Amazon DSP 2.0 Explained: From Retargeting Tool to Full-Funnel Platform

What 'Amazon DSP 2.0' actually refers to

First, the disambiguation: this is Amazon DSP, the Demand-Side Platform — programmatic display, video and audio buying across Amazon and the open web. It is not Delivery Service Partner, the courier franchise. Different Amazon, different letters, same acronym.

Second: there is no software release called DSP 2.0. Amazon didn't ship a version 2.0. The term got popularized by agencies, Acadia among them, to describe a real change in how the platform gets used well — moving from bottom-funnel retargeting (cart abandoners, detail-page viewers) toward full-funnel buys that reach shoppers before they're in-market, measured through Amazon Marketing Cloud instead of last-click alone. The label is marketing shorthand. The underlying shift is real and worth understanding on its own terms, without the version number.

Three things actually changed: the inventory available (streaming TV, audio, more premium placements), the audience-building tools (AMC lets you build segments from purchase behavior, not just Amazon site activity), and the measurement expectation — brands increasingly know last-click DSP reporting overstates or understates impact, and are asking for something closer to proof.

The part every explainer skips: last-click can't tell you what DSP did

Here's the honest problem underneath all of this. Last-click attribution assigns a sale to whatever ad was clicked last. It cannot tell you whether that shopper would have bought anyway. It never could — this isn't a DSP 2.0-era discovery, it's a structural limit of the model. Adding streaming TV or AMC audiences on top of last-click reporting doesn't fix that; it just adds more inventory being measured the wrong way.

The fix isn't a better dashboard. It's a different test design: holdouts and matched controls, where you compare an exposed group to a deliberately unexposed one and measure the actual difference. Amazon Marketing Cloud is where this gets reconciled properly — it's the one place DSP and sponsored ads stop double-counting the same shopper, so you can say honestly whether display added incremental sales or just took credit for ones that were already happening.

A worked example: what full-funnel DSP looks like in real numbers

Definitions are cheap. Here's what the shift looks like on an actual book of business. Across 30 advertisers in July 2026, a full-funnel DSP program delivered 6.04x return on ad spend, 78.4 million impressions at a $4.00 CPM, and a blended $1.42 cost-per-click. Blended cost per acquisition was $5.49 across 57,137 attributed purchases — and 20.1% of those purchases came from shoppers new to the brand.

That 20.1% new-to-brand figure is the number that separates DSP 2.0 from DSP 1.0 in practice. A retargeting-only book, by design, mostly reconverts people who already know the brand. New-to-brand share near a fifth of purchases means real reach beyond the existing customer base — which is the actual argument for spending on DSP instead of just shifting more dollars into search.

Scope matters here: these are results from a specific book of advertisers in a specific month, not a universal guarantee. Anyone quoting a DSP number without saying whose spend, over what window, is asking you to trust a figure you can't check.

The mistake almost everyone makes — including agencies that should know better

The common failure mode: a brand launches upper-funnel DSP (streaming TV, prospecting audiences) and then judges it against the same ROAS bar as bottom-funnel retargeting. Upper-funnel media doesn't convert on the same timeline or at the same rate as a retargeting ad shown to someone who already has the product in their cart. Measured the wrong way, it looks like a failure. Killed on that basis, the brand loses the exact reach it was trying to build.

We've made a version of this mistake too: moving a client into full-funnel spend before the measurement plan was actually in place to judge it fairly. The fix isn't more caution about DSP — it's sequencing. Measurement plan first, spend second, not the other way round.

The other mistake is the opposite one: keeping DSP in ROAS jail permanently — running only retargeting, forever, because it's the safest number to show finance. That protects the metric and starves the growth the platform was actually built to deliver.

What to do when the number is bad news

Sometimes the honest answer is that DSP isn't working for a given brand or category, and no amount of reframing changes that. Here's how to tell the difference between a measurement problem and a real problem:

• If last-click ROAS looks weak but you haven't run a holdout: that's a measurement gap, not proof of failure. Test before you cut.
• If a matched control shows no lift over the unexposed group: that's real. Cut the spend, don't just reallocate it to a different line item and call it fixed.
• If new-to-brand share is near zero after months of prospecting: the targeting is retargeting existing customers under a different name. Fix the audience before touching the budget.

Every change worth making should carry three things before it runs: the evidence behind it, a measurement plan, and a rollback trigger. If a recommendation doesn't have those three, it's a guess wearing a strategy deck.

Where this leaves you

If you're running DSP self-serve and want to understand what to build next, the crawl-toward-full-funnel logic in this piece applies whether you manage it yourself or hand it off. If you want it run as a managed product with holdout-based measurement built in rather than bolted on, that's what Dr. DSP does — Fable 5, from Full Circle, which has managed over $500M in revenue across 100+ brands, with Orbit (the reporting and automation suite) included at no extra cost and no published flat rate; pricing is set on a call against your actual spend and scope, with the first 30 days free.