What Amazon Ads announced at Unboxed 2026, and what it means for your brand
At Unboxed in San Francisco, Amazon retired the screens most of us have spent the last decade working inside. The Ads Console and the DSP interface are giving way to a single AI assistant. Any AI agent can now reach more than 115 Amazon Ads tools through one connection. Display, video, and audio buying have collapsed into a single flow, and sponsored ads are turning into a conversation.
While many would file that as a feature list, we see things differently. When every advertiser gets the same agent, the agent stops being the advantage. What still delivers an edge is what you give that agent to work with and how carefully you govern it. That edge is where we put our focus during the two days of Unboxed and what this recap is about.
KEY NUMBERS FROM THE FLOOR
|
Big Stats |
|
|
350M |
Customers who used Alexa for Shopping in the last 12 months |
|
40% |
Higher average order size when shoppers buy inside a shopping conversation |
|
115+ |
Amazon Ads tools any AI agent can use through one connection |
|
67% |
Higher long-term ROAS for full-funnel campaigns in early open beta results |
|
29% |
Lower cost to acquire a new-to-brand customer, same beta |
|
3 hrs to 3.5 min |
How long a complex Amazon Marketing Cloud query takes, before and after the new Expert Tool |
All figures as presented by Amazon Ads and its partners at Unboxed, September 2026
WHAT'S NEW FROM AMAZON
Six announcements matter most. Here they are in plain terms, because a few of these names will be on every call with your agency for the next year.
|
Area |
What changed |
|
Amazon Ads Agent |
One AI assistant replaces the Ads Console and the DSP interface. Ask it about any of 1,000+ metrics in plain English and it answers with peer benchmarks. Users can also give the Amazon Ads Agent an instruction and it will act on it. |
|
Amazon Ads MCP server |
MCP is a standard way for AI agents to use a platform's tools. Amazon's server, now in open beta, gives an agent 115+ actions across campaigns, reporting, Amazon Marketing Cloud, and billing through a single connection. More than 700 connections already use it. |
|
AMC Expert Tool |
Amazon Marketing Cloud (AMC) records every step a shopper took, from first ad to purchase. Getting answers out of it has meant writing structured query language (SQL). Now, the AMC Expert Tool writes the query from a business question. |
|
DVA+ |
Display, video and audio buying in one four-step flow, now open to every advertiser. Users can build a campaign by filling in a form or by describing it in a sentence, and nothing goes live without approval. |
|
Full Funnel Campaigns |
Sponsored ads, display, video, and streaming TV in one campaign, with ROAS floors and a new Long Term Sales metric that estimates the future value of new-to-brand shoppers. Open beta. |
|
Alexa for Shopping |
Sponsored Products and Sponsored Brands placements can now open into a multi-turn conversation, written from a brand's own conversation brief. The whole decision, and the purchase, can happen inside it. |
One architectural decision underpins all six announcements. Amazon rebuilt its ads business so everything can be done through code, then handed those same capabilities to agents. Thinking through this like arithmetic:
Amazon brings the signals, the supply and the measurement, the builder brings the data, the orchestration and the customer experience, and the sum is the differentiator.
TAKEAWAY #1: WHEN EVERYONE HAS THE SAME AGENT, CONTEXT WINS
Every brand on Amazon now has access to the same assistant, the same 115 tools and the same query writer. If the agent were the advantage, that advantage will evaporate by the end of the quarter.
Amazon's own presenters kept landing on the same idea in different sessions: an agent is only as good as its context. Context, in practice, is specific; examples are:
- The search terms a brand will never bid on
- The ROAS floor below which brand campaigns get protected instead of cut, the margin on each product
- The fact that lip balm sells hardest in winter
- The record of what worked last time
In a hands-on workshop, Amazon built that context layer as its own component, a store of brand rules and past wins that the agent consults before it plans anything.
The most common failure? An under-specified prompt. A vague instruction produces a surprising result, and a surprising result is almost impossible to audit after the fact. We recommend a folder of tested instructions and constraints; every agent should have instructions to read it.
Agents change where an agency's value sits:
- Operating the controls will get cheaper every quarter. The agent can build the campaign, pull the report, and draft the query.
- Writing the context matters more. Knowing which rules matter for which brand, and why, is the expertise, and agents don't replace that judgement so much as multiply it.
- Context compounds. Every rule written down and every result logged makes the next decision better. A brand that starts building that record now will be hard to catch in two years.
TAKEAWAY #2: THE TIME SAVINGS ARE REAL, AND THEY MOVE THE WORK UPSTREAM
The most concrete numbers of the week were about time. Pacvue showed a basic AMC query dropping from about 30 minutes to three, and a complex one from three hours to about three and a half minutes. Nestle walked through its Garden of Life brands: a workflow that took five or six steps and two to three hours ran in under ten minutes on the first attempt, and the query library its team used to maintain by hand mostly stopped needing maintenance.
An AMC question that used to need a specialist and an afternoon is now a question anyone on the account can ask. Insight has always been rationed by time. When pulling an audience overlap or an on-Amazon conversion path out of AMC costs an afternoon, it happens once a quarter. When it costs four minutes, it can happen whenever a client asks a good question.
The bigger point came from a panel with Omnicom, AWS, and Amazon. Teams that bolted AI onto their existing workflows reported efficiency gains in the 10 to 30 percent range. Teams that rebuilt the process from scratch, assuming an agent would do the mechanical parts, reported 90 percent gains or more(!) Their stated goal was to move most of a team's time away from systems and toward an even deeper consumer understanding, toward telling an even stronger story.
We agree. The hours that used to go into writing queries now go into deeper thinking: which questions matter, and what to do with the answers.
TAKEAWAY #3: FULL FUNNEL ISN'T JUST FOR BIG BRANDS ANYMORE
The best case study of the week came from a company of about 90 people. Wind River Chimes makes hand-tuned wind chimes, a hard category to advertise. Nobody searches "wind chime" when shopping for a wedding or memorial gift, so their search campaigns had hit a ceiling.
With Intentwise, Wind River Chimes ran three ad tiers at once: awareness ads showing chimes in gifting moments, consideration ads aimed at shoppers who had seen those but hadn't searched yet, and search campaigns underneath. AMC tied everything together. Shoppers who had seen the awareness ads got higher Sponsored Products bids when they did search, and detail page visitors who didn't buy were remarketed to. Awareness ads used to be a leap of faith. Here, AMC turned each impression into a signal that raised search bids on the shoppers most likely to buy.
The results:
- 5.7M shoppers reached against a 4M goal
- Search clickthrough up 75.4% year over year
- 76.2% of sales from new-to-brand customers, and more than 78% from the Performance Plus campaigns
- Impressions fell while clickthrough and revenue rose, which is exactly the trade desired
Amazon's early full-funnel beta results point the same way: 67% higher long-term ROAS and 29% lower cost per new-to-brand customer, with one beta brand, K9 Naturals, experiencing 41% of campaign sales from new-to-brand shoppers.
For years, AMC was a black box. Getting anything out of it took SQL, a technical team, and the budget to pay for both, so in practice it belonged to the biggest advertisers. That barrier is gone, and we feel AMC is a highly underused Amazon asset. AMC captures every shopper event with five years of purchase history, and most brands still treat it as a reporting tool. Now, AMC should be table stakes, as a brand of any size can now use AMC audiences to reach new customers and grow sales and reach, without a data science team or a big-brand budget.
TAKEAWAY #4: THE SHELF IS A CONVERSATION
Amazon said 350M customers used Alexa for Shopping in the last 12 months. While that figure seems wildly high, adoption did near double in Q2, interactions are up five times year over year, and orders placed inside a shopping conversation are 40% larger.
The advertising change is the important part. A Sponsored Products or Sponsored Brands placement can now open into a back-and-forth conversation. The brand supplies a conversation brief covering its voice, which products to recommend and which bundles to suggest, and Alexa runs the conversation from the brand's own detail pages and content. The shopper can decide and buy without leaving it.
That turns product content into a media input. The Creativity Keynote put it neatly: marketers now write for two audiences: people and agents. People still need emotion and story. Agents need structure, specifics, and a clear sense of what the brand stands for.
A weak product page used to lose sales one shopper at a time. Now an agent often reads that page first, and if it doesn't find what it needs, it recommends a competitor before the shopper ever hears your brand name.
Amazon Ads: The Stella Way
Our rule for agents is short: agents propose, people approve. In practice that means three things.
- We write the context before the prompt. Brand rules, margin floors, seasonality, and approval lanes, written down and tested, before any agent touches a budget.
- Put a person on every dollar. Amazon's own reference architecture in the hands-on workshop had a human approving every significant change, with every tool call and decision logged with a reason. We hold to the same standard, and we'd be wary of any partner who doesn't.
- Earn autonomy in steps. The MCP session's advice was to treat an agent like a new team member, and it's the right frame. You wouldn't hand a new hire a multimillion-dollar budget on day one. Start with proposals, measure how good they are, and widen access as trust is earned.
We're not saying this in theory. Within a few days of leaving the hands-on workshop, we rebuilt its campaign loop on our own stack for one of the consumer brands we manage. The agent reads performance, scores every campaign against ROAS targets that step up over time, and proposes budget changes for a person to review. Nothing applies without sign-off. It's an early version, and it's propose-only on purpose.
A NOTE ON WHOSE NUMBERS THESE ARE
Every figure in this recap came from Amazon or a partner on Amazon's stage. That doesn't make them wrong, but it does mean a few deserve a careful read before they show up in a plan.
- The full-funnel results are early open-beta numbers. Beta participants choose to be there, and the 67% and 29% figures haven't been through anyone else's measurement.
- Case studies are chosen for a reason. Wind River was presented by its agency, and the query time savings came from Pacvue, which builds the tools that deliver them. Both are real, and both are best cases.
- The 40% larger orders are a correlation. Shoppers who start a conversation may already be building a bigger basket. The behavior shift is the signal. The order size premium isn't yet proof the conversation caused it.
- The 90% efficiency figure was a panel claim, not a study. We believe the direction. We'd hold off on putting the number in a forecast.
- The AMC Expert Tool's status was described two ways. One session called it launched, another said it was expanding into a closed beta. Check access before you plan around it.
WHERE WE PUSH BACK
We think Amazon is mostly right about what these updates will deliver, while pushing back on three things:
- Agents optimize what you measure, including the wrong thing. An AWS panelist gave the example of a learning system offering 20% discounts when the business intended 10 to 15. Watching the end result isn't enough; you have to watch how the agent gets there.
- One platform's agent sees one platform. Amazon Ads Agent knows Amazon. Most of the brands we work with also sell on Walmart, Target, Instacart and a dozen grocery and club retailers, and the right budget decision often depends on what's happening on the others. Amazon's own multiplier slide says as much: the builder brings the cross-channel layer. Skai showed the AMC Expert Tool running inside its own agent environment on the keynote stage, which is the shape we expect to win.
- Q4 is not the time to migrate. DVA+ arrived about a month before holiday peak. Test it on one account. Don't move a peak-season plan onto a new buying flow because the demo was smooth.
THE TLDR FOR YOUR BRAND'S AMAZON PROGRAM
Amazon has made the controls nearly free to operate, which means the controls no longer deliver an advantage. The edge is the context given to the agent and the discipline used to govern it. Both take work, and both compound.
What we recommend for brands in the next ninety days:
- Connect before you build. Get the Amazon Ads MCP connection in place and use it read-only first, so you learn what the agent sees before it acts.
- Write your context down. Brand rules, margin floors, seasonality and who approves what. This is the work that makes everything else safe.
- Test DVA+ on one account before peak. Small, measured, and easy to pull back.
- Pilot Full Funnel against a new-to-brand goal. Judge it on Long Term Sales and new-to-brand cost, not on week-one ROAS.
- Write your conversation brief. If an assistant is going to recommend your product, decide what you want it to say.
Looking to put agents to work on your Amazon program without handing them the keys? Connect with us.
Based on sessions attended at Amazon Unboxed, September 29 to 30, 2026, in San Francisco, including the opening keynote, the Creativity Keynote, Building with Amazon Ads MCP, the AI Ad Stack session, the DVA+ hands-on workshop, a full-funnel spotlight with Wind River Chimes and Intentwise, and the hands-on agentic campaign automation workshop. All statistics are figures Amazon Ads and its partners presented at the event. Beta features were described as in development and may change before general release.
COMMENTS