Amazon DSP Mistakes: 7 to Avoid in 2026
Archit Arora
TL;DR
Key takeaways
The costliest Amazon DSP mistakes in 2026 are missing frequency caps, missing purchaser exclusions and judging prospecting on last-touch ROAS.
Since January 1, 2026, DSP ads in the Amazon store use a stricter view model, so compare years with "all views" metrics.
Fewer, bigger ad groups with separate supply learn faster than a dozen small ones, and a holdout shows what was incremental.
Test DVA+ quick setup and Full-Funnel Campaigns beside a structure you control, never as a wholesale switch.
Most Amazon DSP accounts we audit aren't failing because DSP doesn't work. They fail on a handful of settings and measurement habits that quietly waste a third of the budget.
This is the list of Amazon DSP mistakes we fix most often in 2026, each with what goes wrong, why, the fix and the arithmetic. After it, we correct the myths that still keep brands from trying DSP at all.
One naming note: "Amazon DSP" also means Amazon's Delivery Service Partner program for package delivery. This post is only about the advertising platform.
What Changed in Amazon DSP for 2026
Several of these mistakes are new, because the platform changed under advertisers this year. If your DSP playbook was written in 2024 or 2025, these are the changes that break it.
Date | What changed | Mistake it creates |
|---|---|---|
Jan 1, 2026 | New view attribution model for DSP ads in the Amazon store | Reading 2026 results against 2025 |
Apr 2, 2026 | De-duplicated reach and frequency across sponsored ads and DSP | None, if you use it. Most accounts don't |
Jul 22, 2026 | Audiences in Brand+ and Performance+ became "Audience Signals" | Assuming an audience fences in the AI |
Sep 29, 2026 | Amazon Ads Agent; DSP campaigns become Display, Video and Audio (DVA) campaigns; Full-Funnel Campaigns in the US | Handing everything to AI without a test |
Late Oct 2026 | DVA+ open beta in the US | Using quick setup where you need controls |
The vocabulary moved too. Orders are now campaigns, line items are ad groups and creatives are ads. Outside the US, DSP users keep the classic flow for now, as our guide to what Amazon DSP is and how it works explains.
The 7 Amazon DSP Mistakes We Fix Most Often
These are ordered by how much money they waste in the accounts we take over, not by how often blogs mention them. The numbers in each example are illustrative unless a source is linked.
1. Running With No Frequency Cap, or One Blanket Cap
What goes wrong. A small remarketing audience sees the same ad 20 or 30 times a month. The budget spends, the reported ROAS looks fine, and most of the impressions bought nothing.
Why it happens. Advertisers leave the cap blank to "help delivery", or copy one cap across every ad group. Amazon's own frequency capping guide says "there is not one set frequency cap that works for everyone" and that a blanket cap "isn't a recommended practice."
The upside of getting it right is measurable. When Amazon launched frequency groups, it said advertisers using its frequency cap controls saved up to 26% of campaign budgets from duplicate impressions and drove up to 21% incremental reach.
The arithmetic. A remarketing ad group targets 30,000 people with $6,000 a month at an $8 CPM. That's 750,000 impressions, or 25 per person.
At a cap of three a week, the pool can absorb about 360,000 impressions in four weeks, which is $2,880. The other $3,120 is buying the same people a fourth, tenth and twentieth time.
The fix.
Set a campaign-level cap and a tighter one per ad group. We start remarketing at 3 to 5 a week and prospecting at 1 to 2 a day.
Use frequency groups to cap a household across campaigns, so awareness and remarketing don't stack on the same person.
Check total frequency in de-duplicated reach reporting once a month. A household seeing you 25 times a week is invisible in any single campaign report.
Our Amazon DSP frequency capping guide has the starting caps by tactic, and our bidding strategies guide shows why raising the bid on a capped pool only raises the price of the same people.
2. Retargeting and Prospecting Without Excluding Recent Purchasers
What goes wrong. A "prospecting" ad group reports a strong ROAS, and most of the buyers already bought from you last month. Remarketing keeps chasing people who converted yesterday.
Why it happens. Exclusions are optional and nobody owns them. Amazon credits the sale either way, so the campaign report rewards the mistake.
The arithmetic. Without exclusions, $10,000 of prospecting drives 400 purchases and $60,000 in sales, a 6.0 ROAS. Only 15% are new to brand, so you paid about $167 for each of 60 new customers.
Exclude anyone who bought in the last 365 days and anyone who viewed your detail pages in the last 30. Purchases drop to 220 and ROAS to 3.3, but 70% are new to brand: 154 new customers at about $65 each.
The fix.
Exclude purchasers of the advertised products from every prospecting and consideration ad group, with a window that matches your repurchase cycle.
Exclude detail page viewers and cart abandoners from prospecting, so they get the remarketing message instead.
In remarketing, exclude anyone who bought in the last 7 to 14 days.
Watch new-to-brand share per ad group. A prospecting line under 30% new to brand has broken exclusions.
This matters more since July 22, 2026. Amazon says the audiences you add to Brand+ and Performance+ are now Audience Signals, "not hard targeting constraints", while exclusions "function independently." Exclusions are the only fence you still own in those campaigns.
Our guides to excluding audiences in Amazon DSP and Amazon DSP retargeting have the exclusion builds and lookback windows we use.
3. Judging Prospecting on Last-Touch ROAS, With No Holdout
What goes wrong. Remarketing shows a 9.0 ROAS, prospecting a 2.5, and the budget moves to remarketing. Three months later the remarketing pool has shrunk and total sales are flat.
Why it happens. Last-touch attribution credits the ad closest to the sale. Remarketing sits next to purchases that would often have happened anyway, so it always wins that comparison.
Attribution says who saw an ad before buying. Only a holdout, where a random part of the audience is kept from seeing the ads, says whether the ad caused the purchase.
The arithmetic. Each tactic spends $5,000. Remarketing gets $45,000 attributed (9.0 ROAS), prospecting $12,500 (2.5 ROAS).
The holdout shows 20% of remarketing sales were incremental, against 72% for prospecting. Both are worth $9,000 of incremental sales, an incremental ROAS of 1.8. Cutting prospecting would have shrunk the business for no gain.
The fix.
Score by stage. Judge prospecting on cost per detail page view, new-to-brand share and long-term sales, not 14-day ROAS.
Run a holdout once a quarter. Split an AMC audience at random, withhold ads from one part for four to six weeks and compare purchase rates. If exposed shoppers convert at 2.4% and the holdout at 1.8%, a quarter of attributed sales were incremental.
Read the path in AMC. Amazon opened Amazon Marketing Cloud to all sponsored ads advertisers in September 2025. Path-to-conversion and overlap queries show what DSP did before the sponsored ads click.
Judge the account on [TACOS](https://adbrew.io/blog/amazon-tacos) and total new customers, not the DSP report alone.
Our full-funnel strategy guide has a stage-by-stage scorecard, and our AMC custom audiences guide shows how to build the audience you split.
4. Reading 2026 Results Against 2025 Without "All Views"
What goes wrong. DSP ROAS fell in January 2026 and someone cut the budget. Performance didn't change. The way Amazon counts view-based sales did.
Why it happens. From January 1, 2026, Amazon DSP campaigns that serve ads in the Amazon store, plus vCPM Sponsored Brands and vCPM Sponsored Display, use what Amazon calls a "shopping-signal enhanced last-touch attribution model" with a shorter view window. Click attribution is unchanged, and so is DSP supply off the Amazon store.
The old count survives. Amazon's announcement says "all views" metrics, such as Purchases (all views), include "all ad views for eligible campaigns within a 14-day window."
Metric for one ad group | Q1 2025 | Q1 2026 |
|---|---|---|
ROAS, default view model | 4.2 | 3.1 |
ROAS, all views | 4.2 | 4.0 |
What it means | Old model | Measurement changed; performance roughly flat |
*Illustrative numbers.*
The fix.
Compare any period that crosses January 1, 2026 on all-views metrics, pulled from unified reporting or the API.
Re-base KPI targets for goal-based bidding. A ROAS target set from 2025 view-attributed results will leave budget unspent in 2026.
Tell finance before the quarterly review, not after. "The ruler changed" lands better as a heads-up than as an excuse.
Unified reporting has been generally available since June 8, 2026, and the legacy DSP report pages become read-only on December 17 and retire on December 31, 2026. Our Amazon DSP reports guide maps the old reports to the new ones.
5. Spreading the Budget Too Thin Across Ad Groups
What goes wrong. A $12,000 monthly budget is split across 12 ad groups: four audiences times three supply types. None of them gets enough data to optimize, and none can be compared with another.
Why it happens. More ad groups feel like more control. In DSP, each ad group bids and learns on its own outcomes, so slicing the budget slices the learning.
The arithmetic. At an $8 CPM, $1,000 buys 125,000 impressions. At a 0.4% detail page view rate and 10% of viewers buying, each ad group gets about 500 detail page views and 50 purchases a month, roughly a dozen a week.
The same $12,000 across four ad groups gives each about 375,000 impressions, 1,500 detail page views and 150 purchases. That is enough for the bidder to learn and for you to see a difference between tactics.
The fix.
Build one ad group per job: a funnel stage, an audience and a supply type that genuinely bids differently.
Give every ad group at least a few hundred thousand impressions over a four- to six-week flight. Below that, merge.
Use one campaign budget with automated budget allocation across ad groups, and keep pacing profiles consistent inside it.
If the whole DSP budget is a few thousand dollars a month, run one remarketing and one prospecting ad group, or start with display ads (formerly Sponsored Display).
Our Amazon DSP campaign setup guide shows the structure we build for a first campaign.
6. Leaving Supply Controls on Defaults, or Locking Them Too Tight
What goes wrong. Amazon-owned inventory and open-exchange supply sit in one ad group. The blended CPM looks cheap, most impressions land on low-performing sites, and nobody can tell.
The opposite failure is just as common: an allowlist of 40 publishers and the strictest suitability tier, and the ad group can't spend.
Why it happens. Supply settings live several screens away from audiences and bids, so they get set once and forgotten. Each control narrows the auction, and a narrower auction costs more per impression.
The arithmetic. One ad group buys a million impressions: 700,000 on the open exchange at a $3.50 CPM and 300,000 on Amazon-owned supply at $11. The blended CPM is $5.75 and looks fine.
At a 0.1% detail page view rate, the exchange impressions cost $3.50 per view. Amazon-owned supply at 0.5% costs $2.20 per view. The "cheap" supply was the expensive one.
The fix.
Control | What to do |
|---|---|
Supply type | Give Amazon-owned, third-party exchange and streaming TV supply their own ad groups and bids |
Inventory tier (since Nov 14, 2025) | Pick Expanded, Standard or Restrictive for Twitch and third-party supply to match your brand's risk tolerance |
Domain and app lists | Use blocklists to trim waste; use allowlists only when delivery forecasts still look healthy |
Third-party verification | Pre-bid brand safety and viewability segments are billed as a CPM fee, so add them where they pay |
Amazon's built-in protection | Amazon says it blocks made-for-advertising sites and removes invalid traffic "from reports and invoices" |
Amazon's supply quality page covers the built-in protections. In July 2026 Amazon also described its Certified Supply Exchange program, which certifies five SSPs, including Magnite, PubMatic and Index Exchange, across eight technical areas.
When an ad group under-delivers, we check audience size and bid first and loosen brand safety last.
7. Handing Everything to DVA+ Quick Setup or Full-Funnel Campaigns Without a Test
What goes wrong. A brand moves its whole DSP budget into an AI-managed campaign on day one. Results look fine in the new report, and there is nothing left to compare them with.
Why it happens. The new flows are genuinely easier. Amazon's DVA+ announcement says quick setup handles "targeting, inventory selection, format allocation, bidding, creative generation, and continuous optimization."
Advanced settings bring back what quick setup hides: frequency caps, publisher lists, deals, manual bids, lookalikes, suppression and multiple ad groups. DVA+ entered open beta in the US in late October 2026, with general availability in 35 markets targeted for Q1 2027.
Full-Funnel Campaigns run sponsored ads, display, video and streaming TV on one budget. They are available in the US (Amazon still labels them a beta on its product page), only for advertisers who sell on Amazon.
Amazon's launch note reports 67% higher long-term ROAS and 29% lower cost per new-to-brand purchase. The footnote compares against "the blended average of other campaign performance across beta advertisers," from June to August 2026, which is a comparison that flatters any new campaign type.
The fix: a controlled test.
Split by product, not by mood. Put a matched set of products, or 20% to 30% of the budget, in the AI campaign. Keep the rest in your own structure.
Keep exclusions in both cells. Past purchasers and your CRM list go on the AI campaign too, or it will "find" customers you already have.
Allow the learning period. Amazon's Full-Funnel Campaigns page describes about two weeks of learning, so judge either flow on the six weeks after that.
Compare like with like. Total sales, TACOS, new-to-brand customers and, if you can, incremental sales from a holdout. Not each campaign's own ROAS.
Move budget in steps. If the AI cell wins, shift another 20% to 30% and test again.
Our view: quick setup is fine for a first Prime Video reach buy. Anything you need to explain to a finance team belongs in advanced settings, because a blended AI result can't tell you which audience earned the money. Our Amazon DSP vs Sponsored Display guide covers where DVA+ fits next to display ads.
Amazon DSP Myths, Corrected for 2026
Many brands never make the seven mistakes above, because they never try DSP. These are the myths we still hear in sales calls, with what's true today.
"Amazon DSP Is Only for Big Brands With Big Budgets"
Not anymore. Amazon's Ads Agent page says DVA campaigns are "available self-service with budgets you set," while managed service "typically requires a minimum spend that varies by country." Amazon's DSP guide puts that managed minimum at $50,000.
No minimum isn't the same as no practical floor. A test needs four to six weeks and a few hundred thousand impressions per ad group, as mistake 5 shows. Older posts, including our earlier myths article, suggested starting at $100 a month; a budget that small won't teach you anything.
"You Have to Sell on Amazon to Use Amazon DSP"
False. Amazon says Ads Agent "is available to advertisers whether or not they sell products in the Amazon store," and that those who don't "can run Display, Video, and Audio campaigns."
An insurer can still reach households that just bought a car, and a fitness app can reach shoppers in-market for running shoes. What a non-seller can't buy is sponsored ads or Full-Funnel Campaigns, which promote products listed on Amazon. Our guide to Amazon DSP for non-Amazon sellers covers the setup.
"Amazon DSP Takes Credit for Organic and Sponsored Ads Sales"
Partly true, if you let it. Without exclusions, DSP serves people who were already going to buy and claims the sale, which is mistake 2.
Exclude recent purchasers and detail page viewers from prospecting, then check new-to-brand share and AMC overlap between DSP and sponsored ads. A holdout settles the argument with a number instead of a debate.
"Amazon DSP Is Only for Upper-Funnel Awareness"
DSP covers the whole funnel. Streaming TV and Prime Video build reach, in-market and contextual audiences drive consideration, and remarketing and loyalty audiences close and repeat sales.
Amazon keeps adding to the bottom of the funnel. Since April 30, 2026, Performance+ remarketing and retention tactics can also optimize to consideration KPIs such as detail page view rate.
"Amazon DSP Only Drives Sales on Amazon"
False. DVA campaigns can send traffic to your own website or app and optimize on off-Amazon conversions recorded by the Amazon Ad Tag and the Events API, managed in Ads data manager.
That's how D2C brands, apps and service businesses use DSP, and AMC can join those conversions to ad exposure for path analysis.
"It's Just Another Display Platform, the Same as Sponsored Display"
Less true every year. Amazon DSP buys Prime Video, Fire TV and Twitch, plus partner supply from Roku, Disney, Netflix, Spotify and SiriusXM, with Amazon's shopping signals attached.
In 2026 the lines did blur: Sponsored Display is now called display ads inside Amazon Ads Agent, and DVA+ brings display, video and audio into one flow. DSP still adds frequency groups, deals, supply controls and audiences off Amazon. Our Amazon DSP ad types guide and targeting guide show what that means in practice.
A 30-Minute Amazon DSP Audit
When we take over an account, we run these seven checks before changing a single bid. They map one to one to the mistakes above.
Check | Where to look | Red flag |
|---|---|---|
1. Frequency | Ad group settings, de-duplicated reach report | No cap, or average frequency above 8 a week in remarketing |
2. Exclusions | Each prospecting ad group's exclude group | New-to-brand share under 30% |
3. Measurement | Your weekly report and AMC | Prospecting judged on 14-day ROAS; no holdout this year |
4. View model | Year-over-year comparisons | Default view metrics compared across January 1, 2026 |
5. Budget spread | Impressions per ad group per flight | Ad groups under a few hundred thousand impressions |
6. Supply | Supply sources and brand suitability | Amazon-owned and exchange supply mixed in one ad group |
7. AI campaigns | Budget share in DVA+ quick setup or Full-Funnel | All of it, with no control cell |
If an account fails three or more, we fix exclusions and caps first. They are the fastest money back.
How We Avoid These Mistakes in Adbrew
Everything above can be done by hand in Amazon Ads Agent. The hard part is checking it every week, across DSP and sponsored ads, and often across several brands.
Adbrew is an agentic retail media management platform for Amazon and Walmart, and DSP runs inside it next to Sponsored Ads, AMC and Marketing Stream:
A team and a seat. Run Amazon DSP through Adbrew's own seat, with no minimum spend, or connect a DSP account you already hold.
Automation you control. Rules on bids, budgets, audiences and frequency caps, with the thresholds and guardrails you set, and every change logged.
AMC without SQL. Adbrew's AMC tool runs path-to-conversion, overlap and new-to-brand reports and builds exclusion audiences, such as recent purchasers, that go straight to DSP.
[Adbrew Intelligence](https://adbrew.io/adbrew-intelligence). Ask for "every DSP ad group with new-to-brand share under 30% this month" or "frequency by campaign for the last four weeks," and the agent builds it from your data.
[Adbrew MCP](https://adbrew.io/mcp). The same checks run from Claude or ChatGPT, with any change applied only after you approve it.
More than 5,000 brands, agencies and aggregators use Adbrew, and customers average a 46% decrease in ACOS. On DSP, the value is less about access and more about catching these seven mistakes before they cost a quarter.
Fix the Settings, Then Judge the Channel
Most "DSP doesn't work for us" verdicts we hear trace back to these seven mistakes, not to the channel. Fix caps, exclusions and measurement first, and the numbers you judge DSP on start telling the truth.
Want a second pair of eyes on your account? Book a demo with our team and we'll run the seven-check audit on your DSP campaigns with you.
Frequently Asked Questions
What are the most common Amazon DSP mistakes?
The most common Amazon DSP mistakes are running without frequency caps, prospecting without excluding recent purchasers, judging prospecting on last-touch ROAS, comparing 2026 results with 2025 without "all views" metrics, spreading budget across too many ad groups, poor supply controls and moving everything to DVA+ or Full-Funnel Campaigns without a controlled test.
What frequency cap should I use on Amazon DSP?
Amazon says no single frequency cap works for everyone, so test by tactic. We start remarketing at 3 to 5 impressions a week per person and prospecting at 1 to 2 a day, then use frequency groups to cap households across campaigns. Raise a cap only when delivery is short and performance holds.
Is there a minimum spend for Amazon DSP in 2026?
Amazon DSP self-service runs on budgets the advertiser sets, with no minimum stated by Amazon. Amazon's managed service typically requires $50,000, varying by country. Partners with their own seat set their own terms; Adbrew has no minimum. A useful test still needs four to six weeks and enough budget per ad group to learn.
Can I use Amazon DSP if I don't sell on Amazon?
Yes. Amazon DSP, now Display, Video and Audio campaigns in Amazon Ads Agent, is open to advertisers who don't sell in the Amazon store. They can reach Amazon audiences on Prime Video, Fire TV, Twitch and third-party sites and track conversions on their own website. Sponsored ads and Full-Funnel Campaigns require products sold on Amazon.
Why did my Amazon DSP ROAS drop in 2026?
Amazon DSP ROAS often dropped in January 2026 because Amazon changed how it credits view-based sales. From January 1, 2026, DSP ads in the Amazon store use a shopping-signal enhanced last-touch model with a shorter view window. Compare periods on "all views" metrics before deciding performance actually fell.
How do I know if Amazon DSP is incremental?
Run a holdout. Split an Amazon Marketing Cloud audience at random, withhold DSP ads from one part for four to six weeks and compare purchase rates. The difference, applied to attributed sales, gives incremental sales and incremental ROAS. Brand lift studies and AMC path-to-conversion reports add evidence for upper-funnel campaigns.
Should I use DVA+ quick setup or advanced settings?
DVA+ quick setup suits a first reach campaign when you will judge the blended result. Advanced settings suit anything you need to explain or optimize, because they restore frequency caps, publisher lists, deals, manual bids and multiple ad groups. Test quick setup beside a structure you control rather than moving the whole budget.
Does Amazon DSP cannibalize Sponsored Products sales?
Amazon DSP can claim sales that Sponsored Products or organic search would have made, mainly when prospecting doesn't exclude recent purchasers and detail page viewers. Exclusions, new-to-brand share, AMC overlap analysis and a holdout test show how much DSP adds. In most accounts we manage, well-excluded DSP lifts branded search rather than stealing it.
