Document classification
Labs, imaging reports, consult notes, hospital discharge summaries, immunisation records, insurance cards, legal requests and junk.
05 · Communication
AI Document & Referral Agent
Fax is still how a huge share of clinical information moves. The Document Agent reads it, decides what it is, files it in the right chart, and puts anything actionable in front of a human.
Agent at a glance
- StatusIn development
- CategoryCommunication
- Deployment2–4 weeks, shadow mode first
- Review modelHuman approval per action, configurable
- Headline metric85%+ Auto-filed
- EvidenceFull action log, exportable any time
85%+
Auto-filed correctly
Classified and matched without human touch on the first pass.
< 5 min
Fax to chart
Versus the hours a pile waits on a desk.
100%
Abnormal findings flagged
Anything critical goes to a clinician queue, not a folder.
What it does
Everything in this agent, in writing.
Patient matching
Name, date of birth, MRN, phone and address scored together; ambiguous matches go to a human rather than being guessed.
Structured extraction
Dates, ordering provider, results, medications and follow-up instructions pulled into searchable chart fields.
Referral intake and closing
Referrals created with the reason, urgency and insurance, then tracked until the consult note comes back.
Orders and results routing
Unsigned orders, results requiring acknowledgement and critical values queued to the right clinician by policy.
Fax-out and prior records
Incoming requests for records and outgoing faxes handled with the same logging and consent checks.
Deployment
Live in weeks, not quarters.
01
Volume and type audit
We sample your inbound documents to quantify type, volume and the current handling cost.
02
Classification tuning
Your document types and filing rules trained on your own sample.
03
Match threshold set
You choose the confidence level required to file without review.
04
Continuous correction loop
Every human correction feeds back, and the exception rate is reported monthly.
Guardrails
What it will never do.
These limits hold regardless of configuration. They exist so that you can put an agent in front of patients without putting your licence next to it.
- Never guesses a patient match — ambiguous documents queue for a human
- Critical and abnormal results route to a clinician immediately, never silently filed
- Filing rules are versioned and reversible; any misfiled document is traceable and movable
- Malware and macro scanning on every inbound file before it is processed
- Retention and destruction rules follow your schedule, with deletion logged
Human in the loop, by your choice
Every action can require approval before it commits. Start with all of document agent queued to your staff or to ours, watch the accuracy numbers, then widen autonomy workflow by workflow — or leave it queued forever. Both are supported.
Escalations land with certified people
Our review queue is staffed by the same certified coders, billers and credentialing specialists who run our managed services. Not a support desk, and not a chatbot watching a chatbot.
Full evidence trail
Input, decision, action, confidence, model version, reviewer and timestamp — retained and exportable as CSV or JSON, including after you leave us.
What we report against
Operational targets for this agent. They are internal commitments measured on your own data, reported monthly — including the misses.
85%+
Auto-filed
< 5 min
To chart
99%+
Match precision
100%
Critical flags
Works with your systems
Your EMR, your numbers, your data.
- Athena
- eClinicalWorks
- NextGen
- Epic
- ModMed
- Updox
- Concord fax
- RightFax
We integrate through FHIR and HL7 where the vendor supports it, and through secure automation or structured document intake where it does not. No API required, no vendor commission taken.
FAQ
Common questions.
Is our fax line changing?
No. Documents keep arriving on your existing fax number. We intercept, classify and file them, and your staff keep working in the EMR as usual.
What stops it filing to the wrong patient?
A high match threshold, multiple identifiers scored together, and a rule that ambiguous matches always go to a human. Every filing records which identifiers matched, so a mistake is findable and correctable.
Can it read handwritten referrals?
Handwritten documents are processed with extraction accuracy reported separately. If confidence is low the document is queued for a human rather than filed with guessed data.
Do we need to change our fax vendor?
No. We work with the major fax and document platforms, and with plain inbound fax-to-email where that is what you have.
Related agents
Usually deployed together.
06
In development
AI Scribe
Listens to the visit and drafts the note in your format — SOAP, APSO or your own template — with suggested codes, orders and follow-up. The provider reviews, edits and signs.
- Ambient capture in the room or on telehealth
- Drafted in your template, in your voice
- Provider reviews and signs — the note is never auto-filed
07
In development
AI Coding & Charge Agent
Turns the signed note into codes and charges, checks them against payer rules and your fee schedule, and queues anything ambiguous for a certified human coder.
- ICD-10, CPT and HCPCS suggested from the signed note
- Payer-specific edits applied before the claim is built
- Certified coder reviews before release — always
08
In development
AI Intake & Eligibility Agent
Collects history, consents and insurance before the visit, verifies eligibility and benefits in real time, and writes clean structured data into the right EMR fields.
- Digital intake completed before the patient arrives
- Eligibility, deductible and co-insurance verified in real time
- Clean structured data written back — no retyping
See it run on your own workflow.
A 30-minute demo using your scenarios, your payer mix and your systems. We will show you the guardrails and the audit log, not just the happy path.
A 30-minute call with an operations lead. No obligation, no sales sequence, and we will say so if you do not need us.