Whose seat is it, and why that comes before strategy
Amazon answers a question most agency pages skip. Its own product page says the platform is available to brand advertisers, ad agencies and tool providers, and that there are two ways in. Self-service, where in Amazon's words customers "are in full control of their campaigns". And managed-service, which "typically requires a minimum spend of USD $50,000 (minimum may vary per country)". For a brand doing £20,000 to £100,000 a month on Amazon UK, the second door is not open. So your programmatic runs inside somebody's self-service seat, and the first question is whose.
It matters more than it sounds, because the seat is where the campaign history, the audiences and the measurement live. Two publishers describe the mechanics plainly, and neither of them is an agency selling the service. A transition checklist published by Velocity Sellers in June 2026 puts it in a line: "DSP and AMC access, if you run them. These are frequently agency-seat-owned and don't transfer cleanly." Its advice is to write down, asset by asset, who owns and who can reach each one, because "everything you don't own is a negotiation point, and your leverage is highest before you announce you're leaving".
The second is newer and less comfortable. SellerApp published a piece this month describing white-label DSP, where an agency resells another provider's programmatic seat under its own branding, and where one of the structural decisions is "whether each client gets an isolated workspace with separate budgets and creative, or whether accounts share infrastructure under one umbrella seat". That is ordinary industry practice and there is nothing improper about it. What is striking is that we read the first page of results for three separate versions of this query and not one agency page mentioned any of it.
So ask before you sign rather than after. Whose entity is the advertiser account created under, is the seat the agency's own or resold from somebody else, does your account sit in its own workspace, what happens to the audiences and the campaign history if the engagement ends, and what notice does either side give. We went looking for an Amazon-owned page setting out how a DSP advertiser account moves between agencies and could not find one, and no publisher we checked states a standard minimum contract term for this work. That is a genuine hole in the public record rather than something being withheld from you, which is exactly why the answer has to arrive in writing from whoever is going to run it. Ours included.
None of that is the question of whether to start at all. That one has its own post, with the published minimums, the sixteenfold spread between them and the cases where the honest answer is not yet: when Amazon DSP is worth it, and when it is not. This page assumes you have read it and decided.
What we run, and the number we do not publish
DSP comes last in our sequence, and that is a decision rather than caution. Programmatic earns its place when there is enough demand to retarget and enough margin to fund prospecting, and both of those are made by the ranking, catalogue and sponsored work underneath it. On the collagen brand Ancient + Brave, DSP is where the step change happened: custom audiences bringing in new customers at a positive return, retargeting to bring existing customers back, Fire Tablet and Prime TV placements, and imagery built for programmatic rather than cropped from the listing.
It runs as one of five disciplines rather than as a separate media buy: RippleTarget custom audiences, new-customer acquisition campaigns, retargeting and basket recovery, AMC reporting and custom builds, Fire Tablet and Prime TV ads, and creative made for the format instead of inherited from the product page. Our reason for keeping it in the same hands as the listing and the sponsored account is that DSP is the discipline most able to disguise a problem sitting somewhere else. A programmatic budget will buy traffic to a page that does not convert, and the report will look like advertising working.
We do not publish a DSP-specific figure for any account and we are not going to estimate one here. Retargeting and prospecting are two different jobs, a blended return hides which one was done, and we have made that argument in public often enough that inventing a number for our own service page would be poor form. What the published record supports is the sequence and the direction, not a figure attached to this stage alone.
Three kinds of audience, and only one is built from your own data
Amazon sets out the shape of it plainly. The platform "offers audience targeting in three main categories: 1) Amazon Audiences, based on first-party buying, browsing and streaming signals; 2) advertiser audiences, driven by custom inputs like product relevance and remarketing; 3) third-party audiences, sourced from broader internet activities". Those three are not interchangeable, they are not equally hard to get right, and only the middle one is assembled out of anything you own.
So the thing to demand is that your monthly report says which of the three your money went to. A campaign leaning on remarketing is advertising at people who already found you; one leaning on Amazon's in-market signals is buying people who have not heard of you. Both are legitimate and a growing account usually needs both. Only one of them brings new money in, though, and the case for keeping the two apart in the reporting, with the published return ranges either side of it, is set out in full here.
Amazon names the metrics. Insist on its names.
The reporting vocabulary already exists, it belongs to the platform, and it costs nothing. Amazon says campaign reporting "includes industry-standard metrics and Amazon-proprietary metrics such as detail page view rate (DPVR), add-to-list (ATL) counts and new-to-brand (NTB) metrics, along with reach, frequency and viewability metrics", and that you can "conduct custom analytics across your Amazon DSP campaigns using event-level data sets in Amazon Marketing Cloud". Every one of those is a standard field rather than a bespoke build.
Which is why a DSP report carrying one blended return figure is not describing a limit of the platform. New-to-brand is there for the asking. If it is missing, either nobody built the report or somebody decided the split was better left out. Ask for detail page view rate, add-to-list, new-to-brand and frequency, month on month, per campaign rather than for the account as a whole. Our own view is that frequency catches waste fastest of the four: it says how many times the same person saw the ad, and frequency climbing while new-to-brand sits flat is a budget buying the same audience twice.
Amazon publishes the job description, and it is eight and a half hours long
Amazon runs two certifications for this product and publishes both syllabuses on the same page as the rest. The beginner Amazon DSP Certification covers "Amazon DSP audience solutions, supply quality, campaign setup, supply, ad policy and campaign performance and optimisation" and runs 4 hours 2 minutes. The advanced one covers "bid strategies, conversion tracking, audiences, third-party integrations, private marketplace deals, attribution and bulksheets" and runs 4 hours 32 minutes. Eight hours and thirty-four minutes, beginning to end, is the entire published curriculum of a discipline the market sells as specialist.
Use the advanced list as an interview script, because that is effectively what it is. Ask whoever wants to run your DSP to talk you through private marketplace deals and attribution with nothing in front of them. Those two are where an operator either has the repetitions or does not, and neither survives long in a conversation with somebody who has only sat the course.
We are not going to pretend the knowledge is scarce, because you can establish in an afternoon that it is not. What you are paying for is someone running it every week on an account shaped like yours, and more usefully someone who has decided not to run it at the points where it would not have worked. That is a claim worth making any agency evidence, this one included.
Where we are the wrong answer
If you are already spending at the managed-service level, Amazon's own account team alongside a specialist trading desk is a real option and we would say so rather than compete with it. We are a five-discipline consultancy that runs programmatic inside a whole account. Above a certain scale that is a different product from a programmatic buying shop, and pretending otherwise would waste a quarter of your time.
If you want DSP bought as a standalone media line, held apart from the listing and the sponsored account, we are the wrong supplier for that too. And if the listing does not convert yet, or the sponsored account has never been split between defending your own name and buying new demand, our answer is going to be not yet, which is the least profitable sentence an agency can say and most of the reason the audit is free.
What it costs to have run is published rather than quoted. The retainer and commission are on the site with a calculator, in dollars as well as pounds. DSP sits inside that fee where it is warranted rather than being sold as an upgrade, and the commission is charged on your sales rather than on your advertising spend, so a larger programmatic budget does not pay us more. Media is paid by you, directly to Amazon. Whether your account is anywhere near the point where this is worth starting is one of the first things the free audit answers, and it answers no often enough to be worth having.
Brands we've grown
