Sorting theMorning Mail
How an insurance team used claims processing automation to eliminate manual document triage for 400+ daily documents.
Handlers start each morning with fully contextualized, pre-sorted cases. Zero manual triage. Each document is uploaded and linked to its extracted data (category, key fields, and file reference), ready for the claims system.
An insurance operations team receives 400+ documents daily, all related to claims. Claim forms, medical reports, police reports, repair estimates, and correspondence all arrive mixed together via email and scanned mail. A clerk opens each one, reads it, figures out which claim it belongs to and what type of document it is, then forwards it to the right handler. It takes until noon just to sort the intake. This is insurance document processing before automation.
Ingestion
With the pipeline in place, documents flow in from an S3 bucket (where the email gateway deposits attachments) and an HTTP API (connected to mailroom scanners).
Structure Classification
Each incoming document is evaluated by multiple Structure Classifiers in parallel, one per document type. A claim form classifier defines the fields expected on a claim form, a medical report classifier defines medical report fields, and a repair estimate classifier defines repair estimate fields. Each classifier checks whether the document matches its structure. The matching classifier identifies the document and assigns it the corresponding schema.
Extract & Map
Each Structure Classifier feeds its own Extractor configured to extract and map the data for the claims system. The key: each extractor adds a hard-coded category value ('claim', 'medical', or 'repair') so the claims system knows exactly what type of document it's receiving without needing to classify again. The payload is shaped for the target system, not just forwarded raw.
Delivery
The pipeline delivers to two separate endpoints. The file itself is uploaded via a dedicated file upload API; the claims system returns a file ID. The extracted data (category, key fields, and the file ID referencing the uploaded document) is then sent to a second API endpoint. This separation means the claims system receives both the original document and extracted data that links back to it. Handlers start their day with pre-sorted, fully contextualized cases: insurance claims automation from intake to delivery.
Multiple Input Sources
Documents arrive from different systems (email gateways, scanners, APIs) and converge into a single pipeline. The processing logic stays the same regardless of where the document came from.
Classify-First Routing
Documents are classified by structure before any extraction happens. The classification result determines which extraction schema is applied to the document.
Extract & Map
Each extractor shapes the payload for the target system, adding hard-coded values like a category tag alongside extracted fields. The downstream system receives ready-to-use data, not raw documents.
Structure-first classification: each document type has its own structure definition. The matching structure drives extraction, and per-type extractors shape the payload with hard-coded values. File upload and extracted data are delivered separately, linked by file ID.
Claims Questions
What insurers ask before automating claims document intake.
The pattern above is the answer in production form: every incoming document is classified to its claim and type, structured data is extracted against a claim schema, cross-document checks validate names, dates and amounts, and adjusters receive complete structured claims in minutes instead of after a morning of manual sorting.
Yes. The pipeline runs on-premise or in Swiss hosting with local AI models. Claims forms, medical reports and police records never leave your controlled environment, which keeps GDPR and revDSG compliance architectural rather than contractual.
See a Pattern That Fits Your Workflow?
Every pattern is built from the same modular components: classifiers, extractors, validators, and connectors. Let's find the right combination for your documents.