Decide what Amazon’s ad agent may touch before DVA+ arrives

At unBoxed two weeks ago, Paul Kotas, senior vice president of Amazon Ads, described the new platform in one line I keep rereading: “You can ask questions, gain insights, consider detailed recommendations, or simply give instructions.”

Simply give instructions. That is the part that matters for anyone who buys media on Amazon. The tool you used to click through is now a tool you can talk to, and it can do what you tell it. It can also, in a couple of places, do what it decides is best on your behalf.

Amazon calls the whole thing Amazon Ads Agent. It is the new name for the unified platform that merged the old ads console and the DSP last year, and Amazon says it is “now powered by agentic AI.” I don’t think the interesting question is whether the chat is smart. The interesting question is who on your team has decided what it is allowed to change, and how much money sits behind each campaign type it can reach. My guess is that in most accounts, nobody has written that down yet.

What Amazon actually shipped, and on what clock

It helps to separate the three pieces, because they arrive on different dates and give the machine different amounts of rope.

Sponsored ads stay where they were. What changes is the layer on top. Amazon’s announcement lists “AI-powered guidance for sponsored ads” that finds opportunities in Sponsored Products campaigns, with recommendations for targeting, bids and budgets, and lets you “apply them in one click.”

Full-Funnel Campaigns is live now for all advertisers in the United States, confirmed by Amazon and by Marketing Dive. You give it products, creative and a daily budget (and an optional return on ad spend threshold). Amazon’s AI spreads that budget across sponsored ads, display, video and streaming TV in real time. You approve the creative before launch. Per Digiday, you cannot pick individual channels or specific audiences. Amazon’s own product page says it “runs alongside existing campaigns,” needs about two weeks to learn, and carries a minimum daily budget.

DVA+ is the one with a clock on it. It folds Sponsored Display, Sponsored TV and programmatic display, video and audio into one campaign type, run by Amazon’s Brand+ and Performance+ optimization. Amazon says it begins rolling out in late October. Its UK product note is more precise: extended closed beta in the US today, open beta expected late October, international expansion through Q4, and general availability across 35 markets targeted for Q1 2027. Marketing Dive’s version is late October through the end of the year.

Sitting across all three is the conversational layer. Amazon describes it as now in “expanded closed beta,” and says it helps advertisers “find answers, make decisions, and take action.” The new pieces are media planning with budget allocation recommendations, plain-language targeting that comes back “ready to apply,” natural language analytics with peer benchmarks, and the sponsored ads guidance above. Amazon says these roll out over the coming months, and Kelly MacLean, a vice president at Amazon Ads, told Marketing Dive the beta covers eligible advertisers across 20 marketplaces in 21 languages.

Amazon’s numbers are worth knowing and worth labeling as Amazon’s. It says advertisers who used natural language to apply targeting recommendations last year got more than 25% additional unique customers and paid more than 10% less per thousand impressions on average. It says Full-Funnel beta campaigns saw 67% higher long-term return on ad spend and 29% lower cost per new-to-brand customer. Those come from Amazon’s internal data on its own beta. They may well be true. They are also the vendor grading its own homework, and long-term sales is a 12-month estimate, not cash in the bank.

A chat box that can apply changes needs a written scope

Here is the shift as I see it. When a buyer changed bids last year, there was a person, a login and a change history. When a buyer types “shift budget toward the campaigns with the best new-to-brand rate” into a chat window and clicks apply, there is still a person and a login. But the instruction was vague, the agent decided what it meant, and the action may have touched dozens of campaigns at once.

That is fine if you decided in advance that it should be fine. It is a problem if the first time anyone thinks about it is when a client asks why their branded search budget dropped by a third on a Tuesday.

So the first job is a short written scope for the chat. Not a policy deck. One page that says, for your accounts:

  • What the chat may do with no second person: questions, analysis, reports, audience sizing, forecasts. Reading is cheap. Let it read everything.
  • What it may draft but a human must apply: new targeting, new campaigns, budget reallocation recommendations, media plans.
  • What it may apply directly, and within what limits: for example, bid changes inside a set range on non-branded Sponsored Products campaigns, or budget moves under a dollar amount per day.
  • What it may never touch through chat: branded defense campaigns, anything tied to a retailer co-op or vendor agreement, campaigns with a contractual spend floor, and anything a client has asked to approve personally.

Amazon gives you the shape for this already. The platform offers manual building, AI acting on your behalf, or a mix, and MacLean described the approach to Digiday as “human-led AI supported”. Good. Write down which one each account is, so the mix is a decision and not a habit that formed one click at a time.

For agencies, put the scope in the client agreement. A dentsu executive is quoted in Amazon’s announcement saying the chat lets teams “make human-informed optimizations directly.” That is the right description of the upside. Your client should know which optimizations, and who the human was.

Overhead cyanotype print of a stack of blank slips tied with red string and a red wax seal, next to a closed copper cash box.

Pick which campaign owns the dollar

The second job is about budget, and I think it’s the one most teams will get wrong without noticing.

Full-Funnel Campaigns runs alongside your existing campaigns. Amazon says so plainly. So the day you launch one, the same products can be in a Sponsored Products campaign you manage by hand, a Sponsored Display or programmatic campaign your DSP team runs, and a Full-Funnel Campaign where Amazon’s AI decides the mix. Late October, DVA+ joins that list.

Each of those reports performance its own way. Full-Funnel reports new-to-brand customers and long-term sales alongside clicks and sales. DVA+ promises domain-level reporting by publisher, format and geography. Your manual campaigns report what they always did. If you read all three dashboards and add up the wins, you risk counting some customers twice and giving credit to whichever system happens to tell the best story.

Before you scale anything, I’d make one table. Every product or product group in rows. Columns for sponsored ads, Full-Funnel, and DVA+. In each row, mark which campaign type owns that product’s incremental budget this quarter, and which ones are only allowed to run at a maintenance level. Then decide how you’ll judge the owner, ideally with something outside Amazon’s own reporting, like a holdout, a geo test, or your own blended sales number.

That table answers the question the chat will keep asking you. When Amazon’s media planning recommends a budget split across channels, you’ll know which campaign type you’ve already decided is in charge, and you can tell it no without an argument.

Two practical notes on Full-Funnel. Give it the two-week learning period Amazon describes before you judge it. And since you cannot steer channels or audiences inside it, decide up front what result would make you stop. Digiday listed three things buyers should be able to check: what the agent does with budgets and creative, how it works with third-party data and measurement, and whether fees, supply and auction mechanics stay clear. If you can’t check one of those for your account, write that down as a known gap.

Close-up cyanotype print of a hand turning one key in a console lock while two matching keys hang unused on a copper hook.

Get ready for DVA+ before it shows up

DVA+ is the change with the most operational surface, and the rollout window is close. If you run Sponsored Display, Sponsored TV or programmatic campaigns on Amazon today, these are the things I’d sort out this week, all from Amazon’s own DVA+ documentation.

Who gets advanced settings. Simple DVA+ asks for an objective, products, creative and a budget, then handles targeting, inventory, format mix and bidding. Advanced settings bring back manual targeting, deals, publisher lists, frequency caps, format-level budgets and flighting. Existing programmatic advertisers and agencies with entity access get both. New advertisers and sponsored ads advertisers get simple DVA+ and have to request advanced settings through their account manager. If you’re a brand whose programmatic was run by an agency, check which bucket you’re in before you need the controls.

Which mode each campaign uses. Decide this per campaign, not per team. A test budget on a new product line can run simple. A campaign with deals, brand safety lists or frequency rules from a client contract should run advanced from day one.

Billing. Amazon says payment is invoice-only during closed beta, with card support planned for Q1 2027. If finance expects card billing on a new line, tell them now.

Your scripts and tools. Amazon says API access for display, video and audio campaigns is targeted for Q1 2027, and campaign creation for now happens inside Amazon Ads Agent. If your team or your agency builds and edits DSP campaigns through tooling, plan for a stretch where some of that work happens by hand or by chat. That is exactly when the written scope from earlier earns its keep.

Creative. Both Full-Funnel and DVA+ can generate display and video assets from your product pages. Decide who reviews generated creative and against what brand standard, before the first batch shows up in an approval queue at 5 p.m. on a Friday.

Questions I’d put to Amazon or your account team

The answers will vary by account, so ask directly:

  1. Does every change applied through chat show up in the change history, with the user who approved it and the instruction they typed?
  2. Can an account admin limit what the chat may apply, by campaign type, campaign or dollar amount?
  3. When a Full-Funnel Campaign and a manual Sponsored Products campaign carry the same product, how does Amazon handle the overlap, and how is credit split in reporting?
  4. When will the explanations Amazon says it is working on, the ones that say why the agent recommended a targeting or budget change, reach our account?
  5. What date does DVA+ open beta reach our advertiser account, and what happens to our existing Sponsored Display and Sponsored TV campaigns when it does?

What to do this week

Amazon has done the hard part of making its ad stack simpler. The work left over is ours, and none of it is technical.

Write the one-page scope for what the chat may read, draft, apply and never touch. Build the product-by-campaign-type table and name one owner for each budget. Pick the outside measure you’ll use to judge Full-Funnel after its learning period. Find out which DVA+ access bucket you’re in, decide which campaigns need advanced settings, and warn finance about invoice billing. If you’re an agency, put the chat scope into each client agreement.

Then let the agent do what it’s good at. It can read nearly a thousand metrics faster than any of us. It should still be told which buttons it may press.