AI Claims & Document Automation Bot
Problem
Back-office teams that handle client claims — legal, insurance, and financial departments — spend most of their time on paperwork rather than decisions. Each request rarely depends on a single document: a case is resolved only by cross-referencing several sources at once — a certificate or ID scan the client provides, together with the client's records held inside the company. Doing this by hand is slow, easy to get wrong, and hard to keep consistent with legal requirements. Our client needed a way to let people submit their documents through a channel they already use, and to have the routine cross-checking and document drafting handled automatically.
Result
Claims that used to take a specialist hours of manual cross-checking are now prepared in minutes. Clients submit their documents in a familiar messaging app and receive a correctly formatted result they can act on, while staff are freed from repetitive paperwork to focus on the cases that genuinely need judgement. Because the AI reasons over every relevant document together and follows the legal requirements built into its instructions, the output is more consistent and less error-prone than manual handling — and the same system scales to new request types without rebuilding the workflow.
How did we achieve it?
We built a messaging bot — delivered on both Telegram and WhatsApp — that walks the client through submitting the documents a given request needs. Unlike simple OCR, the system does not process one document in isolation: it parses the client-supplied files alongside the company's own records (for example, spreadsheets of the products and services a client holds, read with dedicated Excel/data-parsing libraries) and passes everything to a language model so it can reason across all of them at once. The AI is orchestrated through OpenRouter/LiteLLM, which lets us route to the most suitable model and tune it for the task. A carefully engineered prompt handles the data extraction and alignment, makes sure the result respects the relevant legal requirements, and fills a document template that is sent back to the client so they can proceed.
Example scenarios it handles
The same engine drives a range of document-driven requests — these are two of the scenarios we implemented, with others built on the same pattern:
- Inheritance & asset export. A relative submits a death certificate; the bot pulls the deceased person's holdings from the company's records, reconciles the two, and produces a document listing what the person held — ready for the client to continue the process.
- ID / passport change. A client uploads a scan of a new passport plus a confirming document; the bot matches it against the person's existing data and drafts an inquiry confirming it is the same individual, with the same assets, under updated identity documents.
- And beyond. The same approach extends naturally to address and contact updates, beneficiary or ownership transfers, and general claim intake and routing — any request that means combining a client's documents with internal records to produce standard paperwork.