What Is Amazon Marketing Cloud (AMC)?

What Amazon Marketing Cloud actually is

AMC is a clean room, not a dashboard. It runs on AWS infrastructure and holds pseudonymized signals — impressions, clicks, purchases, and whatever first-party data an advertiser chooses to upload. Neither side gets to see the other's raw records. Amazon can't export your customer list; you can't export Amazon's shopper-level data. What comes out of a query is aggregated and de-identified.

That distinction matters because it explains what AMC is not. It's not a media-buying tool — it doesn't serve ads. It's not a reporting dashboard you check daily. It's a query environment, mostly SQL-based (though natural-language querying via Ads Agent is rolling out), that sits underneath your DSP and sponsored ads accounts and lets you ask questions those platforms' native reports can't answer — mainly because those reports are built to explain one channel, not the overlap between channels.

Access is free for eligible advertisers with DSP or sponsored ads spend, either through the self-serve UI or by request through an account exec. The cost isn't the license. The cost is the analyst time to write useful queries, or a partner who already has that built.

How it works — a worked example

Say you're running both DSP and sponsored ads for one brand. AMC lets you stitch a shopper's exposure across both — did they see a display ad, then click a sponsored product, then buy — and see the whole path instead of two separate, overlapping reports each claiming credit for the same sale.

Here's what that looks like on a real book. Across 30 advertisers we manage, July 2026 ran 78.4 million impressions at a $4.00 CPM and a blended $1.42 cost-per-click, landing at a $5.49 blended CPA across 57,137 attributed purchases, with 20.1% of those purchases coming from shoppers new to the brand. The whole portfolio returned 6.04x ROAS, measured across the full book rather than pulled from a single best-performing line item.

One thing an AMC query surfaces fast: that $1.42 blended CPC isn't one number, it's several ad types averaged together. The $0.41 CPC figure that circulates in this category is online-video only — quoting it as a general CPC misrepresents the mix. AMC lets you break the blend apart by ad type instead of reporting on the average, which is the whole point of having the tool.

What AMC is actually for

• Custom audiences. Build audiences from purchase history, engagement events, or your own uploaded segments, then push them straight into DSP, sponsored ads, or video targeting.
• Holistic measurement without double-counting. This is the one that matters most and gets skipped most. If a shopper is exposed to both DSP and sponsored ads before buying, native reporting for each channel will often claim that sale independently. AMC is where you reconcile the two and see one number instead of two inflated ones.
• Incrementality testing. AMC can support holdout groups and matched-market designs — showing what happened to a control group that saw no ads versus a group that did. That's the only honest way to know whether display added anything beyond what would have sold anyway.

Note what's missing from that list: AMC does not itself prove incrementality just by existing. It gives you the environment to run the test. The test still has to be designed and run.

The mistake almost everyone makes with AMC (we've made it too)

The most common error is reading an attributed number out of AMC and treating it as an incremental one. Last-click and last-touch attribution — even inside a clean room — tells you which ad was in the path before a purchase. It does not tell you whether that purchase would have happened anyway. Those are different questions, and AMC's default query templates answer the first one, not the second.

We've reported attributed ROAS as if it were proof of lift before running a proper holdout, and the holdout came back lower. That's not a reason to distrust AMC — it's a reason to distrust attributed numbers on their own, from any platform, in any clean room. The fix isn't a different dashboard. It's a matched control group or a geographic holdout, built as a query, not assumed from the reporting.

When the numbers in your AMC query look wrong

Three things to check before assuming the tool is broken:

• Suppressed small cells. AMC enforces minimum aggregation thresholds to protect shopper privacy. If a segment is small, the query may return nothing or a suppressed value — that's privacy protection working, not a bug.
• Lookback window mismatch. AMC now supports an extended traffic lookback of roughly 25 months. If your query window doesn't match your attribution window, the numbers won't tie back to what DSP or Sponsored Products reports natively.
• Double-counted purchases. If your DSP and sponsored ads ROAS, added together, look too good, they probably are — check whether the same conversion is being counted in both channel reports before you trust a combined number.

If none of those explain it, the next step is checking whether your first-party upload matched correctly on the pseudonymized join keys — a bad match will silently shrink your audience sizes without an error message.

Where reMKTR fits

reMKTR runs Amazon DSP as a managed service and reconciles the results in AMC, because that's the only place DSP and sponsored ads stop double-counting each other and you can see whether display actually added something incremental. We hold 109 live DSP advertiser seats and are part of the Full Circle group, which has managed more than $500M in Amazon spend across 100+ brands. Whether or not you ever work with an agency, the questions in this piece — is that number attributed or incremental, is the same sale being counted twice — are the ones worth asking of any AMC setup, including your own.