Amazon Full-Funnel Campaigns report Long-Term ROAS. How much of it is a forecast?

KEY TAKEAWAYS

1. Long-Term Sales is booked sales plus a 12-month forecast. Report the two parts separately.
2. The 67% headline compares against a blended average, not a holdout. Metric mix alone can produce a gap that size.
3. Work out the share of the forecast that must arrive to break even before you scale.
4. Decide on total sales of test products against controls, not on any ROAS Amazon computes.

Amazon's Full-Funnel Campaigns put sponsored ads, display, video and streaming TV under one budget and let Amazon's AI decide the mix. The campaign steers towards a metric called Long-Term Sales. Part of that number is sales you can see in your orders this month. Part of it is a forecast that Amazon computes and does not publish the method for. Before you move budget, separate the two.

What did Amazon launch on 29 September?

At unBoxed 2026 Amazon said Full-Funnel Campaigns were open to every advertiser in the United States. The product page limits that to advertisers who sell products in the Amazon store, including vendors, Brand Registry sellers and agencies.

You give the campaign products, a daily budget, creative and an optional ROAS threshold. Amazon's AI then spreads the budget across formats on Amazon and on third-party sites and apps. According to the product page, the system shifts money between those formats in real time to grow Long-Term Sales, after a learning period of about two weeks. Billing is per click or per thousand impressions, depending on placement.

Two things are missing. You cannot choose channels or audiences, as Digiday reported after speaking to Amazon. And Amazon does not publish the minimum daily budget. Netpeak saw $200 a day during the beta, while Incrementum Digital cites MediaPost's report of $100. Check your own console.

What is Long-Term Sales made of?

Amazon's measurement pages give the definition. Long-Term Sales is immediate ad-attributed sales plus the estimated value of new-to-brand shoppers over the next 12 months.

A shopper is new to brand if they have not converted with the brand in the past 12 months. The estimate starts from what that shopper did after seeing the ad: a detail page view, a branded search, an add-to-cart or a purchase. Amazon then applies the brand's historical 12-month return for shoppers who took the same action. Its 2024 launch release gave the example of crediting a first product page visit with value from past visitors who later bought.

So the metric has a booked part and a forecast part. Skai describes the same split as return-on-engagement models plus 14-day attributed sales. Amazon calls the forecast incremental but does not disclose the model.

Some guides get this wrong. SellerVine's 2026 guide adds revenue from improved organic rank to its definition. Amazon's pages do not mention organic sales. Use Amazon's definition.

Does the 67% mean the campaign sells more?

Amazon says beta advertisers saw 67% higher long-term ROAS and 29% lower cost per new-to-brand purchase. The footnote says the comparison is against the blended average of other campaigns run by beta advertisers, using Amazon internal data from June to August 2026. Another Amazon page describes the base as independently configured campaigns. Neither describes a holdout.

That base matters because the two sides earn forecast credit at different rates. A campaign that chases new-to-brand detail page views collects forecast value. A typical Sponsored Products campaign, which mostly captures shoppers already searching, collects little.

A hypothetical shows how far the metric alone can move. Say the other campaigns return $3.00 in booked sales per dollar with almost no forecast credit, so their long-term ROAS is 3.0. Full-Funnel returns the same $3.00 booked, plus $2.00 of forecast. Its long-term ROAS is 5.0, which is 67% higher by our calculation, with no difference in booked sales.

We are not saying that is what happened. We are saying the published comparison cannot rule it out. Stella Rising adds a second caution: beta participants chose to join.

The booked part is not like-for-like either. Since 1 January 2026, Amazon has used a machine-learning view model for Sponsored Brands, viewable-impression Sponsored Display and in-store DSP. Off-Amazon DSP still counts views over 14 days, per Code3 and Feedvisor. Sponsored Products remains click-based, with a 7-day window for sellers, per Feedvisor. Sponsored Brands and Display also credit sales of your brand by any seller, resellers included.

Who is right, Amazon or the sceptics?

Amazon's case is that last-touch windows miss real value. Lily Tong of Amazon Ads told Digiday that about 20% of a brand's sales value arrives after the usual 7 to 30-day window. That figure comes from Amazon's study of more than 2 million campaigns in June.

Agencies quoted by Digiday are split. Radoslav Mazganov of Hyperzon said "we haven't seen it move the needle". Bryce Hamlin of Power Digital said visibility depends on where the data comes from. Digiday also reports that Amazon gives agencies 90 days of historical sales data, while marketing mix models need about 18 months.

Both sides can be right. Lagged value exists on average. But a 20% average across all brands does not show that the forecast for your campaign is accurate. Forecasting new customers is hard. Google researchers Wang, Liu and Miao noted in 2019 that many customers never return, so lifetime value has a spike at zero and a long tail. They proposed judging forecasts by calibration, meaning whether predicted groups end up spending what was predicted.

Amazon gives you a tool for exactly that check. The product page describes an accumulated sales report that compares estimated Long-Term Sales with what campaign cohorts actually bought. The limit is time. Each forecast covers 12 months, and the beta data Amazon cites starts in June 2026, so no cohort from it has a full year yet.

How much of the forecast has to come true?

Use one sum per campaign:

Required realisation = (break-even ROAS − booked ROAS) ÷ forecast ROAS

Break-even ROAS is 1 divided by your contribution margin before ad spend. Booked ROAS is attributed sales already in the report divided by spend. Forecast ROAS is the forecast part of Long-Term Sales divided by spend. If the result is below zero, booked sales already cover the cost. If it is above 100%, the campaign cannot break even even if every forecast dollar arrives.

A worked example with invented numbers. A brand spends $10,000 in a month. The report shows $40,000 of Long-Term Sales, a long-term ROAS of 4.0. Of that, $22,000 is booked, a booked ROAS of 2.2. The other $18,000 is forecast, which is 45% of the total and a forecast ROAS of 1.8. The brand keeps 30% of each sale after product cost and Amazon fees, so break-even ROAS is 3.33.

Required realisation is (3.33 − 2.2) ÷ 1.8, or 63% by our calculation. Nearly two thirds of the forecast must turn into real orders, at the same margin, within the year.

Now suppose a holdout shows that only 60% of booked sales are incremental. Booked ROAS falls to 1.32. Required realisation becomes 112%. A long-term ROAS of 4.0 then sits on a campaign that loses money.

What can you see, and what can you not?

Reporting is now by format. Amazon promises a unified view across formats, and Trellis and Digiday describe format-level visibility. Common Thread Collective says results show at goal level only, which matches what Netpeak saw in the beta. Expect format detail but no audience detail.

The ROAS threshold is the main unknown. Amazon does not say whether it applies to booked or long-term ROAS. If it applies to long-term ROAS, a threshold set at your break-even can still lose money on booked sales. Ask your account team, and until you know, set it as if it uses the larger number.

Price is the other blind spot. The complaint filed by the FTC and 22 states on 31 August 2026 says Amazon added hidden surcharges, called soft reserve prices internally, from 2019 and that by 2024 advertisers paid their full bid about 80% of the time. Amazon disputes this and says cost per click for Sponsored Products search ads was flat after inflation from 2019 to 2024. Relevant Audience's advice is to compare the price charged with the bid you set. In Full-Funnel Campaigns you set no bids, so that check is not available. Total sales are the safer measure.

Which reading calls for which action?

Trellis reported that one beta advertiser suspected its early results were mostly retargeting. That row of the table is not hypothetical.

A seven-step test

1. Record a baseline. Export eight weeks of total sales by product, not attributed sales.
2. Split the products. Put three to five products in the campaign and keep similar products out as controls. The controls show what would have sold anyway.
3. Set the threshold from booked numbers. Use your break-even ROAS and assume it applies to long-term ROAS until Amazon says otherwise.
4. Wait out the learning period. Leave the campaign alone for the two weeks Amazon gives.
5. Split the report weekly. Record booked sales, forecast sales and required realisation.
6. Compare total sales. After four to six weeks, the duration Feedvisor recommends, compare test and control products against their baselines. The same logic as a go-dark geo test applies.
7. Read the cohorts. Each month, check the accumulated sales report against the forecast for older cohorts. Scale only when realisation is on track.

New-to-brand is also a definition you should check against your own records. We covered how Meta decides who counts as new. Amazon's 12-month rule is set at brand level. Reading channel reports at the margin, as in our note on Performance Max channel sliders, carries over too.

Where the evidence stops

Every performance figure for Full-Funnel Campaigns comes from Amazon or its partners. Canine Naturals said 41% of its sales from the campaign were new to brand over 206 days, per Marketing Dive. That is a share, not a lift. No independent study has tested Long-Term Sales forecasts against realised sales. Until one does, treat the forecast as a claim you can check, and check it.

Retail media is one place buyers meet brands. AI answers are another. If you want to see how we measure brand presence there, join the waitlist.

Questions

What are Amazon Full-Funnel Campaigns?

A campaign type that runs sponsored ads, display, video and streaming TV from one budget. You supply products, budget, creative and an optional ROAS threshold, and Amazon's AI chooses formats and moves money between them. Amazon opened it to all US advertisers who sell on Amazon on 29 September 2026.

What is Long-Term Sales on Amazon Ads?

It is immediate ad-attributed sales plus an estimate of the next 12 months of sales from new-to-brand shoppers who engaged with the ad. Engagements include detail page views, branded searches and add-to-carts. Amazon values each one at the brand's historical 12-month return for that action and does not publish the model.

Does the 67% higher long-term ROAS mean Full-Funnel Campaigns sell more?

Not on its own. Amazon compared beta campaigns with the blended average of other campaigns from June to August 2026, not with a holdout. A campaign built to win new-to-brand engagements earns more forecast credit, so the gap can come from the metric as much as from extra sales.

How do I test Amazon Full-Funnel Campaigns?

Put a few products in the campaign and keep matched products out. Record a baseline of total sales first. After the two-week learning period, compare total sales of test and control products for four to six weeks, and check booked sales and the accumulated sales report against the forecast.

Sources

1. Amazon Ads, Grow total sales with Full-Funnel Campaigns advertising.amazon.com
2. Amazon Ads, Full-Funnel Campaigns product page advertising.amazon.com
3. Amazon Ads, Introducing Amazon Ads Agent advertising.amazon.com
4. Amazon Ads, unBoxed 2026 news and announcements advertising.amazon.com
5. Amazon Ads, Increase sales objective page advertising.amazon.com
6. Amazon Ads, Understand the value of new-to-brand shoppers beyond immediate sales advertising.amazon.com
7. Amazon Ads, Campaign reporting advertising.amazon.com
8. Amazon Ads, View attribution updates for Amazon store ads advertising.amazon.com
9. Amazon Ads, Multi-touch attribution launch announcement advertising.amazon.com
10. Business Wire, Amazon Unveils Simplified Product Launch, Optimization and Measurement Solutions for Full-funnel Performance businesswire.com
11. Digiday, Amazon wants to own the long game in ad measurement digiday.com
12. Digiday, At its flagship unBoxed conference, Amazon Ads collapses buying silos digiday.com
13. Marketing Dive, How Amazon is streamlining ad buys across multiple formats with AI marketingdive.com
14. Relevant Audience, Amazon Ads renames its unified platform Amazon Ads Agent at unBoxed 2026 relevantaudience.com
15. Relevant Audience, The FTC says Amazon charged advertisers their own bid 79 percent of the time relevantaudience.com
16. Trellis, Amazon Full Funnel Campaigns Beta gotrellis.com
17. Trellis, Amazon unBoxed 2026 and Accelerate 2026 recap gotrellis.com
18. Netpeak, What Does Amazon's Full-Funnel Campaigns Beta Mean for My Pet Brand? netpeak.us
19. Common Thread Collective, Amazon unBoxed 2026 commonthreadco.com
20. Stella Rising, Amazon Unboxed 2026 Recap stellarising.com
21. Incrementum Digital, Amazon Accelerate 2026 Updates incrementumdigital.com
22. SellerVine, Amazon Full-Funnel Campaigns strategy guide sellervine.com
23. Skai, Prime Day and Beyond. Forecasting Long-Term Success With Amazon and Skai skai.io
24. Code3, Amazon Quietly Tightened Attribution code3.com
25. PPC Land, Amazon tightens view attribution as ROAS reporting splits ppc.land
26. Feedvisor, Amazon Ad Attribution. Windows, Halo Sales and ROAS Traps feedvisor.com
27. Feedvisor, Amazon Incrementality feedvisor.com
28. PPC Land, Amazon ad surcharge ppc.land
29. Nebraska Attorney General, FTC and states complaint against Amazon ago.nebraska.gov
30. TheWrap, Amazon Disputes FTC Lawsuit Over Sponsored Ads thewrap.com
31. arXiv, Wang, Liu and Miao, A Deep Probabilistic Model for Customer Lifetime Value Prediction arxiv.org

NEXT STEP

See where your brand stands.

Your Digital Authority Score, who AI recommends for your target questions, and the placements that close each gap. Your first report is free.

Get your free report →