Amazon Advertising 2021: What Still Applies and What Doesn't

Why searching '2021 strategy' is the wrong instinct now

Search "amazon advertising strategy 2021" today and you land on pages that are, literally, dated 2021 — or later pages that never got rewritten. eMarketer's own report ranking on this term is titled "Amazon Advertising 2021," published November 17, 2021, forecasting that year's ad revenue. Perpetua's guide covers Amazon SEO tactics — titles, bullets, images, reviews — that are genuinely still correct, but it's a listing-optimization guide wearing a "2021 strategy" headline, not a paid media strategy. Amazon's own resource hub is current, but it's a stream of case studies and product news, not a framework you can plan a quarter against.

What actually changed since 2021: Amazon Marketing Cloud went from a beta feature to the standard way serious advertisers reconcile DSP and sponsored ads instead of double-counting the same purchase twice. Streaming and connected TV inventory expanded well past the display ads that older forecasts anticipated. Non-endemic categories — the financial services and services brands that were flagged as an emerging trend back then — are now a normal part of the advertiser base, not a prediction. None of that means the 2021 fundamentals were wrong. It means those fundamentals now sit inside a measurement layer that didn't exist when those pages were written.

The framework that hasn't changed: capture, build, prove

Strip out the dates and the durable structure is three layers, and it hasn't moved much since 2021 — it's just measured better now:

• Capture — Sponsored Products and Sponsored Brands catch demand that already exists: someone searched, you show up.
• Build — Amazon DSP puts the brand in front of people who weren't searching yet, on and off Amazon, through display, video, and streaming TV.
• Prove — Amazon Marketing Cloud reconciles the two so a purchase touched by both a sponsored ad and a DSP impression only counts once, and so you can run a holdout to see what DSP actually added.

Here's what that looks like on real numbers, not a hypothetical. Across 30 advertisers reMKTR manages, the DSP portfolio delivered 6.04x return on ad spend in July 2026 — measured across the whole book, not the best line item — on 78.4 million impressions at a $4.00 CPM and a blended $1.42 cost-per-click. That produced 57,137 attributed purchases at a blended $5.49 cost per acquisition, and 20.1% of those purchases came from a shopper new to the brand. That last number is the one a Sponsored Products report can't show you: whether the spend grew the customer file, or just found people who would have bought anyway.

How to allocate budget across the funnel

The allocation question — how much to Sponsored Products versus DSP — depends on where the brand actually is, not on a fixed ratio. A brand with thin reviews and low search share should be spending almost entirely on capture; DSP without demand to capture just buys expensive awareness for a product nobody can find yet. A brand with a mature Sponsored Products program and flat growth is usually the one where DSP earns its keep — the search demand is already being caught, so the only way to grow is to create more of it.

The mistake almost everyone makes with these numbers

The most common mistake right now is quoting a single line item as if it describes the whole account. The $0.41 CPC figure that circulates in Amazon advertising benchmarks is real, but it's an online-video number, not a blended cost per click — the blended figure across a full DSP book, ours included, runs closer to $1.42. Budgeting against the cheap number is how a media plan comes in short by the back half of the quarter.

The second mistake is judging DSP on last-click attribution, full stop. Last-click was never built to prove incrementality, and it can't — it only shows who touched an ad last before a sale that may have happened anyway. Holdout tests and matched controls answer the question last-click was never designed to answer. We've seen accounts where the last-click number looked mediocre and the holdout showed real lift, and accounts where the reverse was true. You don't know which one you're looking at until you test it.

When the numbers look bad: what to check before you panic

First: is the DSP campaign's reported conversions overlapping with Sponsored Products conversions in a shared attribution window, effectively counting the same sale twice? Amazon Marketing Cloud will show the overlap directly. Second: is DSP retargeting the same audience Sponsored Brands already reaches, in which case it's buying a sale that was coming regardless? Third, and worth doing before touching a budget slider: run a holdout — hold spend back from a matched slice of the audience and compare. If the holdout group converts at nearly the same rate as the exposed group, the number wasn't wrong. The campaign was.

A 5-point audit for any Amazon advertising 2021 plan

If you still have a plan written in 2021, or inherited one, run it through these five checks before spending another dollar against it:

• Does it separate capture from build? A plan that is only Sponsored Products has no way to create new demand.
• Does it name a new-to-brand target? If the only goal is ACOS, the plan cannot tell growth from relabeled repeat sales.
• Does it say how DSP and sponsored ads are reconciled? Without Amazon Marketing Cloud or similar, the same purchase can be counted twice.
• Is there a holdout test written in? If nobody designed one, the plan assumes incrementality instead of measuring it.
• Are the benchmarks dated? Any CPC, CPM or ROAS figure without a window and an advertiser count is a number to ignore.

A plan that passes all five is current regardless of the year on the cover. One that fails two or more is worth rebuilding from the capture, build, prove structure above.

Where this leaves you

Most of the framework above works whether you run it yourself, hire an agency, or bring in a managed DSP partner. reMKTR runs Amazon DSP as a managed service, reconciling in Amazon Marketing Cloud because last-click can't answer the incrementality question — and sits inside the Full Circle group (100+ brands managed, $500M+ in Amazon revenue managed). That's one way to run the build-and-prove layers above. It isn't the only way, and a brand that hasn't yet got the capture layer working is better off fixing that first, with us or without us.