 948%, depending on the model and outcome. That is the risk of trusting any attribution model that is not checked against experiments.

Here are the tools Amazon offers to go further.

Multi touch attribution in the DSP console

Amazon's multi touch attribution divides credit for a purchase across the Amazon Ads touchpoints a shopper saw, in proportion to their likely contribution. It combines randomized experiments, machine learning and shopping signals. For US advertisers, metrics such as Purchases (multi touch), Sales (multi touch) and ROAS (multi touch) appear next to the standard last touch metrics in the Amazon DSP campaign manager and in DSP custom reports.

How to use it: Compare each campaign's multi touch sales with its last touch sales.

Credit shift ratio = Multi touch sales ÷ Last touch sales

A ratio above 1 means the campaign is undervalued by last touch (typical for streaming TV, video and prospecting). A ratio below 1 means it is overvalued (typical for remarketing that catches shoppers at the finish line). If your streaming TV campaign shows $8,000 last touch sales and $20,000 multi touch sales, its ratio is 2.5. Before cutting it, look at that number. Read more in our guide to multi touch attribution.

Conversion path reporting

Conversion path reporting shows the most common sequences of ad touchpoints in the 30 days before a purchase, across Sponsored Ads and Amazon DSP, ranked by sales. Use it to spot which DSP tactics tend to appear early in winning paths, even when they rarely take last touch credit.

Amazon Marketing Cloud

Amazon Marketing Cloud is Amazon's clean room. It gives you event level, anonymized data so you can write your own attribution logic in SQL: custom lookback windows (including windows longer than 14 days for considered purchases), first touch or position based models, frequency to conversion analysis and exposed vs unexposed comparisons. Our Amazon Marketing Cloud guide explains how to get started.

Omnichannel Metrics

For brands sold at other retailers too, Omnichannel Metrics estimates how Amazon DSP exposure drives sales beyond Amazon. Since July 2026, US advertisers on self service and managed service DSP can see retailer level breakouts of units and sales, based on Amazon Shopper Panel data and third party signals.

Long term sales

Amazon's long term sales metric estimates the sales a brand can expect over the next year based on how well a campaign moves new to brand shoppers down the funnel. It is useful for justifying acquisition focused DSP campaigns whose 14 day ROAS looks thin.

How to Audit Your Amazon DSP Attribution: A Five Step Framework

Reported ROAS in DSP can look extraordinary. Remarketing ad groups reporting ROAS of 20, 50 or even more than 100 are not unusual. Those numbers are rarely wrong in a technical sense. They are simply answering a different question from the one your CFO is asking. Run these five checks on every DSP account before you scale it.

Step 1: Check view share

View share = View attributed purchases ÷ Total purchases

If an ad group shows a view share above 85% and it targets past detail page viewers or past purchasers, most of its credit comes from shoppers who saw an ad and bought anyway. That does not mean the ad did nothing, but it does mean the reported ROAS is an upper bound.

Step 2: Check halo share

Halo share = Brand halo sales ÷ Total sales

A high halo share is fine for a brand building campaign. It is a warning sign for a campaign meant to launch one specific product. If 70% of a launch campaign's sales are for your bestseller, the launch is riding on the bestseller's demand.

Step 3: Check new to brand share

Campaigns built for acquisition should show a meaningfully higher new to brand % than your remarketing. If prospecting and remarketing report similar new to brand shares, your audiences probably overlap and need exclusions.

Step 4: Compare against the Purchases (all views) and multi touch versions

Put standard, all views and multi touch results for each campaign side by side. The campaigns whose numbers swing the most between models are the ones where attribution choices, not performance, are driving your decisions.

Step 5: Test incrementality

Attributed ROAS tells you what DSP touched. Incremental ROAS tells you what DSP caused.

iROAS = (Purchase rate of exposed group minus Purchase rate of holdout group) × Exposed shoppers × Average order value ÷ Ad spend

Example: A remarketing ad group reaches 150,000 shoppers and reports ROAS of 9.0. A randomized holdout of the same audience, never shown the ads, purchases at 2.1%, while the exposed group purchases at 2.6%. Average order value is $40 and spend is $6,000.

(0.026 minus 0.021) × 150,000 × $40 ÷ $6,000 = 5.0

The ad group is still profitable, but the true return is closer to 5 than 9. Compare iROAS with your break even ROAS (1 ÷ contribution margin) when deciding whether to scale. Holdouts can be built with AMC audiences or Amazon's lift studies.

Common Amazon DSP Attribution Mistakes

How Atom11 Helps Brands and Agencies Manage DSP Attribution at Scale

Every check above is manageable for one advertiser. It gets heavy for an agency running DSP for a dozen brands, or a brand running dozens of orders across markets. The work is not the analysis itself. It is pulling promoted, halo, total, all views and multi touch metrics from several reports, lining them up with Sponsored Ads and retail data, and doing it again every week while the numbers keep maturing.

Whether you use Amazon's native console, AMC directly or a third party platform, look for these capabilities:

• Unified reporting: DSP and Sponsored Ads side by side, with promoted, halo and total metrics, so you can see the full funnel in one place.
  
  

• Retail context: inventory, price and Buy Box data next to ad metrics, so a drop in attributed sales can be traced to a stockout or a lost Buy Box rather than blamed on the campaign.
  
  

• AMC access without heavy SQL: templates for path to purchase, frequency and overlap analysis, and the ability to push audiences back into DSP.
  
  

• Operational controls: bulk creation, cross order pacing and frequency caps by audience, so fixes found in attribution analysis can be applied quickly.
  
  

• Plain language analysis: a way to ask why performance changed instead of rebuilding the same pivot table.
  
  

Atom11 is one example of a platform in this category. Its Amazon DSP module brings DSP and Sponsored Ads reporting into one dashboard with promoted, halo and combined metrics, supports AMC audiences, and lets teams bulk create orders and ad groups, pace across orders and set frequency caps per audience. Teams that prefer working in an AI assistant can also connect their Amazon Ads data through the Atom11 MCP. For agencies, the time saved on report stitching is usually where the value shows up first.

As with any tool, trial it against your own reporting workflow, and check how it handles the 2026 view attribution metrics and multi-touch data before committing.

Conclusion

Amazon DSP attribution is not complicated once you know the rules. Conversions count within 14 days. Clicks beat views, promoted products beat halo products, and the most recent interaction in the winning tier takes the credit, once, across every Amazon ad. Sales land on the ad interaction date, so fresh data always looks worse than it will. Since January 2026, some view conversions follow a stricter shopping signal model, and Purchases (all views) is your bridge to past performance.

The brands and agencies that get the most from DSP treat last-touch ROAS as one lens, not the verdict. They check view share, halo share and new-to-brand share, compare last touch with multi-touch, look at conversion paths in AMC, and run holdout tests on the ad groups that matter most. Do that consistently, and you will stop cutting the campaigns that build demand and stop overfunding the ones that simply show up at checkout.