Skip to content
be-human.ai
← Back to the journal

AI by industry

AI Agents for Oral Surgery Offices: Referrals, Consent Forms, Recalls

Learn how AI agents can handle oral surgery referral intake, consent form tracking and post-op recall admin, with clinical decisions kept with the surgeon.

Be Human typographic cover with the title: AI Agents for Oral Surgery Offices: Referrals, Consent Forms, Recalls
Conceptual illustration created with AI · Be Human

AI for an oral surgery office works best as a set of bounded administrative agents: one that intakes referrals from general dentists and other referring offices, one that tracks consent forms against scheduled procedures, and one that runs post-operative recall messages and follow-up tasks. The surgeon keeps every clinical judgment, including candidacy, the surgical plan and when a patient can return to normal activity. Done this way, the agent removes paperwork chasing from the front desk without touching diagnosis.

Key takeaways

  • Start with three administrative lanes: referral intake, consent form tracking and post-operative recall. Keep diagnosis, treatment planning and clinical triage with the surgeon.
  • A referral agent should check every packet against the same required-field list and route gaps back to the sender instead of guessing.
  • Consent tracking means knowing which form was sent, which version was signed and which procedure it covers. It does not mean judging what the patient understood.
  • Recall messages should follow wording the practice wrote and approved, with an immediate human escalation path for any symptom language.
  • Define the owner, permissions and audit logs for each agent before the first live patient message goes out.

What an oral surgery office needs to hand off

Oral surgery offices run on a few repeated flows that consume staff time. Referrals arrive by fax, portal, email or phone, and each one needs to be complete before the surgeon can review it. Each procedure requires one or more consent documents, and each document has a version and a signing date that must match the record. Post-operative follow-up depends on calendar facts, such as when the check-in is due and when a review visit should be booked, more than on clinical notes.

Most of this work is coordination rather than clinical judgment. The surgeon's time belongs to examination, planning and surgery. The front desk belongs to scheduling and patient communication. An agent fits in the space between them, but only if its job is written down precisely enough that everyone can tell where it stops.

Be Human starts with one role, not a platform. A first engagement names the single lane, the records and tools the agent may touch, the approval boundary and an observable success criterion. For referrals, that criterion might be whether every packet reaches the surgeon with its required fields filled in before a consult is booked.

Referral intake: from inbound packet to surgeon review

A referral usually includes patient identifiers, the referring clinician, imaging, the reason for referral, the medication list and relevant medical history. When one of these is missing, the consult slips and someone on staff spends time on the phone. An intake agent can log what arrived, compare it with the list the surgeon defines and request what is missing using approved wording. It should not rank urgency, assess severity or suggest a procedure.

A workable intake flow looks like this:

  • Record the source, the channel and the date received, then open one file per referral.
  • Compare the packet with the surgeon's required-fields list: identifiers, referring clinician, reason for referral, imaging, medications and relevant history.
  • Send the referring office a request for each missing item, using template wording the practice approved in advance.
  • Route any message that mentions pain that is worsening, swelling, bleeding, fever or breathing difficulty to a staff member for same-day human review.
  • Place the complete packet in the surgeon's review queue with a short factual summary of what is on file and what is still open.

The summary lists documents and gaps. It is not a clinical assessment, and the surgeon's team should be able to verify every line against the source documents. Clinics with intake-heavy workflows face similar tradeoffs, as described in AI for Speech Therapy Practices: Evaluation Scheduling and Session Notes.

Consent forms: track what was signed, not what was explained

Consent creates a record problem more than a conversation problem for the office. Staff need to know which document was sent, which version the patient opened, whether it was signed, by whom and before which appointment. Tracking those facts is administrative. Whether the patient understood the risks, whether an alternative should be discussed and whether an existing consent still covers a changed plan are decisions for the surgeon and, where relevant, the practice's legal advisor.

A consent tracking agent can work from a short checklist:

  • Map each procedure type to its required forms, using a list the practice maintains and the surgeon approves.
  • Record the form version sent, opened and signed, with timestamps taken from the signing system, not from the agent's own guess.
  • Flag every upcoming appointment where a required form is unsigned, using the window the practice sets for reminders.
  • When the procedure plan changes, flag the existing consent for surgeon review instead of reusing it automatically.
  • Keep signed copies in the record system and make sure the agent has no permission to edit a signed document.

This is a useful place to see the difference between a helpful agent and an unsafe one. The helpful version reports a gap and waits. The unsafe version fills the gap by inventing a signature status or by sending a consent to a patient for a procedure the surgeon has not confirmed.

Post-op recall and follow-up admin in practice

Recall work covers the check-in message after a procedure, reminders for review visits the surgeon has specified, requests for stitch or suture removal timing that the surgeon has already defined, and follow-up with the referring office after a visit. Each of these depends on dates in the surgeon's instructions. The agent's job is to execute the schedule the surgeon wrote, record replies and keep the calendar honest.

Illustrative scenario: a patient had a surgical extraction on a Tuesday. The surgeon's post-operative instructions call for a check-in message on the second day and a review visit the following week. The recall agent sends the approved check-in text on schedule and offers an available review slot. The patient replies that the bleeding has not stopped and the swelling is growing. The agent does not answer with clinical advice. It sends the message to the on-call staff path immediately, logs the exchange and marks the file for surgeon attention.

The same recall logic appears in other clinical offices, such as the visit recalls described in AI Agent for a Chiropractic Practice: Intake and Visit Recalls. The principle is the same: the schedule is the agent's responsibility, and the medical content belongs to the clinician.

Permissions, patient data and the approval line

Each agent should have its own identity, access limited to the records its lane needs and a log of every read and write. Authorization guidance from OWASP points to least privilege and clear separation of duties, which in this setting means the referral agent can read packets and scheduling but cannot change clinical notes, and the recall agent can send only approved templates. Anything new, such as a message the template does not cover, goes to a person first.

Patient records are sensitive under health privacy rules, including HIPAA in the United States and LGPD in Brazil. Requirements vary by location and by the practice's own obligations, so the office should confirm them with its compliance advisor before connecting any system. Referral documents and patient messages are also untrusted input. A PDF attached to a referral might contain text that tries to instruct the agent. The agent should treat such text as content to summarize, not as a command, which is the kind of risk covered in the OWASP Top 10 for LLM Applications. The NIST AI Risk Management Framework offers a shared vocabulary for mapping these risks, measuring them and deciding who is accountable.

A practical approval line for this office usually looks like this: read access to referral packets and scheduling, no write access to clinical notes, outbound messages limited to approved templates and staff confirmation for any change to consent status.

Be Human runs this as managed operations. We design the referral, consent and recall roles with the surgeon and the office manager, define the memory and tools each role may use and configure the approval boundaries. After launch we monitor activity, run recurring evaluations on sample referrals and messages, review permissions when staff change and maintain integrations. If an agent acts outside its lane, we handle the incident and record the correction. Changes are made in small, controlled steps.

How to choose the first task and measure it

Choose one flow with clear inputs, a written rule set and a named owner. Referral completeness checks are often a reasonable first candidate because the required fields are known and the result is visible to the surgeon's team. Consent gap reporting is a close second. Recall automation usually comes after the team has seen the intake and consent logs behave well for a while.

Measure with criteria the office can check by hand. Useful examples include whether each referral reached review with its required fields filled in, whether consent gaps were flagged before the appointment and whether every recall message went out on the scheduled day with a log entry. Build a small test set from past referrals and messages, have the surgeon's team grade the agent's output and repeat the test whenever the templates or rules change. Evaluation methods such as those outlined in [OpenAI's evals guide](https://platform.openai.com/docs/guides/evals) can guide the structure, although the office's own rules remain the standard.

If you want to find the first task in your office worth handing to an agent, take the free Be Human AI diagnostic.

Frequently asked questions

Can an AI agent decide which patients need surgery or how urgent a referral is? No. In this design the agent checks completeness, routes messages and tracks dates, while the surgeon makes every candidacy, urgency and treatment decision. Any message that signals a possible complication goes to a human immediately, and the agent does not give clinical advice in reply.

Does an AI agent replace the surgical coordinator or the front desk? Usually it removes repetitive chasing, such as requesting missing referral items, checking whether a consent is signed and sending scheduled reminders. Staff still own exceptions, patient conversations that need judgment and every decision about records. Most offices find the coordinator's time shifts toward those exceptions.

How is patient data protected when an agent works on referrals and recalls? Each agent gets its own identity, access limited to its lane and a full audit log. Data handling must follow the rules that apply to the practice, such as HIPAA or LGPD, and the office should confirm the setup with its compliance advisor. Incoming documents are treated as content, not as instructions, to reduce the risk of manipulated input.

How does the office know the agent is working correctly? Set a small success criterion before launch, such as whether referral packets reach the surgeon complete or whether consent gaps are caught in time. Check a sample regularly against records that staff can verify by hand. Review permissions and templates whenever staff or procedures change, and treat any wrong output as an incident to fix, not a detail to ignore.

Your next step

See where an AI agent fits in your company first.

Take the free Be Human AI diagnostic. A few questions about your work show which task to hand to an agent first, what it needs to know and where a person approves.

Be Human editorial guidance. Examples are illustrative unless explicitly attributed. Sources provide background; implementation choices depend on your workflow.

Less busywork. More room for you.

Bring us the work you wish you had time for.

Book a call to try a personalized AI agent free for 7 days ↗Or take the free AI diagnostic →