Amazon DSP Agency Comparison: Score Them Yourself

Read the comparison lists knowing who wrote them

Search this phrase and you get ranked lists of agencies. Look at the domain each list is published on, then look at which agency sits at number one. In a striking number of cases they are the same company. That is not a scandal and nobody is being deceived on purpose — an agency writes an article about its market, and its own view of that market naturally puts itself near the top. The format simply does what it was built to do.

The rule that follows applies to this page too. reMKTR is an Amazon DSP agency and this page is published on our own domain. So treat everything below as a method rather than a verdict, use it on us as readily as on anyone else, and give more weight to what a page will tell you about its own limits than to where it places itself.

What genuinely useful comparison content looks like: it defines the criteria before naming any company, it says which situations each option loses in, and it hands you something you can run without the author present. That is the shape of the rest of this page.

One disambiguation first, because the acronym collides. Amazon DSP is the demand-side platform — the programmatic buying product for display, video and audio. Amazon Delivery Service Partner is the courier franchise programme with vans, routes and drivers. If you wanted the delivery programme, you want Amazon's logistics site and not an advertising agency.

Seven dimensions, weighted

Score each candidate out of five on each dimension, multiply by the weight, total it. The weights below reflect what actually moves the outcome on a DSP engagement; adjust them if your situation genuinely differs, but adjust them before you see the proposals rather than afterwards.

• Measurement design — weight 5. Can they describe a specific holdout they ran, what was withheld, from whom, for how long, and what it showed? A partner who has never withheld display from a matched group has never tested whether display works.
• Audience craft — weight 4. How they build remarketing, in-market, lifestyle and conquesting segments, how often they rebuild them, and what they do with your own first-party lists.
• Fee architecture — weight 4. The rate, the basis it applies to, what sits inside it, whether it is capped, and where it steps down. The undisclosed variable matters more than the headline.
• Team continuity — weight 4. Who runs the account weekly, how many other accounts they hold, and whether you meet them before signing.
• Inventory control — weight 3. Which placements run on and off Amazon, what is excluded by default, and who decides.
• Creative capability — weight 3. Display and video units in the right sizes, refreshed on a schedule, and whether production is inside the fee or billed on top.
• Exit terms — weight 3. Seat ownership, audience portability, clean-room instance ownership and notice period.

Anything scoring one or two on measurement design should be disqualified regardless of its total. It is the dimension that determines whether every other number you are shown means anything.

The three fee architectures you will meet

Almost every proposal is a variation on one of three, and the differences between them matter more at the edges than in the middle.

• Percentage of media spend. The most common structure in this category, and the one we use. It aligns the agency with growth and it scales automatically. Its weakness is that it also scales when spend rises for reasons that have nothing to do with the agency's work — so ask for a cap, or a step-down schedule, in writing.
• Flat monthly fee. Predictable for finance, and it does not punish you for growing. Its weakness is the opposite: at low spend it is expensive per dollar managed, and at high spend the agency's incentive to push for more budget disappears.
• Hybrid — a base plus a percentage. Covers the fixed cost of servicing an account and keeps some upside alignment. Its weakness is that two numbers are easier to obscure than one, so make sure both are named and both are capped.

None of these is dishonest. What causes trouble is not the model but the parts left unstated: whether creative production, clean-room analysis and audience development are inside the fee, and what the percentage is calculated on — media only, or media plus platform costs. At $100,000 a month in spend, a single undisclosed percentage point is $12,000 a year. Ask the question in the first meeting, not the fourth.

Minimums: what Amazon requires versus what agencies require

These get conflated constantly, and the conflation costs mid-market brands months of assuming they are not eligible.

Amazon's own managed service — where Amazon's team runs the campaigns — is described on Amazon's product page as typically requiring a minimum investment of USD 50,000, with the note that the figure varies by country. That is a real floor, and it is Amazon's, not an agency's.

Self-service is a different arrangement: the advertiser keeps full control of campaigns after setup. Access arrangements have widened over the past year, so confirm the current position with Amazon rather than with a two-year-old article.

Agencies holding their own seats set their own minimums, and those are generally lower and more negotiable than the managed-service floor. Ask each agency for its minimum monthly media spend in writing, and ask separately whether there is a minimum fee that applies underneath it, because those are two different constraints and only one of them usually gets quoted.

How to tell real evidence from a good slide

Every agency will show you a campaign with a large multiple on it. Three tests separate the ones that mean something.

Ask for a denominator. A return on ad spend without a scope is decoration. "6.04x" means one thing across a whole book and something else entirely across one campaign in one month. Our own figure is 6.04x measured across 30 advertisers in July 2026, and we scope it that way every time precisely because the scope is what makes it checkable.

Ask what the number excludes. A blended cost per click of $1.42 across a full book will always look worse than a video-only figure from the same book. Cheap-looking metrics are frequently just narrower ones.

Ask about double counting. If display and sponsored ads are both attributed the same purchase, the blended return is inflated and usually nobody has noticed. The reconciliation happens in Amazon Marketing Cloud or it does not happen. This is the question that most reliably separates agencies who measure from agencies who report.

Running the comparison in two weeks

A structure that produces comparable proposals rather than three differently-shaped brochures.

• Days 1–2. Write one brief and send it to every candidate unchanged: category, current Amazon revenue, current ad spend by type, target media budget, margin reality, inventory position, and the one business question you want display to answer.
• Days 3–7. Require every proposal to answer the same seven questions in the same order, and to state fee basis, inclusions, minimum term, notice period and minimum spend on one page.
• Days 8–10. Meet the delivery team, not the pitch team. Ask each named person how many other accounts they hold.
• Days 11–12. Two reference calls each, and ask for one client who left. What they say about that departure is the most informative twenty minutes of the process.
• Days 13–14. Score, weight, total. Then read the exit clauses before you look at your own scores, so the numbers do not talk you out of a bad contract.

What reMKTR looks like against this sheet

Fair is fair — the method above applied to us, including where it is unflattering.

reMKTR is the Amazon DSP arm of Full Circle, a full-service Amazon management company with $500M+ in managed spend across 100+ brands. We run 109 live Amazon DSP advertiser seats. Across 30 advertisers in July 2026 the book delivered 6.04x on ad spend, 78.4 million impressions at a $4.00 CPM, a blended $1.42 cost per click, and a $5.49 cost per acquisition across 57,137 attributed purchases with 20.1% new to brand.

On measurement design we would score ourselves highly and invite you to test it: last-click attribution cannot prove incrementality and never could, holdouts and matched controls can, and we reconcile in Amazon Marketing Cloud so display and sponsored ads stop double-counting each other. On fee transparency we would score ourselves mid-table — we publish a percentage-of-media model and quote the rate before you sign, but several firms in this market print more detail publicly than we do. On breadth we score low by design: if you need catalogue, listings and logistics, we are the wrong hire.

Two places to go instead when that is the case. Dr. DSP — when you would rather buy DSP as a product with a defined scope than run an agency selection at all. Dr. PPC — when the comparison you should really be running is on sponsored ads, because display sits on top of search and amplifies whatever search is already doing.