Amazon Marketing Cloud: What It Answers, Who Can Use It and Where to Start

Amazon Marketing Cloud is a place to ask questions of Amazon's advertising data that the normal reports cannot answer: how shoppers moved between ad types before they bought, which combinations of ads reached customers new to the brand, how long the path from first ad to purchase really is. It works on event level data in a protected environment and returns only aggregated answers, so a brand learns about groups of shoppers, never individuals.

It used to sit behind Amazon DSP. Since 2024 it has been open to sponsored ads advertisers through the ads console, and Amazon says it is available at no cost to eligible advertisers. The cost now is skill and time: the questions have to be asked well, and the answers acted on.

The console reports each campaign on its own, crediting a sale to the last ad clicked within its window. That is fine for bids and wrong for strategy, because shoppers rarely meet one ad. Marketing Cloud can follow the same anonymous shopper across Sponsored Products, Sponsored Brands, Sponsored Display and DSP, so it can answer questions about the combination: whether shoppers who saw a Sponsored Brands video before a Sponsored Products ad bought more often, or whether a display campaign reached people the search campaigns never would have. The reasoning behind those questions is in incrementality testing across Meta, Google and Amazon Ads.

Typical questions a brand starts with are the overlap between campaigns, the share of buyers who were new to the brand by the path that reached them, the time between first exposure and purchase, and how often shoppers saw an ad before buying. None of these is visible in the standard reports, and each changes a budget decision: which campaign is finding new customers, which is buying sales that would have happened, and how long a campaign needs to run before it is judged.

That last point matters for how ACoS is read. A campaign that looks expensive on last click may be the first touch on a large share of the brand's new customers, which is the gap between the ratios set out in ACoS, TACoS and ROAS.

Amazon names three routes. Sponsored ads advertisers can open it themselves in the ads console, under measurement and reporting. Amazon DSP advertisers have long had it and keep it. And agencies registered with Amazon's advertising programme for third parties can run it on behalf of their clients. The requirement that an advertiser also buy DSP was dropped when Amazon widened access in 2024.

Inside, there are three levels of effort. No-code templates answer common questions with a few settings and no code at all. Low-code templates in the use case library start from a written query that can be adjusted. And a query editor accepts custom SQL for anything the templates do not cover. Most brands should live in the templates for their first months; custom SQL is where an analyst earns their keep, once the standard questions have been answered.

Answers come back aggregated, and rows that would describe too few shoppers are withheld. That protects shoppers and occasionally frustrates analysts, because a narrow question on a small brand can come back empty. The fix is usually a broader question or a longer date range rather than a different tool.

Marketing Cloud can now build audiences as well as answer questions. Amazon describes custom audiences created with SQL or with templates and activated in Amazon DSP, in beta in a limited set of countries: shoppers who viewed a product page and did not buy within a set time, for instance, or buyers of one product who have not tried another. It turns an insight into a campaign without the brand ever seeing a list of people. Whether a brand is ready for DSP at all is the question in Amazon DSP, what it is and who it suits.

Two pieces of AI help arrived in 2026. Amazon's ads agent can edit and fix SQL inside the query editor for anyone with Marketing Cloud access, which lowers the bar for custom questions considerably. And SQL generation through Amazon's MCP server, which lets an AI assistant write queries from a plain description, is in closed beta in the US, with general availability planned for early 2027. Both make the queries easier to write; neither decides which questions are worth asking.

Start with three questions, each tied to a decision. First, how much do the campaigns overlap, and are any paying to reach shoppers another campaign already reaches? Second, which paths bring customers new to the brand, and which mostly reach existing buyers? Third, how long does it take from first ad to purchase in the category, and are campaigns being judged before that time has passed? The templates answer all three.

Write down what each answer would change before running it. A finding that would not move a budget, a bid or a campaign's judging window is interesting and not useful, and Marketing Cloud makes it easy to spend weeks on interesting.

Then repeat the same queries monthly. One answer is a snapshot; the same question asked every month shows whether a change to the campaigns moved what it was meant to move.

If nobody in the business writes SQL, the templates and the 2026 help in the editor now cover more than they used to. When a question needs custom work every month, that is the point at which an analyst, or a partner who runs these queries weekly, starts to pay for the time.

Marketing Cloud is built so that no advertiser can see an individual shopper, and the rules that guarantee it shape every answer it gives. Knowing them explains most surprising results.

Only aggregates come out. A query can work over event level data inside Marketing Cloud, but what it returns has to be aggregated. Rows that describe too few distinct shoppers, generally fewer than a hundred, are removed from the output rather than shown. Nothing that identifies a person leaves the environment.

Why a query returns nothing. A question asked about too small a slice, a single small campaign over a few days, a rare path, a narrow audience, can have every row filtered out by that threshold. The answer is not that nothing happened; it is that too few shoppers were involved to report it. Widen the date range or group the results more broadly.

Small brands feel it most. An advertiser with modest volume reaches the threshold less often, so its questions have to be framed at the level of the whole account or of ad types rather than individual campaigns. That is a limit on the detail, not on the usefulness.

The rules are the reason it exists. Amazon can let advertisers ask these questions of its data only because it can promise that no individual is exposed. Treat the thresholds as part of the design rather than an obstacle to work around.

Marketing Cloud answers exactly the question asked, which makes the definitions inside the question matter: the date range, the window chosen for counting a purchase after an ad, and whether a purchase means anything from the brand or only the product advertised. Two queries that differ in one of those will disagree, and both will be right.

Correlation is the trap. Shoppers who saw three ad types may have bought more often because they were keener to begin with, not because of the ads; a path that ends in purchases is not proof the ads caused them. Treat path findings as a pointer to where to test, and prove the important ones with a holdout or a lift study before moving serious budget.

Keep every query and its result in one place, with the date it ran and the definitions it used, so a number quoted in a March meeting can still be traced in September.

On its own, Marketing Cloud answers questions about Amazon's advertising and Amazon's shoppers. With some of the brand's own data added, it can answer questions about the brand's customers as a whole.

What can be added. Advertisers can upload their own data into their Marketing Cloud instance, such as customer lists, with personal details hashed so they can be matched without being exposed. The same aggregation rules apply to every answer that uses it.

What it makes possible. Questions such as how many of the brand's existing customers its Amazon ads reached, whether advertising on Amazon brings in people who are new to the brand everywhere or only new to its Amazon store, and how customers who buy direct behave when they meet the brand's ads on Amazon.

What it takes. A clean, consistent customer list, the right consent to use it for this purpose, and somebody to own the process of preparing and uploading it. The data protection rules that govern the list in the brand's own systems still govern it here.

Start without it. Most of the value of a first project comes from Amazon's own data. Add the brand's data when a specific question needs it, not as a first step, so the work stays tied to a decision.

Marketing Cloud sits between two kinds of work: the analysis that writes and checks the queries, and the advertising decisions that use the answers. It works best when both are in the room.

The analyst. Somebody comfortable with SQL, or with Amazon's templates and its library of instructional queries, who can turn a business question into a query and check that the result means what it seems to. The 2026 AI help reduces the writing, not the checking.

The advertiser. The person who runs the campaigns and owns the budget decisions. They bring the question worth asking and decide what changes because of the answer. Without them, Marketing Cloud produces interesting tables that change nothing.

A rhythm, not a project. One or two questions a month, each tied to a decision, works better than a large analysis once a year. The same questions asked quarterly, with the same definitions, show whether the changes made after the last answer worked.

For a smaller brand. Where there is no analyst in house, an agency registered with Amazon can run Marketing Cloud on the brand's behalf. The brand still needs to own the questions; outsourcing the queries is reasonable, outsourcing the decisions is not.

Marketing Cloud reads Amazon's data. It does not see the brand's own store, its email, or advertising on other networks unless that data is brought in, and it does not replace the weekly work in the console: search terms, bids, budgets. It sits above that work as the place where the strategy questions get answered. How the console and the automated campaigns divide decisions is in Amazon full funnel campaigns.

Running Marketing Cloud alongside DSP and the sponsored formats, asking the questions that change budgets and acting on the answers, is part of our programmatic advertising service.