Amazon DSP is the most misunderstood line on a DTC brand's media plan. Operators hear that it returns four to eight times spend on retargeting, switch it on, watch the dashboard confirm it, and assume the channel is working. Then blended new customer acquisition cost refuses to move, contribution margin stays flat, and nobody can explain why a 6x channel did nothing to the P&L.
The explanation is almost always the same. The audience was built lazily, so a meaningful share of the budget went to shoppers who were going to buy regardless. This is a breakdown of how DSP actually works for a DTC or CPG brand, the benchmarks worth holding yourself to, the three ways budget leaks out of an account, and the honest threshold below which you should not run it at all.
What Amazon DSP Actually Is
Sponsored Products is a search product. Someone types a query, you bid, your listing appears. It harvests demand that already exists and it is the first thing any brand should get right on Amazon.
Amazon DSP is a programmatic buying platform. It places display and video inventory across Amazon owned properties like Fire TV, Twitch and IMDb, plus a wide pool of third party sites, and it targets using Amazon's purchase and browsing data rather than social interest graphs. That data is the reason operators care. Meta knows what you like. Amazon knows what you bought, how often you rebuy it, and which competitor's detail page you were sitting on last Thursday.
For a DTC brand that has an Amazon presence, this creates two jobs DSP is genuinely good at. The first is rescuing leaked demand: a shopper viewed your detail page, added to basket, and walked. The second is stealing considered demand: a shopper is deep in a competitor's product pages in your category and has not yet committed. Everything else DSP can technically do, it usually does worse or more expensively than another channel you already run.
DSP is not a demand creation machine for a brand nobody has heard of. It is an efficiency layer on top of demand that already exists. If nothing is leaking, there is nothing for it to catch.
The Benchmarks Worth Holding Yourself To
Before touching the account, know what good looks like. These are the numbers to judge line items against in 2026.
Retargeting ROAS
4:1 to 8:1
Cart abandoners are the strongest single segment at roughly 0.68 percent click through rate and 4.2x return. If a retargeting line item is under 3x, the audience is contaminated, not the creative.
Prospecting ROAS
2:1 to 3:1
Lookalike audiences run near 0.38 percent click through rate and 2.1x return. Prospecting on DSP is a margin decision, not a performance one. Only fund it if contribution margin carries a 2x.
Frequency cap
3 to 5 per user, per day
Past this, click through rate falls while cost per thousand holds. You are paying more to annoy the same shopper. Cap it at the line item level, not just the campaign.
Incremental lift
1.4x reach, 26% better purchase rate
Measured against Sponsored Ads alone, with detail page view rates roughly 45 percent higher. This is the number that justifies the channel, not platform reported ROAS.
The Three Ways DSP Budget Leaks
Waste in a DSP account is rarely dramatic. It does not show up as a failed campaign. It shows up as a channel that reports well and changes nothing. There are three forms of it and most accounts carry all three.
Duplicate reach
Two line items paying separately to reach the same person. It happens because audiences get built by theme rather than by exclusion, so the cart abandoner segment, the category browser segment and the lookalike segment all contain the same shopper, and you buy that impression three times.
The fix is a hierarchy. Decide which audience has the highest intent, serve that one, and explicitly exclude its members from every audience below it. Review overlap monthly, because audiences drift as the account scales.
Dead reach
Spend on people with no path to purchase. Recent buyers of a product they will not need for ninety days. Active subscribers who are already receiving the item monthly. Shoppers who have absorbed a dozen impressions and clicked nothing.
This is the expensive one and it is entirely preventable. Every account needs a purchaser exclusion rebuilt weekly, a subscriber exclusion, and a frequency cap. A seven day stale exclusion list is seven days of paying to advertise to your own customers.
Cannibalised reach
Spend on shoppers who were already converting through Sponsored Products. The DSP dashboard happily claims those orders, which is why the reported ROAS looks excellent while blended acquisition cost does not move a millimetre.
The test is simple and uncomfortable. Hold out a segment, run the channel without it for a fortnight, and compare total orders rather than attributed ones. If total volume is flat, you were paying twice for the same customer.
When You Should Not Run DSP At All
Self service access typically requires a commitment around 35,000 dollars, and managed service sits higher. That alone rules out a lot of brands, but the spend floor is not the real qualifier. The real qualifier is whether you have leaked demand worth catching.
If your annual Amazon revenue is under one to two million dollars, or you are not yet spending around ten thousand dollars a month profitably on Sponsored Ads, the money is better spent on search efficiency and detail page conversion. A brand with a 9 percent detail page conversion rate does not have a retargeting problem. It has a listing problem, and DSP will simply buy more traffic for a page that does not convert.
The order of operations matters more than the channel. Fix the listing, then fix search, then use DSP to catch what those two leave behind. Running it earlier is an expensive way to find out your detail page was the problem all along.
What This Looks Like In Practice
The build order we use is deliberately unglamorous. Exclusions first, before a single line item goes live. Purchasers within the repurchase window, active subscribers, and anyone converting on branded search all come out. That one step usually reclaims a visible slice of budget in the first fortnight and it does not require a new creative asset.
Then the retargeting ladder gets built by intent, highest first. Cart abandoners in the last 7 days, then detail page viewers in the last 14, then category browsers in the last 30, with each tier excluding the one above it so nobody is bought twice. Budget follows that order, not the other way round.
Prospecting only gets funded once the retargeting tiers are clearing their benchmark and the holdout test says the channel is producing incremental orders rather than reclassifying existing ones. Most brands never need to go further than that. The money is in the rescue, not the reach.
Inside the system
How we build this for brands
The suppression layer is the part nobody wants to own manually, because it is only correct for as long as it is fresh. We run a reporting agent against live Shopify and ad data that rebuilds purchaser and subscriber exclusions on a schedule, flags audience overlap between line items, and surfaces the moment frequency climbs past the point where click through rate starts falling. The same agent writes the weekly summary, so the leak gets caught in days rather than at quarter end.
Around it sits the rest of the stack that decides whether DSP has anything worth catching: a VOC engine that mines reviews and support messages into the objections the detail page has to answer, creative built from that language rather than from a brief, and profit dashboards that judge the channel on contribution margin instead of platform reported return. Part of this runs live for portfolio brands today; the full system is what we deploy when we take a brand on.
Amazon Growth Audit
Find Out How Much Of Your Retargeting Budget Is Reaching Existing Customers
I will go through your Amazon account line item by line item, show you where the exclusions are missing, and tell you whether DSP is the right next channel or whether the money belongs somewhere else first. Numbers and a plan, not a pitch deck.
Book Your Amazon AuditFrequently asked questions
What is Amazon DSP and how is it different from Sponsored Products?
Sponsored Products is search advertising. It puts your listing in front of someone already typing a query on Amazon. Amazon DSP is programmatic display and video buying that reaches shoppers off search and often off Amazon entirely, across Fire TV, Twitch, IMDb and third party sites, using Amazon purchase data to build the audience. Sponsored Products harvests demand that already exists. DSP rescues demand that leaked or targets shoppers deep in a competitor listing.
What ROAS should I expect from Amazon DSP?
Retargeting should land between 4:1 and 8:1. Prospecting typically delivers 2:1 to 3:1. Cart abandoners are the strongest segment at roughly 0.68 percent click through rate and 4.2x return, while lookalikes sit nearer 0.38 percent and 2.1x. A retargeting line item under 3x almost always has a contaminated audience rather than weak creative.
How much does Amazon DSP cost to run?
Self service accounts generally require a minimum commitment around 35,000 dollars, with managed service higher. If annual Amazon revenue is under one to two million dollars, or you are not yet spending ten thousand dollars a month profitably on Sponsored Ads, DSP is the wrong next move. Fix search efficiency and detail page conversion first.
What is audience suppression and why does it matter?
Suppression means excluding shoppers who should not see a given ad, most commonly recent purchasers, active subscribers and people already converting through Sponsored Products. Without it you pay to advertise to your own customers and reported ROAS inflates, because those shoppers were going to buy anyway. Purchaser exclusions, subscriber exclusions and a frequency cap should all be live before the first prospecting pound is spent.
Should DTC brands run Amazon DSP alongside Meta retargeting?
Yes, but measure them together. Brands running DSP alongside Sponsored Ads see roughly 1.4x higher incremental reach, 45 percent higher detail page view rates and 26 percent better purchase rates than Sponsored Ads alone. Comparing platform reported ROAS across Amazon and Meta is misleading because attribution windows and models differ. Judge the combination on blended new customer acquisition cost and contribution margin.
How often should Amazon DSP audiences be refreshed?
Rebuild purchaser exclusions weekly at minimum, since a stale list means days of paying to retarget people who already bought. Review audience overlap monthly, because duplicate reach is the quietest form of waste. Refresh creative on a fatigue trigger rather than a calendar, typically when frequency passes three to five impressions per user per day and click through rate falls while cost per thousand holds steady.
About the author
Caner Veli is a DTC operator who has helped 350+ brands fix broken growth engines. He built Liquiproof from zero to 3,000+ global retailers in under 6 years. He now runs the same playbook, supported by AI systems he built himself, for DTC and CPG brands.