Smart Intake Agent

The Smart Intake agent can be used to assess and classify automatically emails (an their attachment) from Brokers that submit Trade Credit requests, often mixing new submissions, renewals, and claims.

Comment

Inspiration

In the Trade-Credit world, Brokers or Insureds submit trade credit requests via email or semi-structured forms—often mixing new submissions, renewals, and claims. These messages include free-text descriptions, buyer lists, requested limits, and attachments in varied formats. Underwriting teams lose time manually parsing, reclassifying, and rekeying data, which leads to delays, missed deals, and potential errors.

What it does

The Smart Intake Agent reads incoming broker/insureds emails and attachments, and performs:

  • Classification: Is it a new submission, a renewal, a limit increase, or a claim?
  • Entity extraction: Captures key buyer names, requested limits, insured amounts, sector, geography
  • Document tagging: Flags whether the required files are attached (e.g., financials, group structure)
  • Routing logic: Forwards to the correct underwriting team queue based on region or deal type

It builds a structured JSON that can be used to:

  • pre-populates the Trade Credit Platform submission interface
  • Integrate with Trade Credit Platform to match existing accounts or buyers

How we built it

This project was born following a study on how to improve and better automate some business processes presents in our current Credit-Insurance SaaS platform. The solution is built on top of Python code, AWS Lambda (Containerized with SAM) connected to Amazon Bedrock (Llama LLM) The following is configurable with external JSON files:

  • System and User Prompts
  • Routing rules
  • Output JSON Schema
  • Inbound/Outbound and Config files S3 folders

Challenges we ran into

A coherence check needs to be done between the content of the email and its attachments to ensure that the provided information is coherent. The request needs to be routed to the specific Underwriting team. (which means an external routing config) The output needs to be a structured JSON file and configurable with external schema.

Accomplishments that we're proud of

More than 80% of typical complex emails from Brokers or Insureds for Commercial, Risk or Claims requests were assessed, classified, parsed and routed successfully.

What we learned

A lot! How rich and complex brokers/insured email requests can be. How much precious time can be saved by our client's underwriters

What's next for Smart Intake Agent

  • Integrate it with Amazon SES for end-to-end email integration
  • Integrate with Trade Credit Platform. This can be done easily by reading the output JSon file on S3.

Built With

Share this project:

Updates