The Claimwell Mail Room replaces the manual indexing operation. Scanned mail, fax, email, portal uploads, and employer feeds go into one queue. Each item comes out classified, matched to its claim, labeled, and written back with the provider, the service dates, and the codes already filled in. Your team handles the exceptions. Every page is traceable.
Documents arrive the way they really do: multi-document scan batches, skewed pages, cover sheets, duplicate faxes, email replies without a reference number. The pipeline renders every page, reads it with layout-aware OCR and a vision model for handwriting and bad scans, classifies the document, extracts the identifiers and the domain fields, and searches your claims system for the match.
Handwritten dates, stamped provider names, degraded fax headers. Field-level confidence on every value.
Name, date of birth, last four of SSN, employer, claim number, reference token. Scored; thresholds are yours.
Duplicate detection across channels. The repeat is discarded with an audit record pointing at the original.
Ambiguous matches, unreadable pages, documents with no claim. Ranked candidates, one click to resolve, every resolution audited.
OCR is a component. The deliverable is a claim file where every document carries the fields your examiners and your downstream systems need, populated, cited, and scored. Mapped to your admin system's document and activity objects during implementation.
| Field | Example | Source | Every value carries |
|---|---|---|---|
| document_label | Attending Physician Statement · claim form · medical records · job description · authorization · employer statement · appeal letter · correspondence · other | classifier | confidence score page citation channel and received time batch and sender model and prompt version |
| claim_match | CW-204118, matched on name + DOB + employer | matcher | |
| claimant_identifiers | name, date of birth, SSN last four, employee id | extraction | |
| provider / facility | treating physician, practice, NPI where printed, fax-back number | extraction | |
| service_dates | visit date, date signed, period covered, surgery date | extraction | |
| diagnosis_codes | ICD-10 codes and descriptions as printed | extraction | |
| key_dates | last day worked, expected return to work, delivery date | extraction | |
| restrictions | as stated on the form, verbatim with page reference | extraction | |
| sender | provider office, employer HR contact, claimant, attorney | classifier + extraction | |
| duplicate_of | document id of the original, when detected | duplicate detector | |
| page_map | which pages belong to which document in a multi-document batch | splitter | |
| activity | "New APS received" on the claim, or an exception activity for a human | router |
It is an indexing product. It does not interpret, recommend, or decide. Your decisioning framework stays exactly where it is.
The standalone Mail Room is deliberately light: the document pipeline, the matcher, the write-back adapter, the exception desk, and the audit store. No workflow engine, no strategy layer, no model gateway beyond a single configured endpoint. It runs in your cloud account as one container stack.
Fuzzy search by name, DOB, SSN last four, employer, reference token. Read the claim to confirm the match.
Post the document with its label, extraction payload, confidence, and original pointer. Originals stay in your repository.
"New document" on the claim for your examiners, or an exception activity for the desk.
Claimant and provider contacts when a first-notice document arrives for a claim that does not exist yet.
Scan batch folder, fax inbox, mailbox, portal upload bucket, employer feed drop. Each is a drop point we read from.
Append-only. Every classification, match, discard, and human resolution, with inputs and model version.
Minutes, not days. Reported by channel and document type.
Share of items indexed with no human. The rest went to the desk with candidates ranked.
Audit-sample accuracy by field and document type, including degraded scans. Measured before go-live on your own mix.
Against the current vendor line. Backlog goes to zero on day one.
Every document the Mail Room indexes is already evidence. The next product reads the whole file, reconciles it, and tells your examiner what changed and where it says so. Same buyer, same adapter, nothing new from IT.
De-identified or synthetic, your choice. We run it through the pipeline and show you accuracy by document type before you commit to anything.