AI System for Filing Insurance Documents Received on WhatsApp

The problem

Customers send damage photos, accident reports, vehicle registration documents, driving licences, old policies, repair invoices and loss adjuster reports through WhatsApp. The files remain on phones and in chat histories.

Staff download, rename and move each attachment by hand. Different naming habits leave the same customer’s documents scattered across folders.

When a document is needed, its customer, policy or claim may be unclear. Colleagues working on the same case have to search the conversations again.

What we set out to improve

• Whether incoming documents can be found under the correct customer, policy and claim
• Consistency with the agency’s folder structure and file naming rules
• Correct handling of review cases and links between duplicate messages and existing files

How the system works

1. 01Receive messages, photos and PDF or Office documents through the agency’s official WhatsApp Business Platform number. Cloud API is the official connection that passes messages from that number to the workflow. Preserve the original file unchanged.
2. 02Match the sender’s number against the agency’s customer system or list. If there is no match, place the record in the approval queue; show any policy number read from the document as evidence for a suggested match.
3. 03Suggest a document type from the agency’s list: damage photo, accident report, vehicle registration, driving licence, policy, invoice or loss adjuster report.
4. 04OCR converts text in an image into readable text. Use it to extract fields such as the policy number, accident date and policy expiry date from images and PDFs, leaving unreadable fields blank.
5. 05Assign eligible files a name and folder using the agency’s rules. An example folder structure is Customers / Customer code / Policy / Claim. An example filename is 2026-10-12_MUS-0142_HSR-0031_HasarFoto_01.jpg, ordered by date, customer, case, type and sequence. HasarFoto is the agency’s example code for a damage photo.
6. 06File the document in the storage chosen during discovery: Google Drive, SharePoint/OneDrive, the agency’s server or S3-compatible storage. S3 compatibility means the storage supports a common file connection format.
7. 07Write the sender, date, document type, extracted fields, file link and reviewer, where applicable, to the archive record.
8. 08Send a receipt acknowledgement within WhatsApp’s 24-hour customer service window. The message confirms receipt of the file only.

Methods we used

• Connect message receipt, customer matching, file naming and storage in an n8n workflow.
• Match the sender’s number against customer records, using the policy number read from the document as supporting evidence.
• Keep type suggestions within the agency’s document list and extract only the fields needed for filing.
• Link a repeat message to the existing archive record instead of creating another copy of the same file.
• Organise documents by customer, policy and claim so they can serve as inputs to a separate claim document pre-review workflow.

Where people stay involved

Data and security

Only fields needed for the archive are processed. Content sent to AI is limited to what is needed to suggest a document type and read those fields. Existing storage permissions are preserved, and retention follows the agency’s policy. Receipt, classification, corrections and approvals are recorded in the audit log. The system does not delete files or overwrite existing ones. Rejecting a file for archiving preserves the original received attachment.

Who this suits