What Amazon DSP actually is (and isn't)
Amazon DSP stands for Demand-Side Platform. It's a programmatic advertising system that buys display, video and audio ad inventory on Amazon's own properties and across thousands of third-party sites and apps. That's the technical definition. In practice, the thing to grasp is that DSP is a different tool from Sponsored Display, and brands that treat it as "Sponsored Display with a bigger budget" are the ones most likely to walk away disappointed.
Sponsored Products, Sponsored Brands and Sponsored Display are auction-based, targeted by keyword or product, and billed per click. DSP works differently. You target audiences instead of keywords, inventory is bought programmatically against real-time bid requests, and billing is usually CPM (cost per thousand impressions). You aren't bidding on a search term. You're bidding on the chance to show an ad to a specific person, wherever they happen to be browsing at that moment.
DSP vs sponsored ads: two different systems
Sponsored ads live inside one well-defined moment: someone is searching or browsing on Amazon, and you're competing for their attention right then. DSP covers the whole customer journey, including the parts that happen nowhere near a search bar. That might be someone streaming a show on Fire TV, listening to the free tier of Amazon Music, or reading an article on an unrelated website that runs Amazon's ad exchange.
That changes what each tool is good at. Sponsored ads convert intent that already exists. DSP can build awareness before intent exists, and win it back after someone has shown interest and left without buying. They sit at different points in the funnel, so one doesn't replace the other. The accounts that get the most from DSP already have Sponsored Products and Sponsored Brands handling the conversion stage well.
We regularly audit accounts where DSP was switched on with the same expectations as Sponsored Products: a target ACoS and a wait-and-see attitude. DSP campaigns judged purely on last-click ROAS almost always look worse than Sponsored Products, because much of DSP's value shows up as extra organic and branded search, which never appears as DSP-attributed sales.
Self-service vs managed-service DSP
There are two ways into DSP. With self-service, you or your agency run the console directly: building campaigns, setting audiences, managing creative and adjusting bids in real time. With managed service, Amazon's own team runs the account for you, usually with much higher minimum spend commitments and less day-to-day control over targeting.
Most mid-sized brands do better with self-service DSP run through an agency partner. You get direct access to performance data, you can iterate on audience segments faster, and you can react to what's working without waiting for a quarterly review. Managed service suits enterprise brands running complex, multi-market programmatic plans who want Amazon's team to carry the operational load.
Whichever route you take, agree up front who owns the advertiser account and the data in it. If an agency runs DSP through its own seat, ask how you'll keep access to campaign history, audiences and reports if you part ways.
The audiences DSP gives you access to
The targeting options are where DSP's complexity pays off. They fall broadly into four groups.
Amazon shopping-signal audiences include in-market segments (people shopping a category right now), lifestyle segments (broader interest-based groups) and category-affinity audiences. These are built from real purchase and browsing behaviour across Amazon, so they reflect what people did, not modelled guesses.
Remarketing audiences are shoppers who viewed your product detail page, or a competitor's, without buying. This is usually the highest-converting audience type in DSP, because you're re-engaging intent people have already shown.
Lookalike audiences are built by Amazon from shoppers who behave like your existing purchasers. They're useful for prospecting once you have enough purchase history to give the model a reliable seed audience.
First-party audiences are lists you upload yourself, such as your own customer emails, matched against Amazon's users for targeting or exclusion. Exclusion is the underrated use: leaving out recent buyers stops you paying to reach people who just bought.
Where your ads actually show up
DSP inventory covers Amazon-owned properties (Amazon.com itself, IMDb, Twitch, Amazon Music's free tier and the Fire TV home screen) and Amazon's third-party publisher network, which reaches well beyond Amazon through the Amazon Publisher Services exchange. By format, that splits into standard display banners, video (including OTT and streaming placements on Fire TV and connected TV) and audio (pre-roll spots on Amazon Music's ad-supported tier).
Video and OTT placements deserve separate thought. In format and cost they're closer to traditional TV advertising than anything else Amazon sells, and they're one of the few ways to buy connected-TV inventory with Amazon's purchase data behind the targeting.
Closed-loop attribution
Programmatic advertising elsewhere on the internet relies on modelled or probabilistic attribution: pixels, cookies and statistical inference about whether an ad exposure led to a sale. Amazon can see the actual purchase in its own systems, so DSP attribution is closed-loop and deterministic. If someone saw your ad and bought the product on Amazon within the attribution window, that conversion is confirmed.
This is DSP's main structural advantage over generic programmatic platforms, and it's worth stating plainly when you discuss budget internally. You're buying reach with the ability to measure, with certainty, whether it led to a transaction on Amazon. Keep in mind what closed-loop attribution doesn't tell you: whether the shopper would have bought anyway. That's an incrementality question, covered under measurement below.
When DSP is worth it (and when it isn't)
DSP earns its budget when at least one of these applies:
- Competitors are winning shoppers from your own detail pages and you need a defensive retargeting layer.
- You're launching in a competitive category and need upper-funnel awareness before sponsored ads can convert efficiently.
- You have enough purchase history to build real lookalike audiences.
- You want to reach customers off Amazon, which nothing else in Amazon's ad stack does.
DSP is premature for brand-new catalogues with no purchase history, for accounts still working out basic Sponsored Products profitability, and for budgets too small to reach the frequency programmatic learning needs. DSP campaigns usually need real budget runway before the algorithm has enough signal to optimise.
A quick readiness check
Before you start, answer these honestly. Are your Sponsored Products campaigns profitable on your core ASINs, with search term harvesting and negatives in place? Are your detail pages strong enough to convert cold traffic, with good images, A+ Content and a healthy review count? Do you have stock depth to support a sales lift for several months? Can you commit to running DSP for long enough to judge it, without switching it off after the first fortnight? If any answer is no, fix that first. DSP sends people to your detail pages, and if those pages don't convert or the product goes out of stock, the spend is wasted.
A sensible first campaign structure
Most brands should start narrow. A good first setup has one remarketing campaign and one prospecting campaign, with separate budgets so you can judge each on its own terms.
For remarketing, split audiences by how recently the shopper viewed your product, for example a short window of a week or so and a longer one of around a month, with recent purchasers excluded. Shorter windows usually convert better and can carry higher bids. Add a separate line for shoppers who viewed competitor products in your category if your budget allows.
For prospecting, start with one or two in-market segments closely matched to your category, and add lookalikes once you have the purchase history for them. Keep frequency caps lower here than in remarketing, and judge the campaign on the upper-funnel measures in the next section.
As an illustration only: a brand with a modest test budget might put roughly two thirds of it into remarketing and one third into prospecting for the first couple of months, then shift money toward prospecting as the remarketing pools get used up. The right split depends on how much traffic your detail pages already get, because remarketing can only reach people who visited.
Common DSP mistakes we see in audits
No frequency capping. Showing the same shopper the same ad 40 times a week builds fatigue, not affinity. Set caps deliberately by funnel stage: higher for remarketing, lower for prospecting.
Treating all creative as interchangeable. Static banners built for Sponsored Display get reused for video and OTT placements, where expectations are completely different. Video has to work with the sound off in the first three seconds and tell a complete story in 15 to 30 seconds.
Judging performance too early. Programmatic algorithms need volume to optimise. Killing a campaign on day 10 because ROAS looks weak, before it has had a real learning window, throws away what you've already spent on that learning.
Ignoring off-Amazon incrementality. Measuring DSP only on Amazon-attributed sales misses the reason for running it off Amazon. Where you can, run a holdout test: pause DSP for a defined market or audience segment and measure the change in branded search and direct traffic.
Retargeting people who already bought. Without purchaser exclusions, remarketing budgets drift toward recent customers who would have bought again anyway, which flatters ROAS while adding little.
How to measure DSP correctly
Judge upper-funnel DSP campaigns (prospecting, lookalikes, OTT and video) on reach, frequency, the lift in branded search volume that follows, and new-to-brand purchase rate. Click-through ROAS is the wrong measure for them. Judge remarketing campaigns on direct ROAS, because that audience already has intent and should convert efficiently. Track new-to-brand metrics closely. DSP's new-to-brand purchase rate is one of the clearest signs of whether the platform is growing your customer base or re-selling to people who already knew you.
For a holdout test, pick something you can separate cleanly, such as one region or one audience segment, and stop serving DSP to it for a few weeks while everything else stays the same. Compare branded search volume, new-to-brand orders and total sales in the held-out group against the rest. If the gap is small, the DSP-attributed sales were mostly people who would have bought anyway. If it's large, DSP is adding sales the attribution report alone can't prove. Brands with access to Amazon Marketing Cloud can run more detailed path-to-purchase and overlap analysis across DSP and sponsored ads, which helps when budgets get large enough that the split between channels matters.
Frequently Asked Questions
It's a different system. Sponsored ads are auction based, keyword or product targeted, and billed per click. DSP targets audiences, buys inventory programmatically against real-time bid requests, and typically bills on CPM. You're bidding on the chance to show an ad to a specific person wherever they're browsing, instead of on a search term.
Last-click ROAS is the wrong yardstick for most DSP campaigns. A meaningful share of DSP's value shows up as extra organic and branded search, which isn't counted as DSP-attributed sales. Judge prospecting, lookalike and video campaigns on reach, frequency, branded search lift and new-to-brand rate, and save direct ROAS for remarketing.
DSP is premature for brand-new catalogs with no purchase history, accounts still working out basic Sponsored Products profitability, and budgets too small to reach the frequency programmatic learning needs. The accounts that get the most value already have Sponsored Products and Sponsored Brands handling the conversion stage well.
Remarketing audiences, meaning shoppers who viewed your detail page or a competitor's without buying, are usually the highest-converting type, because you're re-engaging intent that already exists. Lookalikes and Amazon's in-market and lifestyle segments are prospecting tools, useful once you have enough purchase history to build a reliable seed audience.
