Key takeaways

  • Amazon now gives away campaign building and reporting through Ads Agent and its MCP Server, so an agency's value sits above that layer.
  • An AI that spends is only as safe as its limits: a margin-based target ACoS, an approval step and a change log.
  • Ask any agency five questions about who sets limits, what runs without approval, the change log, account access and stockouts.
AI does itHuman decides
Anomaly alerts across accounts
First drafts of copy and creative
Target ACoS from real margin
Launch at a loss or protect profit
Bid changes inside set caps
Claims and compliance calls

Illustrative worked example in the post: a margin cap turned a $150 monthly loss into a $240 profit on one campaign.

How agencies for Amazon sellers use AI today

Strip away the marketing and agency AI does four jobs. It watches, it drafts, it summarizes and, increasingly, it acts. The first three have been normal for a while. The fourth is what changed this year.

SupplyKick describes the core reason agencies adopted AI first: a team running hundreds of ASINs across dozens of brands cannot check every budget cap, search term and stock alert by hand, so software does the first pass and senior people take the decisions that need judgment. That framing is right, and it is where most articles stop.

  • Watching: anomaly alerts on spend, conversion rate, rank and stock across every account
  • Drafting: listing copy, Sponsored Brands headlines, image and video concepts
  • Summarizing: weekly reports, search term reviews, Amazon Marketing Cloud queries written from plain English
  • Acting: bid and budget changes, campaign builds, negatives and keyword harvesting pushed live

What changed in 2026: Amazon built the agent layer itself

On February 2, 2026 Amazon Ads opened its MCP Server in beta to partners with active API credentials. It lets AI tools such as Claude, ChatGPT or Gemini call Amazon Ads functions in plain language, including creating, updating or deleting campaigns, pulling reports and reaching billing data. Amazon also shipped a tool that builds a full Sponsored Products campaign from one prompt, leaving only review and approval.

At unBoxed in September, Amazon merged the old DSP and the ads console into one platform called Amazon Ads Agent, with Sponsored ads, Full-Funnel Campaigns and DVA+ in one place. Amazon says the agent puts nearly 1,000 metrics behind a natural language question and that generative creative tools now sit inside the same chat. Amazon also reported beta results for Full-Funnel Campaigns of 67 percent higher long-term return on ad spend; that is Amazon's own measurement, not an independent test.

The practical effect for brands: the work that used to justify an agency retainer, building campaigns and pulling reports, is now close to free inside the console. What an agency adds has to sit above that layer.

The gap: an AI that spends is only as safe as its brake

Most guides on this topic list tools. Few ask the question that decides whether AI makes or loses you money: who set the limits the AI works inside, and do those limits come from your margin or from the platform's defaults?

Seller Metrics, testing the Ads Agent, notes that its campaign drafts often lean toward settings that push spend, such as broad match, dynamic bids up and down and generous starting budgets, and that keyword suggestions can drift away from the product. Forrester's unBoxed review makes the same point at a higher level: Amazon now owns the media, the marketplace and more of the decision layer, so advertisers should decide where its optimization authority starts and ends and keep independent measurement.

None of that makes the tools bad. It means an agency's AI is worth exactly as much as the guardrails around it: a target ACoS built from your real margin, a confirmation step before anything spends, a log of every automated change, and someone who reads that log.

A worked example: the same AI, with and without a margin cap

Illustrative numbers only. Take a $30 product with $9 landed cost and $10.50 of Amazon fees. That leaves $10.50 before advertising, so break-even ACoS is 35 percent. Say a new campaign converts 10 percent of clicks.

Without a cap: the AI drafts a broad campaign and bids land around $1.20 a click. A month of 1,000 clicks costs $1,200 and brings 100 orders, or $3,000 in sales. ACoS is 40 percent. Profit before ads is $1,050, so the campaign loses $150 while the dashboard shows healthy sales growth.

With a margin cap: the agency sets a 25 percent target ACoS, which at a 10 percent conversion rate means a maximum of $0.75 a click (30 x 0.25 x 0.10). Fewer auctions are won, so say 800 clicks cost $600 and bring 80 orders, or $2,400 in sales. Profit before ads is $840, so the campaign makes $240.

Same tool, same product. Sales are 20 percent lower and profit swings by $390 a month on one campaign. Multiply that across 40 campaigns and the cap is worth more than any feature the tool advertises.

Where AI still needs a human in an agency for Amazon sellers

The sources agree more than they disagree here. MerchantSpring's webinar with UK consultant Chris Turton argues AI is strongest at narrow tasks such as summarizing data or mining reviews for complaints, and weakest at conversion work, profit analysis and brand strategy. SupplyKick lists the same human jobs: setting ACoS targets, choosing rank versus profit, and compliance calls.

  • Target setting: only someone with your cost, fee and cash data can say what ACoS a product can afford
  • Launch versus profit: an AI cannot know you are buying rank this quarter on purpose
  • Claims and compliance: AI headlines with superlatives or health claims get ads rejected or worse
  • Stock: an agent that scales a winner two weeks before a stockout burns rank and budget together
  • Account access: Amazon updated its Brand Service Agreement on March 4, 2026 with rules on automated systems and AI agent access, so how a tool connects matters as much as what it does

Five questions to ask any agency about its AI

Ask these on the first call. A good agency answers each one with a screen, not a slide.

  • Who sets the limits? Ask the agency's ads manager to show where target ACoS per product is stored and how it was calculated from margin. Metric that proves it works: TACoS falls or holds while total sales rise.
  • What can the AI change without approval? Ask for the list of automated actions and their caps, for example bid changes over 20 percent or new campaigns. Metric: no budget overrun days in the monthly report.
  • Where is the change log? Ask to see last week's automated bid, budget and negative changes with the reason for each. Metric: every change in the ads console history matches a logged rule.
  • How does it connect to my account? Ask whether tools use Amazon's official APIs and user permissions you can revoke in Seller Central and the ads console. Metric: you can list every user and app with access today.
  • What happens before a stockout? Ask the operations lead how stock alerts reach the ads team. Metric: ad spend on an ASIN drops before its days of cover hit zero, not after.

How PPC Spider uses AI

PPC Spider runs accounts on SpiderOS, its own platform for profit per product, live reports and ads automation that runs 24x7. Every campaign gets a profit-capped ACoS target set from margin, and the account is managed to a TACoS that leaves money after fees. Automation changes bids and budgets inside those limits; people own the targets.

Custom AI agents built on SpiderOS run in copilot mode, which asks first, or autopilot mode inside rules you set. You own your Seller Central and ads accounts and data, and PPC Spider works in them as users you can remove at any time.

Prices are public. Full-service Amazon management is a flat $1,500 a month for up to 25 ASINs and $20k of monthly ad spend, with SpiderOS included. PPC management starts at $995 a month. SpiderOS alone starts at $79 a month if your own team runs the ads. A custom AI agent build is $1,950, with care from $480 a month per agent. Every retainer has a 90-day first term, then runs month to month, and none is a percentage of ad spend.

When this advice does not apply

If your catalog is a handful of ASINs spending a few hundred dollars a month, the guardrails above still matter, but you can set them yourself in the console and review weekly; a retainer will likely cost more than the waste it removes. Amazon's built-in tools plus a written target ACoS are enough to start.

It also matters less during a deliberate launch, when you are buying rank at a loss on purpose. In that case the cap is a budget ceiling and a date, not a margin, and the question becomes whether someone ends the launch spend on time.

Sources

  • Amazon Ads: Introducing the Amazon Ads MCP Server, February 2, 2026 (advertising.amazon.com/library/news/amazon-ads-mcp-server-open-beta)
  • Amazon Ads: The 11 biggest takeaways from unBoxed 2026 (advertising.amazon.com/library/news/unboxed-2026-news-announcements)
  • Forrester: Amazon Ads unBoxed 2026, Amazon makes its pitch for outcome-based advertising (forrester.com/blogs/amazon-ads-unboxed-2026-amazon-makes-its-pitch-for-outcome-based-advertising)
  • SupplyKick: How agencies for Amazon sellers use AI in 2026 (supplykick.com/blog/how-amazon-agencies-use-ai-2026)
  • MerchantSpring: The real value of AI for agencies for Amazon sellers (merchantspring.io/resources/the-real-value-of-ai-for-amazon-agencies-hype-vs.-reality)
  • Seller Metrics: Is it a good idea to use the Ads Agent in Amazon Advertising Console? (sellermetrics.app/amazon-advertising-ads-agent)
Get the next change before your competitors do

We send the marketplace changes that matter each week, with what to do next.

By subscribing you agree to our privacy policy. Unsubscribe any time.
Want this done for your account?

We run these playbooks on client accounts every day. Start with a free audit.

Get a free audit