DSP gets pitched to every brand the day it crosses $1M on Amazon. The pitch usually frames it as “Sponsored Display on steroids.”
That framing is wrong.
Sponsored Ads catches demand that already exists. DSP creates demand. Different pricing model (CPM, not CPC). Different funnel stage. Different targeting logic. Different surfaces – DSP runs across Amazon-owned properties and the open internet, while Sponsored Ads stays inside Amazon search and product pages.
You can run DSP without selling on Amazon at all. Banks, insurance brands and telcos use it because Amazon’s audience data is some of the best in the world, whether the conversion happens on Amazon or on your own site.
The single thing most teams get wrong is treating DSP as a performance lever. It is not. It is an awareness machine that pays back in branded search lift, in cheaper SP CPCs over 90 days, and in a wider top of the funnel feeding Sponsored Products downstream. Read the wrong report and DSP looks like it is failing. Read the right report and you see what it is actually doing.
How DSP works under the hood
Someone opens a webpage. An ad slot becomes available. Multiple DSPs receive a bid request describing the impression. Each evaluates whether the user behind that impression is worth bidding on. Each submits a bid or skips. Highest bid wins.
The whole thing takes about 100 milliseconds. Faster than a blink.
That is Real-Time Bidding. And the primary benefit of RTB is not that it saves money or that it bids fast. It is one-to-one targeting. Per impression, per user, evaluated individually.
The old way of buying ads was bulk. You bought 1,000 impressions on a website and you got whatever 1,000 users showed up. RTB lets you evaluate every single impression and decide if this specific user, in this specific moment, on this specific page, is worth the bid.
That is the only reason DSP outperforms display bought any other way. Strip the one-to-one targeting and DSP collapses into expensive display.
Four ways to buy inventory
RTB is the most common method. It is not the only one.
- Open Auction. Per-impression bidding, anyone can join. Cheapest, lowest priority, mixed quality.
- Private Auction. Per-impression bidding, invited advertisers only. Better quality.
- Preferred Deal. Fixed CPM, no auction. You get first right to inventory at an agreed price. You can skip it (then it falls back to open auction). Once you accept, no one outbids you.
- Programmatic Guaranteed. Fixed CPM plus guaranteed delivery. The publisher must deliver. You must pay. Most premium, most expensive.
Think of Preferred Deal as pre-booking a hotel room at a fixed rate. Demand can spike. Your booking is locked. The hotel cannot cancel and give your room to someone paying more.
Most brands run Open Auction for roughly 90% of DSP spend and use the PMP options only for high-stakes flights – a launch week, a tentpole event.
The audience taxonomy that trips most people up
DSP audiences come from three sources. Get the bucket wrong and you target on the wrong signal.
Retail Audiences come from Amazon shopping data. What people search for, browse, add to cart, buy. In-market shoppers, lifestyle audiences (foodies, gamers, fitness), life-event audiences (new parents, new movers), demographic cuts.
Media Audiences come from Amazon’s entertainment properties. Prime Video, IMDb, Twitch, Kindle, Fire TV. What people watch, read, listen to. Content signals, not shopping signals.
Advertiser Audiences come from your own data. Pixel-tracked website visitors. CRM uploads. App users. Anyone you can identify from your own systems.
The most common confusion: people see “Prime Video” listed under audiences and assume Retail, because Amazon also runs Amazon.com. It is not Retail. Prime Video viewing patterns sit in Media, separately from shopping. Get this wrong and your audience math goes off.
The next-buyer principle
Here is where most operators waste DSP budget.
They build audiences from past purchase data – people who bought board games already – and target them with more board game ads. Those people already bought. The conversion rate on showing them the same category again is poor.
The right question is not “who buys my product?” It is “what does someone buy right before they need my product?”
Sell toddler learning games? The audience that converts is not yesterday’s toddler-game buyers. It is people who bought baby formula or diapers six to twelve months ago. They are about to age into your product. Target them now with a brand-awareness flight and you build demand for the moment they go looking.
This shift, from last-buyer logic to next-buyer logic, is the single biggest unlock most teams get when they finally understand the audience taxonomy.
How data gets into DSP
Four ways to bring data in:
- Pixel (Amazon Ad Tag, AAT). JavaScript on your website. Browser-based. Roughly 60-70% reliable, because ad blockers, Safari and iOS restrictions strip out a chunk.
- CAPI (Conversion API). Server-to-server event sending. About 95% or higher, because it bypasses the browser entirely. On Shopify, the Amazon integration is CAPI under the hood. Already wired.
- MMP (Mobile Measurement Partner). An SDK inside your app. Tracks installs, in-app events, in-app purchases. AppsFlyer, Branch, Adjust.
- CRM Upload. Hashed customer list. Roughly 50-70% match rate against Amazon’s user base.
They are not interchangeable. CRM tells you who your customers are. MMP tells you what they do in your app. Pixel and CAPI together tell you what they do on your website. Pick the right tool for the question you are asking.
Campaign hierarchy, and the three optimisation features
DSP organises spending into three levels.
- Order is the campaign. It holds the total budget, flight dates, KPI, bidding priority.
- Line Item is one audience-plus-supply combo. Has its own bid settings, frequency cap, daily budget.
- Creative is the actual ad asset.
One Order has many Line Items. Each Line Item has many Creatives. Do not confuse this with the Sponsored Ads hierarchy (Campaign > Ad Group > Ad). Similar in shape, different in what lives where.
Three “optimisation” features do most of the work:
Bid Optimisation adjusts your bid price per impression. The algorithm bids higher for likely converters, lower (or skips) for unlikely ones. Like a smart bargainer at a fish market who pays premium for the good fish and walks past the stale.
Budget Optimisation reallocates spend across Line Items inside the same Order. Watches which Line Items are winning. Shifts more budget toward winners, less toward losers. Like a kirana shop owner who starts with equal stock across products, then week by week stocks more of what sells.
Goal-Based Bidding is the target. You tell the algorithm what KPI you want (a 4x ROAS, say). The algorithm figures out how to bid to hit it. Like telling an Uber driver the destination without telling them the route.
The rule that catches almost everyone: every “optimisation” feature in DSP is automatic by definition. If you read a description that says “lets the advertiser manually control X”, that is the wrong feature. Optimisation in DSP means the algorithm does it without you. If you want manual control, turn the feature off.
Bidding priority
The Order-level setting most operators ignore.
Two options:
- Prioritise Spending Full Budget. The algorithm’s job is to spend the budget, even if KPIs dip. Use this for fixed budget commitments.
- Prioritise KPI Target. The algorithm’s job is to hit the KPI, even if it underspends. Use this when profitability matters more than spend completion.
Pick “Balanced” and you get the worst of both. Pick a job. Let the algorithm do it.
Frequency caps – daily and weekly
Frequency cap is the lever that decides whether DSP earns its keep or burns quietly. Without a cap, the algorithm spends efficiently in the short term by hitting the same user 20 or 30 times in a week. That kills ROAS at the back end, because no human converts on the 25th exposure.
Set both a daily and a weekly cap. Daily prevents same-day repeat blasts. Weekly prevents 7-day saturation. Segment by audience temperature: cold audiences 3-5/week, warm 7-10/week, hot retargeting 10-15/week for a short window. The account-rituals review cadence is where you actually catch frequency creep before it hurts.
Creative is 50% of the variance
Across DSP campaigns, the lever that drives the most performance variance is not targeting and not bidding. It is creative.
Roughly 50% comes from creative. Targeting drives another 30%. Bidding drives the remaining 20%. Most operators invert that and tune bids while ignoring the asset.
Four creative formats matter:
- REC (Responsive eCommerce Creative). Amazon auto-builds the ad from your PDP. Image, price, star rating, Prime badge – all live from the PDP. Links to the PDP. Use when you want pricing and stock data to stay accurate as it changes.
- ABC (Asset-Based Creative). You upload finished assets. Full creative control. Links anywhere – Amazon, brand store, your own D2C site. Use when you want a specific message or off-Amazon traffic.
- BSC (Brand Store Creative). Links to your Amazon brand store, not a single PDP. Use for multi-product browsing.
- CBC (Component-Based Creative). You upload headlines, images and CTAs separately. Amazon mixes them into variants. Use when you want the algorithm to test combinations.
REC is required for PDP traffic, and the reason is not “REC drives better consideration.” It is policy. REC keeps price and stock data accurate between the ad and the destination. Use a static ABC banner with a hardcoded price, the PDP price drifts mid-flight, you land an ad-policy violation.
The most common creative mistake is reusing listing images as DSP banners. Listing images are calm, white-background, designed for someone already comparing options. DSP banners are loud, contextual, designed to interrupt someone scrolling Prime Video. Different jobs, different visual rules. The crossover almost never works.
A healthy creative slate has five variations per audience: product-led shot with price, benefit-led outcome message, social proof with review snippets, urgency or offer for festive moments, and lifestyle showing real-world use. Run all five. Refresh the losers every 30 to 60 days.
The attribution trap
DSP reports look impressive in a way that is partly real and partly inflated. Three things drive the inflation.
View-through credit. Someone who saw your ad but never clicked, then bought your product within the lookback window, gets credited to DSP. Some would have bought anyway.
Brand halo. A user clicks an ad for ASIN A and ends up buying ASIN B from the same brand. The sale of B gets credited as a “halo sale” to the original ad for A.
Long lookback windows. DSP’s default is 14 days. A 14-day view-through window is generous. Someone who saw the ad two weeks ago and buys today gets full credit.
Read your reports honestly. Always ask:
- What is the click-through ROAS only?
- What is the lookback window?
- What percent of attributed sales is view-through vs click-through?
- Are these Product Sales or Total Sales (Total includes brand halo)?
- What is the New-to-Brand percentage?
If any of those signals are present, your DSP ROAS is partly fiction.
The fix is not to abandon DSP. It is to read the honest slice. Tighten the lookback to 1 day for view-through. Look at Product Sales not Total Sales. Pull click-through ROAS separately from blended ROAS. The numbers get smaller and more honest.
The real test of DSP is not the dashboard. It is incrementality. Pause DSP in one geography, leave it running in a similar one, and watch overall sales over 30 days. If the paused geo drops and the running one holds, DSP is earning real lift. If they look the same, DSP was claiming credit for sales that would have happened anyway.
This is the same instinct that runs through the TACoS vs ACoS reading and the spend ceiling math – the metric on the dashboard is rarely the one that ties to the P&L.
Scaling math – what doubling the budget actually buys you
The most expensive misconception about DSP is that adding more budget to a winning Line Item buys proportionally more reach. It does not.
Spend grows linearly. Reach grows logarithmically. The rest of the extra spend goes into higher frequency on the same users you were already reaching. That is exactly the dynamic that collapses ROAS as you scale.
The rule that saves you: scaling DSP is about giving the algorithm more room, not more fuel.
Before you double the budget on a winning Line Item, expand the audience. Add lookalikes built off existing converters. Add adjacent life-stages or income brackets. Add geographies you have not run in. Once the audience is meaningfully larger, then add the budget.
The cleanest scaling pattern is three phases. Weeks 1 to 4: discovery, multiple small Line Items running in parallel. The goal is signal, not spend. Weeks 5 to 8: validation, bigger budgets shifted onto Line Items that earned through discovery. Cut the losers. Week 9 onward: scale, audiences expanded first and budgets following. Hold an 80/20 split between scaling proven Line Items and exploring new audiences.
Where DSP cannot help
Be honest about what DSP does not do.
It does not replace Sponsored Products for capturing existing search demand. Different auction systems, different jobs. SP catches the user typing “leak-proof water bottle” into Amazon search right now. DSP builds the brand awareness that makes them more likely to type your brand name into search next week.
It does not fix poorly-converting product pages. If your PDP converts at 4% and category benchmark is 12%, sending more DSP traffic burns money. Fix the listing first. The 6-step TACoS diagnostic is where that surfaces.
It does not work well below roughly INR 5L per month per brand. Signals too thin. The algorithm spends three months optimising against noise. Worth reading when DSP is worth spinning up before committing.
It does not show ROI in under 30 days. The optimisation curve needs time. Brands that launch DSP on a 14-day pilot conclude DSP does not work, because DSP genuinely does not work in 14 days.
It does not help new ASINs with no social proof. Driving DSP traffic to a 20-review listing burns money – the user lands, sees thin reviews, leaves. Build the review base on SP first. Layer DSP after.
It does not lower competitor-keyword CPC on Sponsored Products. Branded CPC drops over 90 days of DSP awareness – that is real and measurable. But the cost of bidding on a competitor’s brand name in SP stays high regardless.
The honest takeaway
DSP earns its keep when three things are true at once.
The listings it sends traffic to convert at category benchmark or above. The frequency caps are tight enough that the same user is not blasted 20 times a week. The reports being read are click-through ROAS on Product Sales with a 1-day view-through window – not blended ROAS on Total Sales with the default 14-day window.
Get those three right and DSP earns 5 to 15% incremental revenue at a blended ACoS that barely moves the needle, plus a 20 to 40% drop in branded-keyword CPC on Sponsored Products over 90 days as awareness compounds.
Get them wrong and DSP burns money for 90 days, then gets blamed.
The platform is real. The lever works. The mistakes that kill it are operational, not strategic.
If your account is the kind where the easy wins have already gone, an Amazon Account Audit is the right first move before layering DSP on top.