Amazon Attribution: What It Tracks, What It Misses, and How to Read It

Amazon Attribution tracks the off-Amazon clicks that end in an Amazon purchase. You tag a link in an email, a Meta ad or a creator post, someone clicks it, and if they buy on Amazon within the next fourteen days the tool credits that tag. For a brand that has been running external traffic on faith, that is a real upgrade over guessing.

Then the reports arrive and the numbers do not agree with anything else. Amazon Attribution claims one figure for the week. Meta claims a bigger one. The Amazon payout report shows a third. Nobody is lying, and no setting fixes it, because each system is answering a slightly different question with a different window and a different model.

This guide covers both halves: how to set Amazon Attribution up properly, and how to read the number it gives you without fooling yourself. Start with what each system in your stack can actually see.

Read down the Amazon purchase column and the problem is visible in one glance: three different systems all believe they caused the same order, and only one of them was told about the other two.

What Amazon Attribution is and how it works

Amazon Attribution is a free measurement tool inside Amazon Ads that reports how non-Amazon marketing performs against Amazon sales. Amazon describes it in its own guide to Amazon Attribution as a self-service solution, and there is no media spend attached to it. You are not buying placements. You are buying visibility into traffic you already pay for somewhere else.

The mechanism is a tracking parameter. Amazon Attribution issues a tagged version of the destination URL for a product detail page, a Store page or a search results page. You use that tagged URL as the landing page for a campaign that lives outside Amazon: a Meta ad, a Google search ad, an email, a creator link, a display buy. Clicks through that tag get associated with a campaign name you chose, and Amazon reports what those clicks did on Amazon.

What makes it different from every other report in your stack is the last step: Amazon is the one confirming the purchase. Meta can tell you someone clicked and then guess at the outcome. Amazon Attribution knows whether the order happened, because the order happened on Amazon.

Four things happen in order: you create a tag, you put the tag on an off-Amazon link, someone clicks it, and Amazon looks for a matching purchase inside a fixed window. If it finds one, the campaign gets credit. If the window closes first, the campaign gets nothing.

Attribution tags and links

A tag is scoped to one product and one advertising source. Create a tag for the ASIN you are promoting, pick the publisher and channel it belongs to, and Amazon returns the tagged URL. Amazon documents the setup path in its help article on creating a campaign.

Two practical points that cost people weeks. First, tags do not travel. A tag built for one ASIN reports nothing useful if you point it at a different product, and a tag reused across two channels collapses both into one row. Second, the tagged URL has to survive the trip. Link shorteners, redirect chains and some email platforms rewrite destination URLs, and a tag that gets stripped in transit produces a campaign that reports zero clicks while genuinely sending traffic. Test every link by clicking it and confirming the parameter is still on the URL that loads.

The fourteen day last touch window

Amazon Attribution credits purchases using a fourteen day last touch model, which Amazon states directly in the same guide. Two rules follow from that phrase, and both matter more than the setup steps.

Fourteen days is the outer limit. A customer who clicks your tagged link in March and buys in May is not in your report, and the campaign that genuinely started that purchase looks dead. Last touch means the final tagged click takes everything. If a customer meets your brand through a creator video, comes back through a search ad, and finally converts through a retargeting ad, the retargeting ad is credited with the whole order and the creator video is credited with nothing. Amazon also runs a separate attribution model for on-Amazon ad campaigns, which is why the Amazon Ads console and Amazon Attribution can disagree about the same day.

Who can use Amazon Attribution

Not every Amazon seller can use it. Per Amazon's documentation, Amazon Attribution is open to sellers enrolled in Amazon Brand Registry with brand representative status, to vendors, and to agencies advertising on behalf of those brands. If you sell on Amazon without Brand Registry, the tool is not available to you, and getting registered is the prerequisite rather than a nice-to-have.

Availability is per marketplace, and Amazon lists a specific set of countries where the tool operates. A brand selling in a market outside that set will find the reporting simply absent rather than empty, which reads like a broken setup and is not one.

Amazon Attribution is not a beta, whatever you may have read. Plenty of older guides still describe it that way: Jungle Scout calls it "currently a beta program" in an article last updated in November 2022. Amazon now presents it as a standard part of its measurement suite, so what stands between you and access is Brand Registry, not an invitation.

What Amazon Attribution measures

The metric set is the strongest part of the tool. You get funnel steps rather than a single conversion count, which means you can see where tagged traffic falls over instead of only whether it converted.

The two metrics most people ignore are the two worth building reports around. New to brand separates acquisition from repeat purchase, which is the difference between growth and paying to reach existing customers. Brand halo catches the customer who clicked an ad for one product and bought a different one, which is the most common way external traffic quietly works and looks like it failed.

How to set up Amazon Attribution

The setup itself takes under an hour. Getting it to produce numbers you can trust takes one more step, and it is the one people skip.

Create the tag

In Amazon Ads, open Amazon Attribution and create a campaign. Name it so you can still read it in six months: channel, campaign, product, date. Select the ASINs the campaign points at, define the publisher and channel, and Amazon returns the tagged URL for each product. Bulk creation through a template is available and is worth using past a handful of tags, because hand-built tags are where naming conventions go to die.

Apply the tag to the channel

Replace the destination URL in the live campaign with the tagged version. For paid social and search, that is the ad's landing page field. For email, it is the button target. For creators, it is the link you hand over, and that is the one place to confirm the link was actually used, because a creator who types the plain Amazon URL from memory breaks the measurement without ever mentioning it.

Verify it is recording

Do not wait for the weekly report. Click your own tagged link, confirm the parameter is present in the loaded URL, and then check that clicks appear in Amazon Attribution against that campaign. Reporting is not instant, so allow for a delay before treating a zero as a fault. A tag that was never recording is the single most expensive failure in this whole process, because you find out a month later and the traffic is unmeasurable in retrospect.

The Brand Referral Bonus

Tagging external traffic can also pay for itself. Amazon runs a Brand Referral Bonus that credits eligible brands a share of qualifying sales driven by their own non-Amazon marketing. SalesDuo puts the average at around ten percent of qualifying sales, and notes it varies by category and eligibility. Amazon credits it after the fact rather than at the point of sale.

Treat this as a rebate, not as a measurement feature. It changes the economics of driving external traffic to Amazon and it is a good reason to tag consistently, but it tells you nothing about whether the campaign worked. A bonus arrives whether the attributed sale was incremental or would have happened anyway.

What Amazon Attribution does not tell you

Everything above is the tool working as designed. The gap between what it reports and what your business earned comes from three limits that no configuration removes.

Last touch hides everything before it

A fourteen day last touch model answers one question: which tagged link was clicked most recently before the purchase. It does not describe the path. Every earlier touch, paid or organic, on Amazon or off it, is credited at zero, which systematically overpays whatever sits closest to checkout and starves whatever creates demand.

This is not a Polar opinion about Amazon. Amazon appears to agree: the platform has been rolling out a multi-touch reporting option in beta for United States advertisers, presented alongside the existing last touch figures rather than replacing them. A platform does not add a second lens to a metric it considers finished. The broader trade-offs between single touch and multi-touch are worth understanding before you pick one, and we compared the common attribution models against each other elsewhere.

Attributed is not incremental

Attributed sales are sales that followed a click. Incremental sales are sales that would not have happened otherwise. These are different quantities and the gap between them is usually large, most obviously on branded search and retargeting, where a good share of the credited revenue was already on its way. SalesDuo makes the same point plainly: attribution shows attributed sales, not sales that would not have happened. If you are using attributed revenue to decide whether a channel deserves more budget, you are answering a question the data cannot answer. Holdout tests and geographic lift tests can; attribution reports cannot.

It stops at the marketplace boundary

Amazon Attribution sees Amazon outcomes. If the same Meta campaign also sells on your own store, and for most brands running external traffic it does, those orders are invisible here. The campaign looks worse than it is, and the more of your revenue sits off Amazon the wider that gap gets. Saras Analytics raises the same structural point about fragmented reporting across the Amazon stack.

Reconciling Amazon Attribution with your other channel data

This is the step that decides whether the tool helps you or misleads you. Once Amazon Attribution is live you have a new revenue number that overlaps with numbers you already had. Overlapping is fine. Adding them up is not.

Consider one week on one brand's Amazon channel. The figures below are illustrative and rounded to make the arithmetic legible, not data from any particular brand.

The three ad reports add up to more revenue than the business took, and each one is internally correct. They overlap because the same order is claimed more than once, and because each source uses a different window and a different model. This is the double counted order, and it gets worse as you add channels, not better.

Three moves fix it, in this order.

Anchor on orders, not on platform reports. Start from the order records in your store and your marketplace accounts, which are the only places an order exists once. Then distribute credit across touchpoints. When credit is assigned from a fixed pool of real orders, the attributed totals sum to actual revenue by construction, and no channel can inflate itself.

Hold the window and the model constant across channels. Comparing a fourteen day last touch number against a platform figure on a different window and a different model is not a comparison. Pick one model, apply it everywhere, and keep the platform-reported figures beside it as a reference rather than as a rival truth. Reading the two side by side is also how you find out how much each platform over-reports.

Bring store and marketplace revenue into the same table. This is the step that changes decisions rather than reports. Ad spend from Amazon Advertising, Meta, Google and the rest belongs next to revenue from both the store and the marketplace, so blended return on ad spend and blended acquisition cost describe the business instead of one channel. Polar does this by connecting Amazon Seller Central, Amazon Vendor Central and Amazon Advertising spend alongside Meta and Google, resolving marketplace orders to their own sales channel, and applying one attribution model across the whole set. Ten models are available, from first and last click through linear and time decay to a Shapley value model that estimates each channel's marginal contribution, and switching between them shows how much of your reported performance is a modelling choice.

One honest limit. Customer level metrics behave differently across the boundary. Store orders carry identity, so new versus returning and per-customer acquisition cost are solid there. Amazon's first-party reporting does not expose customer identity the same way, so those metrics are weaker on the marketplace side and should be read as channel level rather than customer level. This is a constraint of Amazon's data, not of the reporting layer, and any tool claiming otherwise is guessing. Brands moving from DTC to omnichannel run into this boundary early.

When Amazon Attribution is enough, and when it is not

The native tool is genuinely sufficient for a real set of cases, and paying for more measurement than you need is its own mistake.

Amazon Attribution on its own is enough when Amazon is effectively your whole business, when external spend is a small share of total marketing, when you are answering a narrow question such as whether a specific creator drove orders, or when you are running the tool mainly to qualify for the Brand Referral Bonus. In all four cases there is no second revenue stream for the numbers to disagree with.

It stops being enough as soon as the same campaigns sell in two places. If a meaningful share of revenue runs through your own store, if you are setting budgets across channels rather than reporting on one, if finance needs a blended number that reconciles to the bank, or if you are being asked whether spend is incremental, then a fourteen day last touch report on one marketplace cannot carry the decision. At that point Amazon Attribution becomes one input among several rather than the answer.

How to read the number without fooling yourself

Five rules, all of which survive contact with a real reporting week.

• Never sum platform-reported revenue. If the channel figures add up to more than the money that arrived, the total is wrong no matter how correct each part is.
• Compare like windows. Before you call a discrepancy a discrepancy, check that both numbers cover the same period on the same window length.
• Hold the model constant when comparing channels. Changing the model changes the ranking, so a channel comparison across two models tells you nothing about the channels.
• Read new to brand separately from total sales. Total sales rewards retargeting existing customers. New to brand is the acquisition signal, and the two often move in opposite directions.
• Use attribution to allocate, use tests to prove. Attribution reports rank and split credit. Only a holdout or geographic lift test tells you whether the revenue was incremental.

Where this leaves you

Amazon Attribution closes a real gap. Before it, external traffic to Amazon was unmeasurable, and a tool that reports clicks, detail page views, add to carts, purchases, new to brand and brand halo against off-Amazon campaigns, for free, is worth setting up carefully.

What it cannot do is tell you what your business earned, because it sees one marketplace through one window with one model. Set it up properly, read it as one input, and reconcile it against store revenue and total spend before any budget moves. The brands that get this right are the ones that stopped asking which report is correct and started asking what each report can see.

If your store and your marketplace are still reported separately, that reconciliation is the next thing to fix.

FAQ