All answers

The Prevailing
Answers for
Business model:
Business model, pick up to two

How should we measure whether Amazon ads are incremental?

The PrevailingPublished Oct 28, 2026

Our answer

Judge Amazon ads on tests you design, not on ROAS or Long-Term Sales. Switch campaigns off and on or hold out similar products, read total sales rather than ad-attributed sales and run each test at least 4 weeks.

Confidence
Medium
Strongest proof
5 Amazon rules
6 practitioners, 3 opinions or claims
Sources
10 sources
14 statements

New here? Take the 1-minute tour

Why this is the answer

3 reasons and our judgment

Amazon's own numbers give credit, not proof. Amazon calls its Long-Term Sales an ad-attributed metric. SalesDuo (Amazon agency) treats new-to-brand, ad spend divided by total sales (TACoS) and organic trends as diagnostic signals, not causal evidence.

Regional holdouts are hard on Amazon. Sponsored Products offers little regional targeting, and a significant holdout needs about $15,000 to $20,000 a month in the campaign type tested (Feedvisor, marketplace pricing software maker). SellerApp gives the same figure for DSP.

Reported ad returns can badly overstate what the ads caused. In a worked example with hypothetical figures, a branded campaign reporting 10.8 times ROAS earns 1.6 times if 85% of its buyers would have bought anyway (Velocity Sellers).

Our judgment

  • Test Sponsored Products and other sponsored search campaigns by time or by product rather than by region. Read them on total units and total purchases on the searched terms. That combines Feedvisor's point on regional targeting and SalesDuo's case for switchbacks with the on and off tests Velocity Sellers, Neato, Marketplace Valet and Nova Analytics describe.
  • Below about $15,000 to $20,000 a month in the campaign type tested, read an on and off test as direction rather than proof. We apply Feedvisor's and SellerApp's spend floor for a significant holdout and SalesDuo's warning that low conversion volume can't separate lift from normal variation.
  • Run each test at least 4 weeks outside October 15 to January 14, and repeat it at least twice a year even when spend and setup are unchanged. Feedvisor, SellerApp and Neato run 4 to 6 weeks, Velocity Sellers runs four weeks minimum outside that peak and retests twice a year, and Neato retests quarterly because incrementality shifts with competition and seasonality.
  • Test branded search and retargeting first when your brand already sells well organically. We follow SalesDuo and Velocity Sellers over SellerApp here, because Feedvisor finds them the least incremental and they are where reported returns most overstate.

Our judgment, built from Feedvisor, Velocity Sellers, Neato, Marketplace Valet, Nova Analytics, SellerApp and SalesDuo: our reading of these sources, not a finding from any one of them.

What the sources say

14 statements from 10 sources
  • AmazonUndatedSupports

    Amazon's campaign reporting includes long-term sales, an ad-attributed figure that adds projected value over the next 12 months, plus new-to-brand metrics and 30-day conversion paths.

    “New-to-brand metrics are generated from first-time customers of your brand and help you understand ad-attributed purchases.”

    Read it on advertising.amazon.com

  • AmazonUndatedSupports

    Amazon says its Marketing Cloud clean room is free for eligible advertisers, offers analytics and audiences with templates and no SQL needed, and returns only aggregated, anonymous outputs.

    “Amazon Marketing Cloud is available at no cost to eligible advertisers via web-based UI and API.”

    Read it on advertising.amazon.com

  • AmazonUndatedCaveat

    Amazon offers a free MMM data feed and API to all active advertisers, even those not selling on Amazon, with up to three years of geo-level ad, sales, promotion and other signals delivered within 10 business days.

    “The Amazon MMM Data Feed and API are completely free for all eligible advertisers.”

    Read it on advertising.amazon.com

  • AmazonUndatedCaveat

    Amazon Brand Lift surveys exposed and unexposed shoppers to measure brand perception, not sales. It is free above spend and impression minimums and excludes Sponsored Products.

    “Amazon Brand Lift is free as long as campaigns meet spend and impression minimums.”

    Read it on advertising.amazon.com

  • PPC LandOct 18, 2026Supports

    PPC Land explains TACoS as ad spend over all sales, not just ad-credited sales, and warns it is not causal: it falls when organic sales rise for any reason and can hide brand ads taking organic sales.

    “A stable TACoS can conceal a shift of sales from free to paid placement.”

    Read it on ppc.land

  • NorthbeamOct 15, 2026Supports

    Northbeam, which sells measurement, says measure Amazon with correlation and incrementality against a baseline set before launch, and don't set Amazon budget from platform ROAS. No data shown.

    “Don't allocate Amazon budget just by looking at platform ROAS.”

    Read it on themediabuyer.northbeam.io

  • Nova AnalyticsOct 8, 2026Supports

    Nova Analytics, which sells Amazon analytics, says ACoS can't show which ad sales would have come organically, and suggests pausing Sponsored Products 7 to 14 days to check, a test with no control group.

    “ACoS cannot tell you which of those paid sales would have arrived organically anyway.”

    Read it on novadata.io

  • NeatoSep 25, 2026Supports

    Neato, which runs brands' Amazon sales, describes geo, on-off and audience holdout tests for retail media, says branded ads have low incrementality, and advises retesting quarterly. Its ROAS ranges cite no study.

    “branded keyword ads have low incrementality because the customer was already searching for your brand.”

    Read it on neato.com

  • SellerAppSep 16, 2026Supports

    SellerApp, which sells managed Amazon DSP, says DSP holdouts need $15,000 to $20,000 a month and 4 to 6 weeks, and warns that testing retargeting or branded campaigns first gives confusing results.

    “Four to six weeks is generally recommended, depending on conversion volume.”

    Read it on sellerapp.com

  • Velocity SellersSep 9, 2026Supports

    Velocity Sellers, an Amazon agency, says branded campaign ROAS misleads, and advises a holdout of at least four weeks read on total branded purchases. Its 10.8x example is hypothetical.

    “The ROAS on a branded campaign is the least informative number in your ad console”

    Read it on velocitysellers.com

  • SalesDuoAug 20, 2026Supports

    SalesDuo, an Amazon agency, says Amazon's attributed metrics, AMC journey analysis, TACoS and new-to-brand figures are signals, not proof of incremental sales; use holdout or switchback tests for that.

    “Amazon describes LTS and LTS ROAS as ad-attributed measures, so they should not be treated as experimental proof of incrementality.”

    Read it on salesduo.com

  • Marketplace ValetJun 16, 2026Supports

    Marketplace Valet, an Amazon agency, says ad reports cannot tell new sales from captured demand, so test by pausing brand campaigns in set periods, while keeping branded coverage for defense.

    “Segment branded and non-branded campaigns. Measure them separately.”

    Read it on marketplacevalet.com

  • FeedvisorMay 21, 2026Supports

    Feedvisor, which sells Amazon ad software, says prove incrementality with a geo holdout (ads off in 20% of regions for 4 to 6 weeks), needing about $15,000 to $20,000 a month in the tested campaign type.

    “Pre-commit to a test duration (four to six weeks minimum) and do not peek.”

    Read it on feedvisor.com

  • AmazonSep 16, 2025Supports

    Amazon says Sponsored Ads advertisers can now open Amazon Marketing Cloud themselves inside the Ads Console, under Measurement & Reporting.

    Read it on advertising.amazon.com

In the full answer

For readers who sign in
  • What to do this week3 steps
  • The numbers behind it2 figures from cited sources
  • Where each source standsA chart placing each source on the question
  • The strongest disagreementThe best case against the answer, from a cited source
  • How sure we areWhat its confidence rests on, and what would change the answer

Read next

More in Measurement