Amazon DSP self-service — what the seat requires, what it really costs, and who should actually take one

What a self-service seat actually is

First, the disambiguation this site makes on every page, because the search results genuinely confuse the two: DSP here means demand-side platform — the software used to buy Amazon's programmatic display, video and audio inventory. It is not Amazon's Delivery Service Partner programme, which is about vans and delivery routes. If you searched "amazon dsp self service" looking for the courier business, this is the wrong page, honestly and completely.

Amazon DSP is one layer of Amazon's retail media stack — the programmatic layer that buys audiences on a CPM across Amazon-owned properties like Fire TV, Twitch and IMDb, plus the open web and app exchanges Amazon has inventory access to. There are three ways to buy it, and "self-service" is one of them:

• Self-service. Your team holds the console. You build the campaigns, pick the audiences, set the bids, rotate the creative, watch the pacing, pull the reports. Amazon describes this option as giving advertisers full control with no management fee from Amazon.
• Managed service. Amazon's own team runs the buying for you. Amazon's page states this option typically requires a minimum spend of $50,000, varying by country.
• An agency or partner seat. A third party that already holds DSP access operates it on your behalf, under whatever fee structure you negotiate with them.

The word "self-service" suggests a sign-up page and a credit card form. That is not how this works, and the gap between the name and the reality is most of what this article is about.

Requirements and minimums — what is actually published, and what is actually true

Here is what you can verify on Amazon's own pages: the managed option carries that typical $50,000 minimum, and the self-service option carries no management fee from Amazon and no published spend minimum. Read quickly, that sounds like self-service is the cheap door. Read carefully, and notice what is missing: there is no public sign-up flow for a self-service seat. Access runs through a conversation with Amazon Ads — typically an account executive — or through a partner that already holds a seat. Amazon decides who gets one, and the budget expectations discussed on that call function as the real entry bar, whatever the marketing page implies.

That access conversation is worth preparing for the same way you would prepare for any vendor call. Ask what spend level makes the seat viable in your marketplace, whether audience and supply costs are billed inside your media or on top of it, what reporting you get natively versus what requires Amazon Marketing Cloud, and what support you actually receive as a self-service advertiser — because "self-service" describes who does the work, and the honest answer is: you do.

One more moving part: the console itself is mid-evolution. Amazon has been rolling out a rebuilt DSP experience — simplified campaign creation, more automation in optimisation, the Performance+ tactic — which we cover separately in Amazon DSP 2.0. The refresh lowers the button-clicking burden. It does not lower the judgment burden, and the judgment burden was always the expensive part.

Self-service vs managed service vs agency seat

The three doors buy the same inventory through the same underlying platform. What differs is who does the work, who holds the knowledge, and what shape the cost takes.

Self-service trades money for time. You avoid a service layer, and in exchange every hour of setup, optimisation, troubleshooting and reporting comes off your own team's calendar. You keep every scrap of institutional knowledge — audience learnings, creative results, seasonal patterns — inside the building, which compounds over years and is the strongest genuine argument for going in-house. You also keep every mistake in-house, at full price, while you learn.

Managed service trades control for coverage. Amazon's team executes, you direct. It suits advertisers who clear the minimum and want the platform's own operators, and it comes with the obvious structural caveat: the party spending your budget and the party grading the spend are the same company, so insist on measurement you can audit — more on that below.

An agency seat sits between the two, and quality varies more than in either other door. The good ones bring cross-account pattern recognition your in-house hire cannot have. The weak ones bring the things this site keeps warning about: a retainer that bills whether or not anything happened, creative fees stacked on top of media, "set it and forget it" autopilot, and a monthly dashboard built to look impressive rather than to be checked. None of those failure modes are about how the fee is calculated — plenty of well-structured fees fund bad service and vice versa. They are about what the fee buys. Whatever structure you are quoted, get the basis, the term, the notice period and the data-access terms in one written document, and confirm you own your campaign history if you leave.

What running it in-house truly takes

This is the section most "how to get Amazon DSP self-service" articles skip, because it is the part that talks people out of the click. Running a seat is a job. Not a task — a job.

People. Someone's calendar has to actually contain the console. At minimum that is one person with real, protected weekly hours who owns campaign structure, audience strategy, bid and frequency management, pacing, and the weekly read of what moved. On top of the trader you need analyst capacity — someone who can work in Amazon Marketing Cloud, or at least read its outputs honestly — and a creative supply line. In small teams these can be one heroic person. They cannot be zero people, and they cannot be a person doing it in the gaps around a full-time job; the accounts we inherit most often are the ones where DSP was somebody's fourth priority.

Hours. Launch weeks are heavy: account setup, pixel and audience plumbing, creative approvals, and the first rounds of learning-phase turbulence. Steady state is lighter but never zero — plan for the seat to consume a meaningful slice of every working week, because pacing drifts, audiences fatigue, creative wears out and inventory shifts whether or not anyone is watching. A seat nobody has time to drive does not produce a discount. It produces the 48.5% problem in the stats above, wearing a display costume.

Creative. DSP is a visual medium with real spec discipline — sizes, formats, video lengths, moderation rules — which we keep current in the creative specs guide. Amazon's responsive eCommerce creative can auto-assemble ads from your listings, and it is a legitimate way to start, but a programme that never graduates beyond auto-generated units is leaving the medium's actual leverage — testing distinct messages against distinct audiences — unused. Budget for a designer's time, whether in-house or freelance, and for video if streaming TV placements are part of the plan.

Knowledge. Amazon's DSP certification is free and worth doing before you commit — it maps the vocabulary and the campaign model. Just be clear-eyed about what a certificate is: proof someone studied the console, not proof anyone is staffed to sit in it daily.

The cost math nobody puts in the deck

The self-service pitch usually compares one line: service fee versus no service fee. The honest comparison is total annual cost, and it has more lines than that.

A seat assumes someone competent drives it daily; a service fee includes that person. So convert every option to the same unit before you choose: total cost per year at your real spend, people included. For in-house, that means the loaded cost of the trader's hours, the analyst's hours, the creative production, the training ramp, and — the line everyone forgets — the three-to-six months of tuition the account pays while a first-time operator learns on live budget. For a managed or agency option, it means the full fee stack in writing: basis, minimums, term, notice period, onboarding, data egress. We are deliberately not printing salary figures or invented "typical cost" tables here, because a fabricated number you can anchor on is worse than no number. Build the math from your own payroll and your own quotes; the framework is the part that transfers.

Run that math and many teams discover the gap between doors is smaller than the fee line suggested — in either direction. A brand with a strong existing media team may find the seat is nearly free capacity. A brand with nobody spare may find the "free" door is the most expensive one on the list.

However you run it, measure it the same honest way

The buying model — self-service, managed, agency — changes who clicks the buttons. It does not change what counts as proof, and this is where most DSP programmes fail regardless of door.

Channel ROAS out of the console is not proof. DSP reporting credits view-through and click-through conversions inside an attribution window, which means it will happily claim shoppers who were already on their way to buy — and when sponsored ads run on the same account, both channels can claim the same purchase in their separate dashboards. Adding those two numbers produces a business that looks like it prints money and isn't. The fixes are boring and specific: reconcile DSP and sponsored ads in Amazon Marketing Cloud, where one purchase is attributed once across the account — Amazon offers AMC to eligible advertisers at no licence cost, so the constraint is analyst time, not budget — and test incrementality with a holdout or matched control rather than reading a lift off a line chart. If that methodology is new, start with what incrementality testing actually is; it is the difference between a display budget that adds sales and one that annotates them.

If you take one sentence from this page into a vendor call or a board meeting, take this one: ask how the programme will prove incremental sales, and treat "our ROAS is strong" as a non-answer. A self-service team, Amazon's managed team and any agency worth hiring should all be able to describe a test design — what's held out, for how long, at what spend, read where. Whoever cannot is offering you a dashboard, not a measurement.

A decision framework you can score yourself on

Strip the vendor noise away and in-house self-service is the right call when four things are simultaneously true: the budget is big enough to generate readable data, a named person has real hours to run the seat, creative can be produced on a cadence rather than once, and the measurement muscle — AMC, holdouts — either exists or is genuinely being built. When one is missing, a hybrid works: an agency or managed seat now, structured so learnings and data stay portable, while you hire and train toward taking the seat later. When most are missing, taking a seat anyway doesn't save the service fee — it converts it into quiet media waste, which is more expensive and harder to see.

The tool below turns that into six questions with weighted scoring. It runs entirely in your browser, sends nothing anywhere, and is deliberately hard to flatter yourself on. Answer it as the team you have, not the team you intend to hire.

Where Dr. DSP fits

Dr. DSP is a fourth door, and we will describe it with the same flatness we used on the other three. It is Amazon DSP — the demand-side platform, not the delivery franchise — run daily by Fable 5 with operators from Full Circle, a full-service Amazon management company with more than $500M in managed revenue across 100+ brands, supervising. You choose the approval level, every change ships with its evidence and its rollback trigger, results are reconciled in Amazon Marketing Cloud rather than read off channel ROAS, and Orbit is included at no extra cost. There is no published price: a demo, the first 30 days free, then a price set on the call against your real budget and scope, month to month.

For a yardstick rather than a promise: across 30 advertisers in July 2026, the pull showed 6.04x return on ad spend, 78.4 million impressions at a $4.00 CPM, $5.49 cost per acquisition across 57,137 attributed purchases and 20.1% new-to-brand. One month of Amazon-attributed data, scoped honestly, guaranteed to nobody.

And the redirect we owe you: if the readiness score above came back strong, take the seat. A brand with the budget, the trader, the creative cadence and the AMC muscle should run its own DSP, and nothing we sell changes that arithmetic. This page exists for the much larger group in the middle — where display is worth doing, nobody owns it as a job, and the honest choice is between a service and a slow leak.