- Start with administrative work: intake packets, referrals, authorization tracking, follow-up.
- Clinical documentation support can help too, but a clinician reviews and signs every note.
- HIPAA's minimum necessary standard applies. The system should see only what each job needs.
- Work within your compliance framework, with your compliance lead involved from the first week.
- Keep a record of what the system did and who approved it.
The queue behind the front desk
It's 7:40 on a Monday. The fax queue has 63 pages from the weekend. Some are referrals. Some are records requests. Two are lab results that belong in a chart today.
Someone sorts them by hand, one page at a time, while the phones start.
That scene repeats across every location. It's where AI helps a practice first: reading, sorting and drafting, so staff spend their time on patients instead of paper.
Where AI actually helps
Intake packets and forms
A new patient sends a packet: demographics, history, medication list, a photo of an ID. A staff member retypes it into the practice management system.
AI can read the packet, fill a draft record, and flag what's missing or unreadable. Staff confirm each field before it's saved. The retyping goes away. The checking stays. Our document processing page shows the general pattern.
Referral intake and tracking
A referral arrives by fax. It needs to be matched to a patient, routed to the right specialty and location, and scheduled. Then someone needs to know if it stalled.
A referral system can read the incoming document, propose the match and the route, and put it on a board with its age. Staff approve the routing. Leaders see which referrals have sat for a week without an appointment.
Prior authorization tracking
Prior authorizations live in spreadsheets, sticky notes and payer portals. Nobody has the whole picture until a patient arrives and the authorization isn't there.
AI won't decide anything here. What it can do is assemble the documentation a request usually needs from the chart, draft the cover note for staff to check, and keep every request on one list with its status and age.
Timing rules are changing on the payer side. Under the CMS Interoperability and Prior Authorization Final Rule, certain government program payers must send decisions within 72 hours for expedited requests and seven calendar days for standard ones, with a compliance date starting January 1, 2026. A clear tracking list helps your team notice when a request has gone quiet.
Documentation support
A clinician finishes a visit and has twelve minutes before the next one. A draft note, produced from the visit with patient consent handled the way your policies require, saves real time.
The clinician reviews, corrects and signs every note. The draft is a starting point, never the record. Many electronic health record systems now offer their own drafting features; check what yours includes before building anything.
Patient messages and follow-up
A shared inbox fills with refill requests, scheduling questions and billing questions. AI can sort them, draft routine replies for staff to approve, and send clinical questions straight to the right care team without a draft.
Records requests
Requests for records arrive from patients, other practices and attorneys, each with different rules about what can be released and to whom.
AI can read each request, sort it by type, check that the needed authorization is attached and draft the response package for staff to verify. Your release-of-information policy decides what goes out. Staff send it.
Where it doesn't help (yet)
Diagnosis and treatment decisions. These belong to clinicians. A tool that suggests diagnoses or treatments raises different regulatory questions and should not be confused with administrative help.
Unreviewed messages to patients about their care. A wrong answer about a medication or a result can hurt someone. Clinical replies need a clinician.
Coding without review. AI can suggest codes from a note. Your coders and clinicians remain responsible for what gets submitted.
Messy source systems. If two locations record the same thing in different places, AI will reflect the mess. Sometimes the first project is agreeing on where things live.
Questions to settle before you build
- Which queue causes the most patient-facing delays, and who works it today?
- What does your compliance lead need to approve before any patient information is used?
- Which systems hold the information each job needs, and who controls access to them?
- How will you log what the system did, and who will review those logs?
- What happens when the system is unsure? Who gets the item, and how fast?
Written answers to these questions become the project's ground rules. They also make the conversation with your compliance lead much shorter.
The NIST AI Risk Management Framework is voluntary, but it gives leadership and compliance a shared language for how the practice manages AI risk.
How to start safely
Bring your compliance lead in first. We work within your HIPAA framework and policies, with your compliance or privacy lead deciding how patient information may be used and where it may go. Don't treat this as paperwork for the end.
Limit what the system sees. HIPAA's Privacy Rule requires reasonable efforts to limit protected health information to the minimum necessary for the purpose. A referral router doesn't need a full chart. Design each job around the least information it needs.
Keep a record. The HIPAA Security Rule calls for audit controls that record and examine activity in systems containing electronic health information. Log what the system read, what it proposed and who approved it.
Start with one queue. Referrals or intake packets are usually the best first choice: high volume, clear rules and a person already doing the checking.
Measure it. Days from referral to appointment. Minutes per intake packet. Authorizations that were missing at check-in. Write down today's numbers first.
Keep review fast. Our guide to human review for AI covers how to design a review step staff will actually use.
What a realistic first release looks like
Take referral intake as the example. The first weeks go to agreeing on how referrals should be handled across locations and collecting real samples, scans and all.
Then the system runs in shadow mode: it reads each referral and proposes a route, while staff keep working as usual and compare. Where it's wrong, the rules get tightened.
Once staff trust it, they start approving its proposals instead of doing the work from scratch. After a month or two, you compare days from referral to appointment against your before number.
If it helped, the next queue is usually close by: intake packets, records requests or authorization tracking.
What changes for your team
Front-desk staff stop retyping and start checking. That's a real change in the day, and it goes better when they help design the screen they'll check on.
Referral coordinators get a list instead of a pile. They can see which referrals are new, which are waiting on records and which have gone quiet, at every location.
Clinicians who use drafting support still own every word in the chart. Some will love it. Some will prefer to write their own notes. Both are fine. Don't force it.
Practice leaders get a view they rarely have today: how long each queue takes, by location, week over week. That's often where the next improvement comes from.
What most guides miss
Fax is still the main road. Many referrals and records still arrive as faxed images. A system that only handles clean digital forms misses most of the work. Plan for poor scans from the start.
Every location does it a little differently. Before anything is automated, the locations usually need to agree on one way to handle a referral. That conversation is often worth as much as the software.
The vendor's terms matter as much as the model. Which AI provider sees the data, under which agreement, is a question for your compliance lead and your counsel before any patient information is used.
Front-desk staff are the experts. They know which referring offices send incomplete packets and which fields are always wrong. Build with them, not around them.
Small wins compound. Saving three minutes per referral sounds minor. Across every location and every day, it's a coordinator's week back each month, and patients seen sooner.
How we do it
We work alongside your staff and your compliance lead, starting with a free intro call to find the queue worth fixing first. We connect to the systems you already run, such as athenahealth or eClinicalWorks, with access your team approves. See our healthcare page and our AI systems work.
