Most sellers know how to evaluate Sponsored Products. You have an ACOS target, you know your break-even ACOS, and you can tell within a couple of weeks whether a campaign is working. The feedback loop is short and the attribution is reasonably clean.
Amazon DSP breaks all three of those conditions. It is priced differently, measured differently, and works on a timescale that makes the usual ACOS instinct actively misleading. Sellers who evaluate DSP the way they evaluate Sponsored Products almost always conclude it doesn’t work — sometimes correctly, often not.
This article covers how to assess DSP as a margin decision.
What Amazon DSP is, and how it differs from Sponsored Ads
Sponsored Ads (Sponsored Products, Brands, Display) are keyword- and product-targeted placements shown to shoppers who are already on Amazon, generally already searching. They are cost-per-click, self-serve, and attributed on a relatively short window.
Amazon DSP is a demand-side platform for programmatic display and video inventory, both on Amazon properties and across the open web and connected TV. It is bought on a cost-per-thousand-impressions (CPM) basis. It targets audiences — behavioural segments, lookalikes, remarketing pools built from Amazon shopping signals — rather than search intent.
The practical differences that matter for margin:
| Sponsored Products | Amazon DSP | |
|---|---|---|
| Pricing | CPC | CPM |
| Trigger | Shopper’s search | Audience membership |
| Funnel position | Lower — active intent | Upper and mid — demand creation |
| Attribution window | Shorter | Longer, and includes view-through |
| Access | Self-serve | Self-serve tier or managed service, with minimum spend commitments |
| Sensible evaluation horizon | Weeks | Months |
Why ACOS is the wrong first metric for DSP
ACOS asks: what percentage of the revenue this campaign generated did I spend to generate it? That question presumes the campaign’s job was to close a sale to someone already shopping.
DSP’s job is usually different. A remarketing campaign to shoppers who viewed and didn’t buy is doing something ACOS can partly capture. A campaign building awareness among a lookalike audience who have never heard of your brand is doing something ACOS captures badly, because the sale — if it comes — arrives later, possibly through an organic search for your brand name, possibly through a Sponsored Products click you also paid for.
That last point is the important one. DSP can inflate the measured performance of your Sponsored Products campaigns by warming demand that then converts through a cheaper lower-funnel click. If you judge DSP on its own attributed ACOS while your Sponsored Products ACOS quietly improves, you will cut the thing that was helping.
The metric that survives this is TACOS — total advertising cost of sales, or total ad spend divided by total revenue. It is indifferent to which campaign gets credit. If DSP is genuinely working, TACOS improves or holds while total revenue grows. If DSP is not working, TACOS rises and nothing else moves.
The profit math
Start from contribution margin, not revenue.
Contribution per unit = Price − Referral − Fulfillment − Storage
− Returns allowance − Landed cost
Then the question DSP has to answer:
Required incremental units = DSP spend ÷ Contribution per unit
Illustrative example. A brand with:
- Average selling price $34
- Contribution per unit $9.50
- Current monthly revenue $180,000
- Current total ad spend $27,000 → TACOS 15.0%
Adding $12,000/month of DSP:
New total ad spend: $39,000
Break-even incremental revenue at 15% TACOS:
$39,000 ÷ 0.15 = $260,000
Required revenue growth: $80,000 (+44%)
That is the honest bar, and it is high — which is why DSP rarely makes sense as a bolt-on to a small ad budget. Now the contribution version, which is the one that determines whether you actually made money:
Incremental units needed = $12,000 ÷ $9.50 = 1,263 units
At $34 ASP: $42,950 incremental revenue
So: $42,950 of new revenue to break even on the DSP spend itself, $80,000 to hold TACOS flat. The gap between those two numbers is the room you have — spend above break-even contribution is profitable even if TACOS deteriorates, provided the deterioration is temporary and the customers repeat.
Which is the real argument for DSP. It only works on repeat-purchase economics or genuine brand building. If your product is a one-time purchase with no line extension, DSP has to pay for itself inside a single transaction, and at CPM pricing it usually can’t.
When DSP is worth modelling
- Consumables and replenishables. Lifetime value materially exceeds first-order contribution, so a first-order loss is recoverable.
- A catalogue with cross-sell depth. Multiple SKUs a customer might buy after the first.
- An established brand with meaningful remarketing pools. Remarketing to detail-page viewers is DSP’s most defensible use, and it is measurable on a shorter horizon than prospecting.
- Sponsored Ads already maxed out. If you own your keywords and impression share is saturated, incremental growth has to come from somewhere else.
- Ad budget large enough that DSP is a slice, not a gamble. If DSP would be more than roughly a quarter of total ad spend, the test is too big to fail safely.
When it isn’t
- Single-purchase products with no repeat or cross-sell path.
- Thin contribution margin per unit, where the required incremental volume is implausible.
- Sponsored Products campaigns still below break-even ACOS — fix the cheaper, better-attributed channel first.
- Catalogues with a handful of SKUs and no brand story to build.
- Any situation where you cannot commit to a three-month read. A one-month DSP test produces a number you should not act on.
Best practices
- Set the success metric before you start, and make it TACOS plus total revenue, not DSP-attributed ACOS.
- Baseline for at least 90 days first. You cannot detect incremental lift against a baseline you never measured.
- Start with remarketing, not prospecting. Shorter feedback loop, clearer attribution, lower risk.
- Hold Sponsored Ads budgets constant during the test. Changing both at once makes the result uninterpretable.
- Watch your Sponsored Products ACOS and branded search volume as secondary indicators. These are where DSP’s spillover shows up.
- Model on contribution margin, not gross revenue. A campaign at 20% ACOS is fine on a 45% contribution margin and fatal on a 15% one.
- Give it a full quarter, then decide once. Mid-flight panic reallocations destroy the read.
Tools like sellerboard track ad spend against net profit at SKU and account level, which is what makes a TACOS-based read possible — the total-spend-to-total-profit relationship is the one DSP has to move.
Common mistakes
- Judging DSP on its own attributed ACOS. The single biggest error, and it cuts both ways: attribution can also flatter DSP by claiming view-through credit for sales that would have happened anyway.
- Running DSP while also changing prices or Sponsored Ads budgets. Now you have three variables and one number.
- Ignoring the managed-service fee. If you’re going through a managed service, that fee is part of the spend and belongs in the math.
- Testing with a budget too small to reach frequency. Underfunded display reaches everyone once and persuades nobody.
- Assuming impressions are progress. CPM buying guarantees impressions. It guarantees nothing else.
- Forgetting that returns eat incremental sales too. Apply your return rate to DSP-driven revenue before calling it profitable.
FAQ
Do I need a minimum budget for DSP? Amazon has historically required substantial minimum commitments for managed DSP, with a self-serve tier available to some advertisers. Minimums and eligibility change, so confirm current terms with Amazon or your account team rather than relying on figures published elsewhere.
Is DSP the same as Sponsored Display? No. Sponsored Display is a self-serve Sponsored Ads product. DSP is a separate programmatic platform with broader inventory, including off-Amazon and connected TV, and CPM pricing.
How long before DSP shows results? Remarketing can show signal in four to six weeks. Prospecting and awareness campaigns need a quarter or more, which is why the budget commitment matters.
Should I use ACOS at all for DSP? As a diagnostic, yes — it tells you something about efficiency within the channel. As the decision metric, no. Use TACOS against total revenue and contribution.
Can DSP make my Sponsored Products look better than they are? Yes, and this is worth watching. If DSP warms demand that converts on a branded Sponsored Products click, that campaign’s ACOS improves for reasons that have nothing to do with the campaign. Reading the two channels together is the only way to avoid drawing the wrong conclusion about either.
Conclusion
DSP is not a better or worse channel than Sponsored Products. It is a different instrument, priced on impressions rather than clicks, aimed at demand that doesn’t exist yet rather than demand that is already searching.
That makes it a genuine option for brands with repeat purchase economics, catalogue depth and an ad budget large enough to absorb a quarter-long test. It makes it a poor fit for single-purchase products on thin margins, no matter how good the audience targeting sounds.
The deciding question is not “what ACOS did DSP deliver.” It is whether total revenue grew faster than total ad spend, measured on contribution rather than revenue, over a period long enough to be real.
