Incrementality testing by platform: Overview · Meta · Google · TikTok · Amazon · Pinterest · Snapchat
Amazon doesn't offer a self-serve conversion lift test. It gives you Brand Lift surveys, a data clean room called Amazon Marketing Cloud (AMC), and modeled metrics. The experiment itself, meaning who is held out and for how long, is yours to design.
The idea in one picture
The logic is the same as on other platforms. The setup work lands on you.
The tools Amazon gives you
- Brand Lift is a survey-based test of awareness, consideration and purchase intent, built on randomized controlled experiments. It reached general availability in the US for Sponsored Display, Sponsored Brands and Sponsored TV on June 30, 2024, after launching for Amazon DSP. You create studies on the ads console Studies page or through the API (Amazon Ads). It measures what people say, not what they buy.
- Amazon Marketing Cloud is a privacy-safe clean room that combines pseudonymized Amazon Ads signals with optional first-party data and returns only aggregated results. Since September 2025, any brand running sponsored ads can open it from the ads console at no cost (Darkroom). AMC supplies the data; you design the exposed-versus-unexposed comparison (Nakoa Digital).
- Multi-touch attribution uses randomized controlled trials that Amazon runs, plus machine learning, to split credit across ad touchpoints, according to Amazon's own researchers (arXiv). It is attribution informed by experiments, not a lift test on your campaign.
- Long-Term Sales combines the sales a campaign drives right away with an estimated value of new-to-brand customers over the next 12 months. It entered closed beta in October 2024. At unBoxed 2026 (September 28–30), Amazon made it the reporting metric for its new Full-Funnel Campaigns, in US open beta (Improvado). The 12-month part is Amazon's forecast, not booked sales, and Forrester calls the metric "not causal measurement" because it doesn't control for outside factors (Forrester).
- Third-party partners. PPC Land reports that since May 15, 2026 Amazon lists more than 50 third-party measurement products in a self-serve catalog, centered on brand lift and offline sales lift from vendors such as NCS (PPC Land).
Also from unBoxed 2026: Amazon renamed its combined ads console and DSP Amazon Ads Agent. None of the measurement news there added a conversion lift test you can run yourself.
Ways to run a holdout yourself
| Method | How it works | Best for | Catch |
|---|---|---|---|
| Audience holdout in AMC | Suppress a share of eligible shoppers (15–20% in agency guides) and compare purchases | DSP and retargeting | Needs SQL skills or a partner; control contamination |
| Geo holdout | Pause ads in matched regions for 3–4 weeks and compare sales | Brands with regional sales data | Sponsored Products-only campaigns rarely have enough regional detail |
| Product-group pause | Pause spend on a matched set of products for about 4 weeks | Sponsored Products | Ranking dips; groups that aren't truly matched |
| Switchback | Alternate ads on and off by day or week | Sellers without AMC | Time and seasonality effects |
Sources: Eva, Think Nectar, SellerApp, SalesDuo. These are agency and software-vendor guides, not Amazon documentation.
Reading the results
Incremental revenue is exposed sales minus held-out sales, scaled to the same size. Incremental ROAS is that figure divided by ad spend. In one vendor's illustrative example, an exposed group made $250,000 and the control made $175,000, so $75,000 was incremental, a 42.8% lift.
One guide suggests that if branded terms drive 40% or more of your Sponsored Products revenue, a small geo holdout on those terms is worth running before you scale spend.
Caveats
- Amazon publishes little guidance on running your own holdouts. Most of the holdout detail here comes from agencies and software vendors.
- An AMC analysis that just compares people who saw ads with people who didn't is observational. Only a holdout you randomize gives you a causal answer.
- AMC shows what Amazon can see. Sales on other retailers or your own site need your own data.
- Long-Term Sales and multi-touch attribution are Amazon's models of Amazon's ads. Check them against a holdout and your own order data.
I design and read incrementality tests for brands, on the platforms and independently of them. See how a Measurement Audit works.
Sources
- Amazon Ads: Measure brand marketing impact using Amazon Brand Lift
- Amazon Ads researchers: Amazon Ads Multi-Touch Attribution (arXiv)
- Improvado: Amazon unBoxed 2026, what changed for measurement
- Forrester: Amazon Ads unBoxed 2026
- Darkroom: Beginner's guide to Amazon Marketing Cloud
- Nakoa Digital: Incrementality with Amazon Marketing Cloud
- PPC Land: Explaining conversion lift
- Eva: Amazon ad incrementality guide