Amazon Just Folded Its DSP Into an Agent, and Your Channel ROAS Goes With It
At unBoxed in San Francisco on September 29, 2026, Amazon merged its ads console and DSP into one platform called Amazon Ads Agent and made Full-Funnel Campaigns available to every US advertiser. You hand over products, creative and budget; Amazon's AI splits spend across sponsored ads, display, video and streaming TV. Channel and audience selection controls are gone, so pull 90 days of per-format baselines before you opt in.
That last sentence is the part I'd underline. The product news is genuinely big, and Digiday's writeup from the floor frames it as Amazon collapsing the buying silos it spent a decade building. But the practical consequence for anyone running Amazon spend is narrower and more annoying: the report you use to defend your Sponsored Products budget in the quarterly review is about to get a lot less granular, at least inside these new campaign types.
Everything Amazon folded together on Tuesday
There's a lot of renaming happening at once, so here is the short version.
The unified buying platform (DSP plus ads console) is now Amazon Ads Agent. The product formerly known as Amazon DSP is now called DVA, for display, video and audio, according to AdExchanger. On top of that sits DVA+, which consolidates Sponsored Display, Sponsored TV and programmatic display, video and audio into one buy, with Performance Plus and Brand Plus as the two campaign types. DVA+ has a simplified mode (pick objective, budget, creative) and an "advanced settings" toggle that keeps targeting, deals, supply sources, frequency and bidding in human hands.
Full-Funnel Campaigns is the more aggressive product. It was previewed at last year's unBoxed and is now live for all US advertisers. It puts sponsored ads, display, video and streaming TV (plus some third-party open-internet inventory, per Digiday) inside a single campaign and optimizes toward new-to-brand customers and Amazon's long-term sales metric. You don't choose channels. You don't choose audiences. You get format-level visibility after the fact.
Kelly MacLean, a VP at Amazon Ads, described the approach as "human-led, AI supported." Which is a nice phrase. On paper, it even describes DVA+ pretty accurately. It describes Full-Funnel Campaigns less well, since the human part there is mostly uploading assets and picking a number.
The case studies are good. Read the metrics they picked.
Amazon brought receipts, and I don't want to wave them away. Marketing Dive reports that pet brand Canine Naturals ran Full-Funnel Campaigns for 206 days, got 41% of sales from new-to-brand customers, and saw higher ROAS than its separately built campaigns. Its digital marketing manager, Brenda Quintana, put it this way: "I don't have to do anything, just give it all the assets and it'll figure it out. That's music to my ears."
For a small team, that quote is completely rational. For anyone who reports to a CFO, it should make you a little itchy.
The Cuisinart DVA+ pilot in Q2 2026 posted an 18-point lift in detail page view rate, a 14% lower cost per detail page view, and a 10-point lift in branded search rate. Those are all real, useful numbers. They're also all mid-funnel metrics. Detail page views and branded search are signals Amazon is very good at moving, and they don't tell you much about incremental revenue on their own (and incrementality is the thing platform-reported numbers tend to be weakest at).
None of that makes the results fake. It just means the case studies are measured in the currency the platform prints.
Why losing the channel split hurts more than it sounds
Think of it like switching from ordering à la carte to a tasting menu. The food might genuinely be better. But when the bill arrives it's one number, and you can't tell whether you paid for the wagyu or the bread.
Today, most Amazon teams defend budget channel by channel. Sponsored Products carries the ROAS story, Sponsored Brands carries the new-to-brand story, DSP and Sponsored TV carry the upper-funnel story that finance tolerates because the other two look good. When an agent decides the split for you, that structure goes away. You get a blended result and a format breakdown the agent chose to produce.
And the audience side is still unresolved. Amit Bhattacharyya, Amazon's VP of full-funnel products, told Digiday: "We are figuring out how to give them more transparency on the audience side." I appreciate the honesty, honestly. But "we are figuring it out" is not the thing you want to hear about a product that is already generally available.
Amazon is asking you to trust its allocation math one month after regulators accused it of rigging its auction math.
That's the context nobody on stage dwelled on. In August, the FTC and 22 state attorneys general sued Amazon alleging secret surcharges in the Sponsored Products auction, which we covered in detail here. Amazon disputes the allegations. Per Digiday, the company also declined to commit to specific new auction transparency at unBoxed. Maybe that's just legal caution during active litigation, which would be normal. It still means the black box got bigger in the same season its walls got questioned.
Snapshot first, then give it a controlled slice
I don't think the right answer is to avoid Full-Funnel Campaigns. For brands without a dedicated DSP buyer, it's probably the easiest route to streaming TV inventory Amazon has ever offered, and the mass-market push is clearly where Amazon wants the growth to come from. Amazon Ads did nearly $20 billion in Q2 2026, according to Marketing Dive. From what I can tell, the company is not building this for the brands that already have DSP teams.
What I'd do this week, in roughly this order:
- Export 90 days of per-format baselines now. ROAS, new-to-brand percentage of sales, cost per detail page view, and total ASIN sales (ads plus organic) for Sponsored Products, Sponsored Brands, Sponsored Display and any DSP line. Save it outside the console. Once campaigns migrate, reconstructing this gets much harder.
- Carve out 15-20% of budget for a Full-Funnel test, on a set of ASINs you're not also pushing through your existing campaigns. Overlapping ASINs will cannibalize each other and you'll learn nothing.
- Run it for at least six to eight weeks. Canine Naturals ran 206 days. Agentic optimizers seem to need a learning period, and a two-week test mostly measures the learning period.
- Judge it on total ASIN sales and new-to-brand share, not ROAS alone. My rough rule: if blended ROAS falls more than 15% below your baseline and new-to-brand share doesn't rise by at least 10 points, the agent is buying you awareness you didn't ask for. Pull it back.
- Stay in advanced settings on DVA+ if you have a real buyer. DVA+ rolls out from late October through the end of 2026. The simplified mode is fine for small accounts. If someone on your team knows what a supply path is, keep the controls.
Agencies should also treat format-level reporting as a negotiation point rather than a given. If a client is asked to move budget into a campaign type with fewer controls, the least they should get back is a format breakdown granular enough to rebuild the old channel view. And if you're already in a fee conversation with your DSP (plenty of buyers are, after The Trade Desk's rough year), this is a good moment to ask Amazon's reps exactly how the agent's allocation is reported.
Where this probably ends up by next unBoxed
My guess: by unBoxed 2027, more than half of new US advertisers will launch their first Amazon campaign through Full-Funnel Campaigns rather than a standalone Sponsored Products campaign. Amazon controls the onboarding flow, and defaults win. I could be wrong on the timing. I don't think I'm wrong on the direction.
And to be fair to Amazon, this isn't a new trick. Google did the same thing with Performance Max, Meta did it with Advantage+, and in both cases the controls came back slowly, one reporting column at a time, after advertisers complained loudly enough. Amazon will probably follow the same arc. The difference is that Amazon also owns the shelf, so the agent deciding your media mix is sitting right next to the algorithm deciding your organic rank.
I'm not sure that's a problem yet. But I'd want my own numbers saved somewhere before I found out.
By Notice Me Senpai Editorial
