What Happens When AI Reads Your Faxes

In our previous post on the fax machine problem, we described the reality most home health agencies face: 50 to 150 faxes per day, one to two staff members sorting them by hand, and a steady stream of misrouted documents creating downstream problems. The question we posed was whether the manual sorting process is the best agencies can do.

The answer is no. Intelligent document triage changes the equation entirely.

Eight Categories, One System

When a fax arrives, our platform reads the document using optical character recognition and then applies AI document understanding to determine what it is. Every incoming fax is classified into one of eight categories:

1. Physician Orders (signed and unsigned plans of care, verbal orders, medication orders)
2. Referral Packets (new patient referrals from hospitals, SNFs, and physician offices)
3. Face to Face Encounter Notes (F2F documentation supporting home health eligibility)
4. Lab Results (blood work, diagnostic reports, pathology findings)
5. Insurance Documents (authorization letters, eligibility confirmations, denial notices)
6. Discharge Summaries (hospital and facility discharge documentation)
7. Clinical Correspondence (physician notes, specialist reports, care coordination communications)
8. Nonclinical / Junk (advertisements, misdirected faxes, duplicate transmissions)

How the Classification Works

The process starts with OCR, which converts the fax image into text that software can process. But OCR alone is not enough. Faxes arrive in wildly inconsistent formats. Cover sheets vary from office to office. Some documents have clear headers. Others are handwritten. Some are partially cut off or transmitted at an angle.

This is where AI document understanding adds value. Rather than relying on keyword matching or template recognition, the system analyzes the full context of the document: its structure, the language used, the presence of specific fields like NPI numbers, diagnosis codes, medication lists, or signature lines. The model has been trained on thousands of real home health documents and understands the patterns that distinguish an order from a referral, or a lab result from a discharge summary.

Each classification comes with a confidence score. Documents that score above the threshold are routed automatically. Documents that fall below the threshold are flagged for human review, ensuring that edge cases still get the attention they need.

Patient Matching

Classification is only half the job. The other half is figuring out which patient the document belongs to. Our platform extracts patient identifiers from the document (name, date of birth, medical record number) and matches them against the agency's active patient roster. When a match is found, the document is linked to the correct patient record and routed to the appropriate team member.

This step alone eliminates one of the most common sources of error in manual fax processing. Misfiling a document under the wrong patient creates problems that can take weeks to surface and hours to untangle.

What Changes for Your Team

The staff members who used to spend their mornings sorting faxes can now focus on the work that actually requires human judgment: following up on unsigned orders, coordinating with physician offices, processing referrals, and supporting clinical staff. The sorting step is gone. Documents arrive classified, matched, and routed before your team touches them.

For unsigned physician orders specifically, the system identifies them immediately upon arrival and flags them in the orders tracking workflow. No more digging through a pile to find out which orders came back signed and which are still pending.

The Learning Loop

Every document the system processes makes it better. When a human reviewer reclassifies a document that was flagged for review, that correction feeds back into the model. Over time, the system learns the specific patterns of each agency's document flow, including the unique formats used by their most frequent referring physicians and hospital partners.

This is not a static rules engine. It is a system that improves with use, adapting to the specific reality of each agency's operations.