Asking for documents one at a time is how claims die
Claims get stuck in a documentation loop. The team asks for a receipt, the customer sends it, then the team asks for the item's serial number, then for photographs. Each round trip takes several days, and the customer loses patience.
All the missing information could have been identified on the first reading, because you have to know what that type of claim requires, depending on the product, the warranty term, and the nature of the damage.
Judging whether the evidence is enough is harder. An out-of-focus photograph can still meet the requirement on paper and still not be enough to decide. If you accept that photograph now, you will have another round trip later.
Some claims must never be automated, because any instance involving injury, safety, or a legal claim must be assigned to a person, no matter how complete the documentation is.
How the agent is built in Agent Studio
The documentation requirements specific to each claim type are the main configuration, and they belong in prompt skills so a claims lead can maintain them without interacting with the agent.
The system prompt states what counts as enough evidence: what makes a photograph usable and what makes a receipt acceptable. It also names the claim types that go straight to a person without assessment.
A flow supplies the material. The claim arrives through a Webhook trigger. Document Upload receives the attachments. PDF4me extracts anything received in PDF format. The Agent Chat node then passes the package on.
The agent returns what is currently there, what is absent, and a single combined request covering all remaining items. That one message is the main aim of the design.
A Human in the Loop node holds the request for review before it is sent to the customer. An Email or Resend node then sends it.
The agent does not make the decision. It checks whether the information is complete. Approving or rejecting a claim has financial and legal implications and stays with a person. That limit is set in the flow, not in the prompt.
What this agent is built from
- Prompt skills: documentation requirements per claim type, maintained by the claims lead.
- System prompt: what makes evidence enough, and which claim types bypass assessment entirely.
- Document Upload (Data): takes the attachments. PDF4me (Data) extracts PDF contents.
- Agent Chat (Util): hands the assembled claim over for a completeness assessment.
- Email (Human in the Loop): holds the consolidated request for review before it reaches the customer.
- Human decisions: approvals and declines stay with a person and are enforced in the flow.
- When an ActionFlow is enough: If a checklist of required fields is enough, build the workflow. Judging whether the supplied evidence is actually usable is the agent's job.
Frequently asked questions
Start building AI workflows
Create a free account, open a template or a blank canvas, and run your first ActionFlow.
Newsletter
Get product updates
New nodes, agents, and product notes. We send mail only when we have something worth opening.