- Document AI classifies each page of a referral packet and extracts the fields your team needs into structured data, validated against the schema you define.
- Missing information is flagged at intake, so coordinators can request it from the referring provider on day one.
- An optional model can predict which referrals are likely to stall, so coordinators follow up on those first.
- UCSF Health used H2O Document AI to process more than 1.4 million faxed documents a year, saving more than 25,000 hours of staff time and 5,000 hours of clinician time.
Every referral is read and keyed more than once
Referrals arrive as faxes and scans from many practices, each with its own layout. Someone in your access center has to work out what each document is, which patient it belongs to, what is missing, and where it goes, then type it into several systems.
That work delays the first appointment and keeps your coordinators on data entry instead of on patients. When volume rises, the backlog grows with it.
How referral automation works
H2O Document AI reads each document with intelligent character recognition, which learns document layouts rather than relying on fixed templates, and returns the fields you ask for.
Step 1: bring the documents in
Faxes and scans land in the queue you already use, and the extractor picks them up from there.
Step 2: classify each page
The extractor labels each page, such as referral, clinical notes, insurance card, or lab result, and groups pages into one packet.
Step 3: extract the fields
Patient demographics, referring provider, reason for referral, and insurance details come back as structured data that matches your schema.
Step 4: flag what is missing
Required fields that are blank or unreadable are listed, so a coordinator can request them right away.
Step 5: review and route
Flagged fields go to a person. Approved referrals go to the right clinic's work queue in the system you already run.
Ranking the queue
An optional model in H2O Driverless AI can predict which referrals are likely to stall, so coordinators follow up on those first. This step is separate from extraction: it adds a priority order to a queue that is already accurate.
Results
In its published case study, UCSF Health used H2O Document AI to process more than 1.4 million faxed documents a year, saving more than 25,000 hours of access center staff time and 5,000 hours of clinician time.