Amazon Marketing Analytics: What It Actually Measures

What Amazon Marketing Analytics Actually Covers

Amazon doesn't have one analytics product. It has a stack, and most people quoting "Amazon marketing analytics" mean whichever piece they last saw a screenshot of. Here's what each one actually does and where it stops:

• Campaign reporting — impressions, clicks, ACOS, ROAS. Sits inside Seller/Vendor Central and the DSP console. Tells you what happened on Amazon during the campaign window, nothing about what would have happened without it.
• Amazon Attribution — tracks how off-Amazon touchpoints (search, social, email, display) lead to on-Amazon detail page views and purchases. Useful for tying non-Amazon spend to Amazon outcomes, useless for measuring Amazon-native campaigns against each other.
• Amazon Marketing Cloud (AMC) — a clean room where you query pseudonymized event-level data across DSP and sponsored ads together. This is the only place that stops DSP and Sponsored Products from double-counting the same purchase, because it can see both.
• Brand Lift — survey-based measurement of awareness, favorability, and purchase intent, drawn from the Amazon Shopper Panel. Directional, panel-based, not transaction-based.
• Marketing Stream — near-real-time hourly campaign data via API. A speed tool, not an insight tool.
• Omnichannel Metrics (OCM) — connects ad exposure to offline point-of-sale activity, mid-campaign. Relevant mostly to CPG brands with retail distribution beyond Amazon.
• Marketing Mix Modeling (MMM) — statistical modeling of incremental sales impact across channels using Amazon's granular ad and retail signals. This is the closest Amazon gets to answering "did this actually work," but it's a model, not a measurement.

Notice what's missing from that list: a single number that tells you, on its own, whether your advertising caused a sale that wouldn't have happened anyway. That's a different question, and it's the one most people actually want answered.

A Worked Example: Reading the Numbers Instead of Skimming Them

Numbers on a dashboard mean nothing until you connect them. Here's how the chain actually works, using a real portfolio figure: across 30 advertisers reMKTR managed in July 2026, the book ran 78.4 million impressions at a $4.00 CPM. That's spend, not an outcome — 78.4M impressions ÷ 1,000 × $4.00 works out to roughly $313,600 in media cost.

That same spend produced 57,137 attributed purchases at a blended $5.49 cost per acquisition — multiply those and you land back near the same spend figure, which is how you sanity-check that a CPA and a purchase count actually reconcile with each other instead of being pulled from two different reporting windows.

The portfolio return was 6.04x ROAS, measured across the whole book — not the best-performing line item pulled out to look good in a deck. And 20.1% of those 57,137 purchases came from shoppers new to the brand, which matters because a high ROAS built entirely on repeat buyers tells you your retargeting works, not that your advertising is growing the customer base.

None of these four numbers — CPM, CPA, ROAS, new-to-brand rate — replaces the others. A dashboard that shows you ROAS without new-to-brand rate can hide a campaign that's just harvesting existing demand.

The Blended-Number Trap

The most common mistake in this category isn't a wrong number — it's a real number applied to the wrong scope. The $0.41 CPC that gets quoted constantly in Amazon marketing content is real, but it's an online-video-only figure. Apply it to a full DSP book and you'll underbudget by a wide margin. The blended CPC across a full book — display, video, and everything else — runs closer to $1.42.

This isn't unique to that one figure. Any CPM, CPC, or CPA you read — ours included — needs a scope attached to it before it means anything. Ask: is this blended across formats or is it the best-performing tactic? Is it the whole book or one advertiser? Is it gross spend or net of fees? If a number doesn't come with an answer to those questions, don't build a budget on it.

Why Last-Click Numbers Lie About Display, and What Fixes It

Sponsored Products and Amazon DSP both claim credit for the same purchase in last-click reporting, because both platforms report against their own last touch, not the customer's actual path. This is not a bug you can toggle off in campaign manager — it's structural. The only way to see whether display added a sale that Sponsored Products would have gotten anyway is to look inside AMC, where both channels' event-level data live in one place and can be deduplicated.

Even then, deduplication only tells you who gets credit for a sale that happened. It doesn't tell you whether the sale would have happened without the ad. That's a different question, and it requires a holdout group or a matched control — some customers see the campaign, a comparable group doesn't, and you measure the gap. Correlation between ad exposure and purchase is not proof of causation, and treating it as proof is a mistake this industry makes constantly, including in ROAS numbers published without stating what's excluded. Ask what's excluded from any ROAS figure you're handed, ours included, before you cite it.

What To Do When The Number Looks Wrong

Bad numbers are more common than clean ones. Before you panic or rebuild a campaign, check these in order:

• Check the attribution window first. A CPA that suddenly doubled might just be a shorter lookback window applied by default, not a real performance drop.
• Check for double-counting. If DSP and Sponsored Products ROAS both look unusually strong at the same time, they may be claiming the same purchases. Reconcile in AMC before trusting either number in isolation.
• Check sample size before trusting lift. A Brand Lift result or a small holdout test with a thin sample can show a swing that's just noise.
• If a fix didn't work, don't re-run the same fix expecting a different result. Change one variable — creative, audience, or bid — and re-test. Changing three at once tells you nothing about which one mattered.
• If the setting was already on, say so. Optimization budgets get wasted turning on features that were already default-enabled. Check current state before proposing a change.

Where This Leaves You

Most of what's marketed as "Amazon marketing analytics" is reporting — accurate, but not proof of anything beyond what already happened. If you're running Sponsored Products alone, campaign reporting and Attribution will cover you. Once DSP enters the mix, AMC stops being optional if you want to know whether display is adding sales or just claiming credit for ones Sponsored Products would have made anyway.

reMKTR runs Amazon DSP as a managed service and reconciles every book in AMC for exactly this reason — last-click reporting can't separate incrementality from overlap, and we don't sell on a dashboard that can't tell you that. Whether or not that's the right setup for you, the questions above — what's excluded, what's blended, what's been deduplicated — are the ones worth asking of any number in this category, from any source.