Amazon Marketing Cloud (AMC): What It Is and How It Actually Works

What Amazon Marketing Cloud actually is

AMC is a clean room, not an ad platform. It's a secure, cloud-based environment where advertisers query pseudonymized signals from Amazon Ads, their own first-party data, and a small set of approved third parties like Experian and NCS. Nobody — including Amazon — can export or view the identifiable data underneath. You get combined, anonymized results back, not raw records.

AMC doesn't decide bids, and it doesn't publish ads anywhere. It sits above sponsored ads, DSP display, video and audio, and answers questions about what those channels did, together, that neither channel's own dashboard can answer on its own. That distinction matters because most confusion about AMC comes from people expecting it to behave like a campaign manager. It's closer to a shared ledger.

Access is free to eligible advertisers. Sponsored ads advertisers get in through self-service sign-up. DSP advertisers request access through their ads tech account executive. Either way, 'free' means no license fee — someone still has to write the queries and know which results to trust.

How an AMC query actually works, worked example

Here's the real use case, not the brochure version. A brand runs sponsored ads and DSP display at the same time. Sponsored ads reporting credits a purchase because a shopper clicked a sponsored product ad. DSP reporting credits the same purchase because that same shopper saw a display ad three days earlier. Both dashboards are correct about their own attribution window. Neither can see the other, so both claim the sale.

AMC is where that gets resolved. You query the pseudonymized event log across both channels for the same shopper, order the events by timestamp, and see the actual sequence: exposure, then click, then purchase. Across 30 advertisers we manage, in July 2026, reconciling this way showed the book returning 6.04x return on ad spend, 78.4 million impressions at a $4.00 CPM and a blended $1.42 cost-per-click, once DSP-influenced sponsored ads conversions stopped being counted a second time.

Blended cost per acquisition came out to $5.49 across 57,137 attributed purchases, 20.1% of them from shoppers new to the brand — a split you only get by joining DSP exposure data against sponsored ads conversion data inside the clean room, because neither platform's own dashboard hands you the new-to-brand rate for a display-influenced sale on its own.

AMC Audiences: building and activating segments

AMC Audiences are built from the same query results: shoppers who saw a video ad but didn't buy, past purchasers who lapsed, category cross-shoppers. Two build methods exist: rule-based audiences you define directly with logic, and lookalike audiences AMC generates from a seed list you supply. Both push straight into DSP, Sponsored Products, Sponsored Brands and Sponsored Display for activation — no manual export required.

• Rule-based: you write the qualifying logic yourself (exposure without purchase, spend threshold, category history).
• Lookalike: AMC finds shoppers who resemble a seed segment you provide.

The part that doesn't make it into the feature list: audience size has a floor. If a segment doesn't clear Amazon's minimum threshold for privacy protection, AMC won't build it — the query runs, the audience comes back empty, and no amount of cleverer SQL fixes that. The fix is broader inputs, not better logic. This is the first wall smaller advertisers hit in AMC, and it's worth knowing before you burn a week debugging a query that was never going to return a segment.

The AMC API, certification, and who can actually get in

AMC's API layer has three parts, and each does a different job. The Reporting API runs queries and pulls results on demand or on a schedule. The Audience Management API creates rule-based and lookalike audiences ready for activation. The Signal Ingestion API streams or uploads an advertiser's own pseudonymized data into AMC so it can be joined against Amazon Ads signals. Together they let a partner or an in-house team run AMC programmatically instead of clicking through the console for every query.

The AMC certification is an eight-course, beginner-level track for people with little or no technical background — it covers what a clean room is, how custom audiences get built, and how first-party data connects in. It's genuinely useful if you're about to write your first queries yourself. It is not a requirement to use AMC, and it won't teach you which query to ask; that's a measurement question, not a syntax one.

Documentation lives on advertising.amazon.com and inside the AMC console itself, alongside case studies from advertisers who've used it for cross-channel measurement. It's a reasonable place to start if you're evaluating whether AMC fits your setup before you request access.

The mistake almost everyone makes with AMC, including us

AMC will answer exactly the query you write. It will not tell you if you asked the right one. The common failure mode is treating a flattering query result as proof, shipping the audience or the budget shift live, and never checking whether the lift was real or just correlated exposure. We've made this mistake ourselves early on — pulled a number that looked good, moved budget on it, and only found out later that a holdout would have shown the same shoppers converting anyway.

Last-click and last-view attribution can't prove incrementality, and joining them in AMC doesn't fix that on its own — it just removes the double-count. Proving incrementality takes a holdout group or a matched control run alongside the query. If the cohort in your AMC result is thin — under roughly 100 purchases — treat the segmentation as directional, not decision-grade, and wait for volume before acting on it.

When the AMC number comes back worse than what either dashboard was reporting separately, that's usually the honest number, not a broken query. The instinct to shorten the lookback window until the result looks better is the wrong move — it's the same instinct that got you the double-count in the first place. Every change we make against an AMC finding carries three things before it runs: the evidence behind it, a measurement plan, and a rollback trigger.

Where Dr. DSP fits

AMC is Amazon's tool, free to eligible advertisers, and worth using whether or not you ever work with an agency — the self-service path and the certification course above will get a sponsored-ads-only advertiser most of the way alone. Dr. DSP is Amazon DSP, the demand-side platform for programmatic display, video and audio — not the Delivery Service Partner courier franchise — run as a managed product by Fable 5, from Full Circle, which has managed over $500M in revenue across 100+ brands, including brands like HexClad and Beardbrand. We reconcile in AMC by default, because last-click attribution can't prove incrementality and holdouts can. There's no published price — a demo, the first 30 days free, and terms set on a call against real spend and scope, with Orbit included at no extra cost.