Quick answer
An AI agent can read a referral fax or portal message, pull out the patient, insurance, referring provider and reason for referral, list what is missing, and queue the appointment request for front-desk staff. A person confirms urgency and eligibility. Before any vendor sees a real document, get its answers on storage, access and retention in writing.
Key takeaways
- The agent reads referrals and finds what is missing. Staff decide urgency and book.
- Get the vendor's answers on storage, access, retention and model training in writing before sharing a real document.
- The agent never gives medical advice and hands off any clinical question.
- If it speaks with patients by phone, it says it is automated and offers a person at any point.
- The 30-minute assessment names your top two or three AI use cases.
Start your 30-minute AI assessment.
What can an AI agent do with a faxed referral?
It can read the referral, extract the patient's name, date of birth, insurance, referring provider and reason for referral, and compare them with a checklist for that specialty. It then lists missing items and queues the request for staff. It can also ask the referring office for what is missing. A person decides when the patient is seen.
Specialty offices often get referrals as faxes, scans and portal messages in different layouts. Retyping them is slow, and a missing insurance card or authorization number delays the patient. The agent takes the first pass:
- Read the document and pull out the key fields.
- Check them against your checklist for that visit type, such as imaging attached, authorization number, and referring provider contact.
- Create a draft intake record in the system your staff already use.
- Send a standard message to the referring office listing what is missing, if your policy allows it.
- Put the case in the queue with the referral date, so older ones rise to the top.
Worked example (hypothetical, not a client case). A cardiology practice in Newport Beach receives referrals by fax and secure email. The agent reads one that has no insurance member number and no recent echo report. It drafts the request for both items and queues the case. A coordinator reads the summary, sends the request and later books the visit. No results are claimed.
What should you ask an AI vendor before it sees referral documents?
Ask where documents are stored, who at the vendor can open them, whether your data trains their models, how long files are kept, how you get them back or deleted, and what access log you can see. Get the answers in writing. Set the system up with made-up test referrals before any real one goes in.
Put the questions in writing and keep the answers with your vendor file:
- Where is the data stored, and who at your company can open it?
- Is our data used to train your models or anyone else's?
- How long do you keep documents and summaries, and how do we get them back or deleted?
- Can we see a log of who opened each record and when?
- What happens on your side if a document goes to the wrong place, and how fast do you tell us?
- Can the agent work inside the system we already use, so files do not pile up in a second place?
Two habits help from day one. First, build and test with made-up referrals, so the setup is proven before a real patient's file goes in. Second, send the agent only what the intake step needs. A referral summary does not need a full chart.
What should staff decide instead of the agent?
Staff should decide clinical urgency, the type of visit, insurance exceptions and any contact with a patient about symptoms or care. The agent organizes paperwork and drafts messages. It should state that it is an automated assistant and hand off at once if anyone asks a medical question or describes an emergency.
Write the handoff rules before the build, not after. A short list is enough:
- Any mention of chest pain, trouble breathing, bleeding or similar symptoms goes to a person immediately, and the script tells the caller to call 911 for an emergency.
- Any question about a diagnosis, a test result, a medicine or whether something is serious goes to clinical staff.
- Any referral marked urgent, or that the agent cannot classify, goes to the top of the human queue.
- Any request to change insurance, cancel a procedure or release records goes to staff.
The agent can collect, sort, draft and remind. A person decides. See What can AI agents do for a small business? for the general pattern.
What changes if the agent also talks to patients by phone?
The phone script needs the same handoff rules, plus a clear greeting. The agent says it is an automated assistant, tells callers if the call is transcribed, offers a person at any point, and hands off at once for symptoms or emergencies. Decide in advance how long transcripts are kept and who can read them.
Three practical points follow:
- Open every call the same way: the practice name, a line saying it is an automated assistant, and how to reach a person.
- Make the path to a person work on the first try, and test it every week.
- Keep transcripts only as long as staff need them, and limit who can read them.
Test the phone script with staff playing patients before launch, including a caller who describes chest pain and a caller who only wants a human. Dental and med spa offices that book consumer visits face a different set of questions. See AI scheduling for dental offices and med spas.
How should a specialty practice start with AI intake?
Start with one referral source and one visit type. Run the agent in shadow mode for two weeks, with staff entering every referral as usual and comparing results. Only after the agent's summaries match staff work should coordinators begin starting from its drafts. Expand to a second source after that.
Shadow mode costs staff little and shows the real failure patterns: faxes that arrive sideways, referrals with two patients' names, and abbreviations your specialty uses that the agent reads wrongly. Keep a log of every difference. After two weeks, count how often the agent's summary needed a correction and what kind. Then decide whether staff can start from the draft.
Pick the first workflow with the tests in What processes should be automated first?.
Who helps Orange County medical practices with AI intake?
Preferred AI Partners is a fractional AI partner in Irvine that works with businesses of 10 to 200 people. It maps your referral flow, writes the checklist and handoff rules with your staff, and runs the shadow period. The 30-minute assessment looks at how referrals arrive today and names where to start.
The Orange County page and the services page describe the work. Ask any vendor to put its data handling answers and its handoff rules in writing before you share a single document.
