Use cases and implementation
Marketplace spam and scam reports
A worked process for account evidence, review, application action, notice, and appeal.
- For
- Marketplace trust and safety, fraud, support, and operations teams
- Last updated
- 29 August 2026
01
Starting point
A buyer reports account 123 for sending repetitive offers and suspicious payment links. The report itself is only one signal: the reviewer also needs account age, prior report volume, listing history, and any previous restrictions before deciding what to do.
02
What must hold
- Treating one unverified allegation as proof of abuse.
- Letting a webhook timeout silently remove an accepted report from the queue.
- Allowing automation to suspend a high-impact seller when evidence is incomplete or contested.
03
Step by step
- Publish the visual workflow with an authority mode, a passing fixture, an immutable version, and explicit failure paths. Pin that version to the active queue policy before new reports enter it.
- Accept the report from an embedded form, public page, or authenticated server API and immediately issue a tracking reference.
- Use the intake-routing agent to classify the allegation and place it in the marketplace spam queue, recording why that queue was selected.
- Read the queue policy and request only the enrichments it requires: account age, prior reports, and recent listing activity.
- Let AI summarize low-risk evidence when the queue allows it, but send uncertainty, failed enrichments, repeat offenders, protected users, and material account actions to a human.
- Record the decision, call the customer action endpoint idempotently, notify the relevant people, and keep any appeal human-decided.
04
Operating boundary
AI may recommend no action, warning, or human review for low-priority spam reports. It may not suspend an account, resolve a contested report, or proceed when account age is missing. Accounts under seven days old with repeated link reports must be claimed by an analyst.
05
Data exchange
The policy requests an account_profile finding. Safest Resolve sends the stable account reference in a signed request. The customer webhook returns the approved age and risk fields. The report keeps the response version and shows timeout or schema failures as review states.
06
Verification
The report records the allegation, relevant context, policy version, evidence source, current owner, reasoning, and allowed actions. The reporter receives an appropriately scoped status. The affected user receives a reason and appeal path when policy requires one.
07
Implementation questions
Can AI close obvious spam reports?+
A published low-priority process may allow that outcome. Uncertain, failed, protected, contested, and high-impact cases move to a person.
What if account data changes after the decision?+
The audit record retains the enrichment result and policy version used at decision time, so the team can reconstruct what the reviewer knew without duplicating the customer’s full account record.
Safest Resolve
Start with one report your team handles today.
Bring its source, required evidence, decision owner, application action, messages, and appeal path.