Amazon Ads Agent: What You Give Up to Get It

On September 29, at its unBoxed conference in San Francisco, Amazon announced that the unified platform merging its demand-side platform and its ads console is now called Amazon Ads Agent. The rename is the least interesting part. What shipped alongside it changes who decides where your budget goes.

Two new campaign types launched, consolidating what used to be five or six separate buying workflows. Both are real products with dates attached. Both trade specific controls for broader reach and less setup. If you spend on Amazon, this is the biggest structural change to how you buy there in several years, and the migration pressure will arrive whether or not you opt in early.

Here is what actually shipped, what you can no longer choose, and — the part that took some reading to arrive at — why "keep the controls" is the wrong response.

What Shipped on September 29

Amazon Ads Agent now organizes everything into three campaign types: Sponsored Ads, DVA+, and Full-Funnel Campaigns. The second and third are new.

Full-Funnel Campaigns — available now in the US

Previewed at last year's unBoxed and now open to all US advertisers, this is the most automated product Amazon has shipped for media buying. You supply products, creative, and a budget. Amazon's models handle planning, channel allocation, audience selection, bidding, and continuous optimization across sponsored ads, display, video, and streaming TV — including third-party open-internet inventory.

It optimizes toward two outcome metrics: new-to-brand customers and Amazon's Long-Term Sales measure. Amazon's own case study cites a dog treat brand, Canine Naturals, running a full-funnel campaign over 206 days, with 41% of sales coming from new-to-brand shoppers. Treat that as a vendor case study, not a benchmark — it is one advertiser, selected by Amazon, with no control group disclosed.

DVA+ — rolling out from late October

DVA+ collapses Sponsored Display, Sponsored TV, and programmatic display, video and audio buying into one workflow. It runs on Amazon's Brand+ and Performance+ models, which handle audience identification and outcome optimization respectively.

Critically, DVA+ ships in two modes. Simplified mode takes an objective, a budget and creative. Advanced settings keep conventional programmatic controls: targeting, deal IDs, supply sources, frequency caps, and bidding. Kelly MacLean, a vice president at Amazon Ads, described the approach to Digiday as "human-led AI supported," and the stated intent is that larger buyers are not forced into the automated path. Rollout runs from late October through the end of the year.

The conversational layer

The chat interface expanded from campaign creation into media planning, AI targeting, campaign guidance, natural-language analytics, sponsored ads guidance, and insight generation. It is in beta across 20 marketplaces and 21 languages. Amazon Marketing Cloud queries no longer require SQL. Amazon reports that advertisers applying targeting recommendations through natural language saw more than 25% additional unique customers at more than 10% lower cost per impression — again, Amazon's figure, Amazon's methodology.

The Three Controls That Disappear

Full-Funnel Campaigns is where the real trade sits. Per Amazon's own product leads, advertisers using it:

• Cannot select individual channels. You do not decide that this budget runs on streaming TV and that budget runs on sponsored products. Amazon's model allocates.
• Cannot select specific audiences. Amit Bhattacharyya, Amazon's VP of full-funnel products, AI and modeling, told Digiday the team is "figuring out how to give them more transparency on the audience side." That is an honest way of saying the transparency is not there yet.
• Get format-level visibility only. Reporting tells you where ads were delivered by format. It does not reconstruct the allocation decision.

You keep creative approval, products, budget, pacing, flight dates, and location targeting. What you give up is the media plan.

The Case for Handing It Over Anyway

The reflexive take — don't delegate budget allocation to the platform selling you the media — sounds prudent and is mostly wrong. Three reasons.

First, this is not new. Performance Max, Advantage+, and AI Max already took channel and audience selection away from buyers, and in aggregate they outperform the manual campaigns they replaced for most advertisers. Refusing on principle has had a measurable cost for three years running.

Second, the controls were mostly not adding value. If you are a mid-market advertiser whose channel split was set once in a planning deck and revisited quarterly, Amazon's model revisiting it hourly is an upgrade. The controls are only worth defending if you were actually using them with information the platform lacks — which is true for some advertisers with offline sales data or brand constraints, and untrue for most.

Third, operational simplicity is a real return. Dentsu, a beta tester, described teams spending hours on manual setup and optimization now making fewer, higher-level decisions. That time is not nothing; it is the margin in an agency P&L.

So the honest position is: Full-Funnel Campaigns will probably beat what most advertisers were doing by hand. Use it.

Control Was Never the Protection

Here is the part that reframes the whole question, and it has nothing to do with AI.

Five weeks before unBoxed, in August 2026, the FTC and the attorneys general of 22 states sued Amazon over its advertising auctions. The complaint alleges that for over seven years Amazon told advertisers it ran generalized second-price auctions — winner pays one cent above the next bid — while in practice charging Sponsored Products advertisers their own full winning bid close to 80% of the time. It alleges an undisclosed surcharge introduced in 2019, described in internal documents as a "soft reserve price," and quotes an Amazon executive saying the price advertisers pay "isn't set by an actual bidder" but is a "proxy 2nd price that we calculate." The FTC alleges the scheme likely extracted tens of billions of dollars from more than a million brands and sellers, including over 500,000 small and medium businesses. Amazon disputes the allegations, and the case is unproven.

Now notice the mechanics. Every one of those advertisers was in full manual control. They set their own keyword bids. They chose their own match types. They had every lever the automation sceptics want to keep — and the alleged harm, if the complaint is right, ran for seven years regardless, because the thing that mattered was not visible from inside the console.

Control over inputs is not protection. It never was. What would have surfaced a hidden reserve price is the ability to reconcile what you were charged against an independent record of what should have happened — auditability, which is a different capability entirely and one almost nobody built.

When Digiday asked whether Amazon Ads Agent would show Sponsored Ads buyers how individual winning prices are determined, Amazon did not commit to that level of auction transparency. Which means the capability gap that matters is still open, in the same place it was before, under a new product name.

What To Actually Do

The conclusion is not to refuse the automation. It is to stop treating platform controls as your safety mechanism and start treating your own records as one.

1. Take the automation where it is strong. Full-Funnel for prospecting and new-to-brand growth, DVA+ simplified mode where you had no programmatic team anyway. These are genuinely better than manual mediocrity.
2. Use DVA+ advanced settings where your constraints are real. Brand safety requirements, specific deal IDs, frequency discipline, supply-path preferences. The controls exist in DVA+; keep them where they do work.
3. Export daily, to storage you own. Spend, impressions, clicks, conversions, and placement-level delivery, pulled through the API into your own warehouse. Platform UIs are not an archive, and a reporting view can be revised. Pull every platform into one schema so comparisons are possible.
4. Keep a record of what changed and when. If the model reallocates and performance moves, you need the change log alongside the outcome. Automated systems generate these events; most teams never capture them.
5. Measure incrementally, outside the platform. Full-Funnel optimizes to Amazon's Long-Term Sales metric. If that metric is also your scoreboard, you cannot evaluate the product. Geo holdouts and independent incrementality tests are the check.
6. Compare like for like before migrating everything. Hold a portion of spend in the structure you can audit for one full cycle. Not because automation is suspect, but because "it performed better" is only a claim until you have a comparison you built yourself.

The Takeaway

Amazon Ads Agent is a good operational deal. Fewer workflows, broader inventory access for smaller advertisers, and models that will likely allocate better than a quarterly planning spreadsheet. The channel and audience controls Full-Funnel takes away are, for most advertisers, controls they were not using well.

The thing to carry out of this week is narrower and more durable. Two Amazon advertising stories landed five weeks apart: one where the platform asked for more decision-making authority, and one where regulators allege the platform's account of its own pricing was wrong for seven years while every advertiser held the manual controls. Those two facts point the same direction. The defensible position is not less automation — it is independent records.

Delegate the media plan. Keep the data. Those are not in tension, and the teams that will be able to answer hard questions in 2027 are the ones treating them as separate decisions now.

Sources: Marketing Dive — How Amazon is streamlining ad buys across multiple formats with AI (September 29, 2026), Digiday — At its flagship unBoxed conference, Amazon Ads collapses buying silos (September 29, 2026), Amazon Ads — unBoxed 2026 announcements, FTC — FTC, States Sue Amazon Over Secret Ad Surcharge Scheme (August 2026)