An account health agent that spots problems before they cost you sales

Brands often find out about an account health problem from the damage: a best seller that stops appearing in search, a stranger's offer on the listing, a red banner in Seller Central on a Monday morning. Amazon publishes most of these changes through its Selling Partner API as notifications and reports, but few teams connect them to anything that tells a person what to do next. Below is the account health agent we build and deploy on a brand's own accounts: what it watches, how quickly each signal reaches it, what it drafts, and which steps stay with a person.

What does an account health agent watch?

It watches the score that tells you how close your account is to deactivation, and the everyday events that move that score or cost sales before they reach it. The score is the Account Health Rating, a number from 0 to 1,000 for the policies it covers. Amazon's AHR program policy sets three bands: Healthy at 200 to 1,000, At Risk at 100 to 199, and Unhealthy at 99 or lower, where the account is eligible for deactivation or already deactivated (checked September 25, 2026). The score reflects the last 180 days, and Amazon can still deactivate an account at any score if it suspects fraudulent, deceptive, illegal or otherwise harmful activity.

The arithmetic shows where the points go. Amazon's AHR FAQ says new sellers start at 200, a violation typically costs two to eight points, repeat violations of the same policy can double in cost, and a critical violation takes the score straight to zero. Points come back when a violation is successfully appealed or disputed, as old violations age out of the 180 days, and at four points for every 200 orders fulfilled in that window (checked September 25, 2026). A brand that ships 1,000 orders in 180 days earns 20 points from volume, so sales alone cannot outrun a string of violations; addressing each one is what wins the points back.

Around that score, the agent reads three groups of signals. At the account level: the rating, the policy categories behind it, such as received IP complaints and product authenticity or condition complaints, plus order metrics and negative feedback. At the listing level: suppressions and other listing issues, new offers on your ASINs, and who holds the Featured Offer, still widely called the Buy Box. At the inventory level: stranded inventory, FBA units sitting in a fulfillment center with no active offer, so nobody can buy them.

Most of the account-level data comes from one place. Amazon's performance reports documentation describes a seller performance report that carries the Account Health dashboard's metrics, including the rating, its status and a count for each policy category, and a feedback report that lists negative and neutral ratings from one to three stars. To size the sales at stake, the agent adds the Sales and Traffic report, which Amazon's analytics reports documentation says carries page traffic and Buy Box percentage (both checked September 25, 2026). Each group of signals arrives in a different way and at a different speed, as the alert map below shows.

Which alerts matter most for a US brand?

The ones that can stop sales fastest come first, then the ones that add up. A critical violation can end selling within days. A best seller hidden from search, or made unbuyable, loses sales from that hour. A lost Featured Offer means Add to Cart no longer selects your offer, and your Sponsored Products ads stop showing. Stranded units and slipping order metrics cost less per day but compound quietly. The table below is the alert-to-action map we start from with each brand: the signal, where Amazon reports it, how fast the agent sees it, what it drafts, and the step a person owns. Your team moves the lines.

The order metrics come with hard lines. Amazon's Order Performance program policy asks sellers to keep the Order Defect Rate under 1% over 60 days, where a defect is negative feedback, an A-to-z claim or a chargeback; and, for seller-fulfilled orders, the pre-fulfillment cancellation rate under 2.5% over 7 days, the late shipment rate under 4%, the valid tracking rate above 95% and the on-time delivery rate at 90% or higher (checked September 25, 2026).

FBA brands are not exempt: the seller performance report splits the Order Defect Rate into FBA and seller-fulfilled orders, each with its own count of claims, chargebacks and negative feedback. Brands that ship some orders themselves have a newer rule too. Since June 29, 2026, the same policy page requires accurate handling times on seller-fulfilled SKUs: a SKU that consistently ships at least a day faster than its setting is flagged for correction, and Amazon recommends Automated Handling Time instead. The agent alerts at a warning line your team sets below each target, not at the target itself, and flags SKUs whose actual handling time drifts from the setting.

Two rules deserve their own alert. A critical violation turns the rating red at zero and starts a three-day grace period to address it before deactivation, according to Amazon's AHR FAQ (checked September 25, 2026). And reaching the maximum number of repeat violations for one policy in 180 days, five for infringement-related policies and two for restricted products, surfaces a critical violation on its own. Amazon labels repeat violations in the Impact column of the Account Health page. The agent also keeps its own count per policy against that maximum, including removed violations that may still count, so nobody has to rebuild it from old emails.

How do you monitor account health without checking every day?

Let Amazon tell you. Most of these signals exist as Selling Partner API notifications that Amazon sends when something changes, so the agent never has to scrape pages. Amazon's notification reference lists one for account status changes between Normal, At Risk and Deactivated; one for listing issues being created, fixed or updated; one for changes to the top 20 offers on an item or to the Buy Box winner and price; and one for when your offer loses Featured Offer eligibility over an uncompetitive price (checked September 25, 2026).

Each listing issue arrives with its severity and the enforcement actions it caused, which Amazon's listings issues guide names as search suppressed, listing suppressed, attribute suppressed and catalog item removed (checked September 25, 2026). That is what lets the agent tell a warning on a slow seller from a best seller that has just vanished from search, and rank the alert accordingly.

The notifications have gaps, and the agent covers each one on a schedule. The offer notification only covers items where you have an active offer, so brand ASINs without one are checked with Amazon's item offers call, which returns the lowest-priced offers for an ASIN at a default rate of one request every two seconds (checked September 25, 2026). That is at most about 1,800 ASINs an hour, so a large catalog is checked in rotation, hero ASINs first.

Reports fill the other gaps. Amazon says not every listing problem has a defined issue, so the agent also reads the suppressed listings report, which Amazon's inventory reports documentation describes as listing each suppression, its reason and how to remove it. The seller performance report and the stranded inventory report can only be requested, not set up as Amazon report schedules, so the agent requests them on its own timer (all checked September 25, 2026).

That setup means the agent monitors continuously, while people, yours and ours, work business days. So the rules have to say who gets a critical alert on a Saturday. We suggest naming a weekend owner for three alerts only: a critical violation, a suppressed top seller and a hijacker on a hero ASIN. Everything else waits for Monday in a ranked queue, with the evidence already attached, so the first hour of the week goes to fixing rather than finding.

If you want to qualify for Account Health Assurance, set the rating floor at 250. Amazon offers it to professional sellers who hold a rating of 250 or higher for at least six months, with no more than 10 days below 250, and a valid emergency contact number on file. Staying in means keeping that number verified and engaging within 72 hours when Amazon asks. Amazon then says it will not deactivate the account as long as you work with it within 72 hours of being contacted, unless it suspects fraudulent or otherwise harmful activity (checked September 25, 2026). An alert at 260 gives your team time to act before the rating crosses the line.

What does it do when it finds a problem?

It turns a notice into a task with an owner, a deadline and a draft. Every alert goes through the same six steps:

1. Classify. A language model reads the notification or email and records the policy, the severity, the ASINs and any deadline Amazon gave.
2. Rank. Rules order the issue by severity, deadline and the sales at stake on the ASIN.
3. Route. The alert goes to the person who owns that fix, in Slack, email or text, and escalates if nobody acknowledges it in time.
4. Assemble. The agent pulls the evidence: invoices and supplier documents, the listing's change history, the orders involved and earlier notices on the same policy.
5. Draft. It drafts the listing fix, the plan of action or the IP response, and says what it could not confirm.
6. Recheck. After a fix, it watches for Amazon to clear the issue and closes the alert only then. Amazon's FAQ says the rating updates within a few hours of Amazon confirming a violation has been addressed.

Take an illustrative case. A brand has had four received-IP-complaint violations under the same policy in 180 days. Two were removed after successful appeals in which the brand did not dispute that the activity happened, so the Account Health page shows two open violations. Amazon's FAQ says violations removed that way may still count toward the repeat-violation maximum, so the agent counts all four against the maximum of five and warns that the next one could surface a critical violation. It alerts the brand owner and the listing lead with the four ASINs, the rights owner behind each complaint and the listings that share the same claims, so the team acts before any three-day clock starts.

Some problems are not policy problems at all. When a best seller runs out at FBA, its offer cannot be featured, and Amazon Ads says in its advertising FAQ that an ad for a listing not eligible for the Featured Offer will not display (checked September 25, 2026). The agent tags that as a stock cause rather than a pricing one, sends it to whoever owns replenishment, and flags the ASIN to your advertising agent or ad manager. Our stockouts page covers what that gap costs, and the inventory agent works to keep it from opening.

Why do appeals and policy responses stay with a person?

Because each one is a statement your business makes to Amazon, and the choices inside it have consequences a model should not weigh alone. We design every account health agent so that no appeal, dispute, plan of action or IP response reaches Amazon without a person reading, editing and submitting it. The agent does the slow parts, finding documents and writing a first draft; your team owns the decision. Four details show why that line sits where it does:

1. Dispute or appeal is a strategic choice. A dispute gives evidence that Amazon detected the violation in error; an appeal accepts that it happened and asks Amazon to forgive it as unintentional. Violations removed after a successful appeal, where you did not dispute that the activity occurred, may still count toward the repeat-violation maximum. The wrong path can leave the account closer to a critical violation.
2. Everything in it must be true. An appeal cites invoices, suppliers and the changes you made. A draft can be wrong about any of them, and only your team knows for sure.
3. Amazon may call. During a critical violation's three days, an Account Health specialist may phone you, and Account Health Assurance asks you to engage within 72 hours, possibly by phone. Amazon's Agent Policy requires an agent to answer truthfully when asked whether an interaction comes from a person or a computer (checked September 25, 2026), and the call needs someone who can commit your business, so it is yours to take.
4. Reports against other sellers carry risk. Amazon lists attempting to damage or abuse another seller among its critical violations. A report through the Report a Violation tool in Amazon Brand Registry, or a removal through Project Zero, is a person's decision, ideally after a test buy confirms the problem. Project Zero asks for at least a 90% acceptance rate on recent trademark reports to join and a 99% accuracy rate to keep access (both checked September 25, 2026).

Our guide to Amazon's Agent Policy covers what Amazon requires of any automated tool on a seller account. The account health agent is designed to follow it: it works through the APIs you authorize, acts only within the rules you approve and leaves everything outside those rules to a person on your team.

How does it fit with your agency or in-house team?

It sits beside the people who already fix these problems and sends each of them only their part. Listing suppressions go to whoever owns the catalog, stranded units to operations, Featured Offer losses to pricing, and policy violations to the brand owner, with your agency's account manager copied where you choose. Every alert names one owner, a deadline and what done looks like. The agent does not replace the relationship with an agency; it makes sure the agency hears about a suppression when it happens, not from a client call days later.

Access follows the same rules as any provider. Your agency gets its own service-provider authorization with only the roles it needs, and the Brand Registry rights-owner role stays with your company; our guide to giving an agency access to Seller Central safely walks through both. The agent runs on your seller account and your systems, so its rules, alert history and evidence files stay with the brand if you change agencies. You keep your accounts, your data and the custom code we build; the Growth OS base is licensed to you. If you stop working with us, your accounts, data and custom code stay with you, and the Growth OS base continues under its license terms.

An agency that runs many client accounts can deploy the same agent on each client's accounts, with alerts routed to its own team. Our page for Amazon agencies covers that setup and the other internal tools agencies run clients on, and account health monitoring sits alongside case management and operations reporting in the operations support on our services page.

When are Amazon's own tools or an alert app the better choice?

Not every brand needs a custom agent, and Amazon's own tools keep improving. On September 23, 2026 Amazon introduced Seller Assistant workflows that run continuously, with prebuilt templates that monitor account health; sellers choose whether a workflow only recommends or takes actions they review and approve first. The announcement gives no store-by-store rollout for workflows, so check Seller Central for yours (checked September 25, 2026). The Account Health page itself, with Account Health Support a Call me now button away for professional sellers, is the right place to start. Alert apps are another route; we checked each one's own page on September 25, 2026.

We are one of the options in this table, so weigh our view accordingly. If Amazon's tools or an alert app cover your catalog and your team has time to act on every alert, use them. A custom agent earns its cost when the hard part is not the alert but the work after it: finding the documents, choosing between a dispute and an appeal, and getting the right person to act before a deadline.

What does it take to deploy it on your accounts?

Five steps, most of them with the person who handles account health today:

1. Diagnose. Start with the free 24-hour diagnostic: connect Amazon with Login with Amazon, no password shared, and get a report within 24 hours of connecting, on business days. It covers stranded or suppressed listings among other gaps. We only read data; we never change listings, prices, stock or ads.
2. Connect. You authorize the agent through the SP-API and subscribe it to the notifications above. Brand Registry stays with your rights owner.
3. Map owners. We record who owns each kind of alert, who covers weekends, your authorized resellers, and where your invoices and supplier documents live.
4. Set the rules. Your team sets the rating floor, the warning lines below Amazon's targets and the drafts it wants for each alert type.
5. Scope and date. The agent is scoped in an architecture doc with a fixed price and go-live date.

We have built the alerting half of this before. AccountDr is rule-based alerting software we built with the AccountDr team: push and email notifications for hijackers, stranded inventory, a suppressed Buy Box, listing faults, ASIN changes, parent-child splits, negative seller comments and performance metrics, built on Amazon MWS with Keepa and camelcamelcamel data. It is not an agent; every alert follows a fixed rule. The account health agent keeps rules like those and adds the model steps: reading Amazon's notices, drafting responses and spotting drops a fixed threshold misses. It starts from Ecomsellertool Growth OS and is one of the AI agents we build on a brand's own accounts.

Ecomsellertool builds and deploys this agent for Amazon brands that have had a suppression, a hijacker or a health warning and want earlier notice next time. We connect your seller account through the SP-API, set the rules with your team and add the model steps that read notices and draft responses, so a person decides with the evidence in hand. We have built on Amazon's seller APIs since 2017.

To see what it would flag in your account, start with the free 24-hour diagnostic or schedule a call from the card below.