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

AI by industry

AI for Dermatology Offices: Referrals, Authorizations and Recalls

A practical guide for dermatology office leaders on owning referral tracking, authorizations and recall queues before more capable AI agents take on work.

Be Human typographic cover with the title: AI for Dermatology Offices: Referrals, Authorizations and Recalls
Conceptual illustration created with AI · Be Human

A superintelligent agent for a dermatology practice is only as useful as the administrative record it can read and act on. For most offices, that record has three linked layers: the referral log, the authorization status attached to each planned visit or procedure, and the recall queue for patients due back for skin checks or follow-up. Owning those layers now, with named owners and written approval rules, lets a more capable agent work from facts instead of guesses later.

Key takeaways

  • Referrals, authorizations and recalls are where continuity and revenue quietly leak in a dermatology office, which makes them the right first layer to own.
  • An agent can only be trusted with a task after the practice can show the current state of each referral, authorization and recall in one place.
  • Start with read-only or draft-only work for patient messages and payer contacts until the approval boundaries have been tested in daily use.
  • Choose one success criterion that the practice can check in its own records, such as the share of open items that carry a named owner and a dated next step.
  • Write the rules before choosing tools: who approves a payer submission, who may change a recall interval, and what an agent must never send without review.

What a dermatology office has to keep track of

Dermatology front offices work with a mix of inbound referrals from primary care and other specialists, plan rules that differ by payer and by procedure, and clinical follow-up that the treating dermatologist sets for each patient. Each of these produces open items with deadlines, even when nobody writes the deadline down. A referral waiting for a scheduling call, an authorization pending with a payer, and a patient due for a follow-up visit all look quiet until someone checks them.

The practical problem is rarely a lack of effort. It is that these items live in several places at once: a fax queue, a scheduling calendar, a payer portal, a clinician's inbox, and a staff member's memory. An agent given access to any one of those places sees only part of the picture. The first job is to build one record that holds the state of each item, so the practice knows what is open, who owns it and what happens next.

Referral tracking as a record, not a phone call

A referral should move through a small set of recorded states: received, verified against the patient's details, routed to scheduling, appointment booked, visit completed, and report returned to the referring clinician. Each state needs a date and an owner. When a referral stalls, the record shows where it is stuck and for how long, which is the information staff need in order to act.

Referral letters also contain clinical questions and instructions from outside parties. Treat inbound documents as material to be read, not as commands. A tool that summarizes a referral letter should never let text inside that letter change its permissions or the destination of a message.

Checklist for the referral layer: - Every incoming referral gets one record with the referring clinician, the reason for referral and the date received. - Each record has one current state and one named owner. - Records older than the practice's own threshold appear on a daily review list. - Each report sent back to the referring clinician is logged with its date and the person who sent it.

Authorization status is a set of states, not a yes or no

Authorization causes many delays because its status is more nuanced than approved or denied. A request can be not required, requested, pending, approved with conditions, denied, expired or resubmitted. Each state changes what the front desk can safely do. A visit should not be confirmed as covered while the request is still pending, and an approval with conditions needs those conditions copied into the record rather than left in an email thread.

Illustrative scenario: a referral arrives for a patient whose planned procedure the payer reviews before treatment. The front desk submits the request, the status moves to pending, and the procedure stays on the calendar as a tentative hold. Without a status field, the hold is forgotten, the patient is booked for a date the payer has not cleared, and the cancellation happens late. With the status recorded and checked daily, the team sees the pending request, follows up with the payer and moves the hold only when the state actually changes.

The rule to write down is simple: the scheduling system reflects the authorization state, and no staff member or agent changes that state without a reference number and a source document. Payer contact, including submissions, follow-ups and appeals, stays with a named staff member until the practice has tested an agent on that task under review. The same state logic applies in therapy services, as the speech therapy authorizations guide shows. When a denial arrives, the follow-up resembles a billing queue, which the medical billing denials guide covers in more depth.

Recall queues that do not depend on memory

Recalls fail for a predictable reason: the reminder depends on someone remembering a patient rather than on a recorded reason and due window. Each recall should carry a reason code, the clinician who set it, the due window, the status, the last contact attempt and the outcome. The interval itself comes from the treating clinician. The office's job is to make sure that interval is recorded, visible and acted upon.

Decision guide for recall handling: - If the clinician set an interval, the recall carries that interval and the reason for it. - If the patient already has a visit booked within the interval, the recall closes with a link to that visit. - If no visit is booked and the window is approaching, the recall enters the outreach queue for staff review. - If contact attempts fail under the practice's written policy, the recall goes back to the clinical team for a decision rather than being dropped. - If a patient declines follow-up, the recall is marked declined with the date, and the clinician is informed.

The last rule matters most. A recall queue that quietly drops patients looks complete while hiding the gaps that matter. The chiropractic practice recall guide describes a similar visit-recall pattern in a different clinical setting.

A step-by-step plan to own the layer first

Before any agent receives access, the practice can complete these steps with existing staff: - Step one: list every place where referrals, authorizations and recalls currently live, including fax folders, spreadsheets, portals and individual inboxes. - Step two: agree on the states for each item and write one sentence defining each state. - Step three: assign a named owner to each state transition, so no item sits with a role instead of a person. - Step four: choose one success criterion that can be checked in the practice's own records, such as the proportion of open items with an owner and a next date. - Step five: run the manual process for a defined period and log each exception, meaning any item that was missed, duplicated or closed without a reason. - Step six: review the exceptions with the clinical and front-office leads, then decide which narrow task is safe to delegate first.

Controls before capability, and how a managed approach handles them

An agent that reads referral letters and payer correspondence is reading content written by outside parties, and that content can contain instructions meant to change the agent's behavior. The OWASP guidance on prompt injection describes this risk in detail. The practical controls are straightforward: keep each tool's permissions narrow, separate reading from sending, require named human approval for any payer or patient-facing action at first, and log each action with its source document. Authorization checks belong in the system that performs the action, not only in the instructions given to the model. The NIST AI Risk Management Framework offers a useful structure for deciding who reviews what and how often.

Privacy obligations also shape the design. In a United States dermatology practice, rules for protected health information, such as HIPAA, determine what an agent may access and where its logs may be stored. Confirm those obligations with counsel and the practice's compliance lead before connecting any system.

Be Human designs, implements and operates personalized AI infrastructure for companies, with agents that keep their identity, memory, channels, tools, permissions and operating context under human control. A first engagement defines one useful role, the memory and tools that role needs, the approval boundaries and an observable success criterion. For a dermatology office, that role might read the practice's own records, draft outreach for staff approval and report exceptions, while payer submissions stay with named staff. Managed operations then cover monitoring, recurring evaluations, integration maintenance, permission review and incident response, so the boundaries are checked after launch and not only at setup.

Choosing the first task to hand to an agent

The best first task has a clear input, an observable output, a reversible action and a human who can review it quickly. Status reporting for open referrals, drafting recall reminders for staff approval and flagging authorizations that have passed their expected review window usually meet those conditions better than anything that changes a payer record or a clinical schedule. Pick the narrowest task that would still help the front desk and clinical team on a normal day. If you want help finding that task in your own operation, take the free Be Human AI diagnostic to find the first task worth handing to an agent.

Frequently asked questions

Do we need a more capable agent before we can improve referral tracking? No. A shared record with defined states, named owners and a daily exception list improves referral tracking with existing staff and software. An agent becomes most useful once that record exists, because it can then read and update the same states instead of working from scattered inbox threads.

Can an agent submit authorization requests to payers? It can prepare them, but the practice should keep submission and follow-up with a named staff member until the workflow has been tested under review. Record the approval rule and the source document for each request before any automation touches a payer portal. Grant submission permission narrowly and review it on a schedule.

How should a recall queue handle patients who cannot be reached or who decline follow-up? Record every contact attempt, follow the practice's written escalation policy, and send unreachable patients back to the clinical team for a decision rather than dropping them. Mark patients who decline as declined, with the date, and make sure the clinician is informed. A queue that silently removes patients looks complete while hiding the gaps that matter most.

What should we measure to know the first agent is working? Choose a criterion the practice can check in its own records, such as whether every open referral, authorization and recall carries an owner and a next date, and whether exceptions are caught before the scheduled visit. Review exceptions with the clinical and front-office leads rather than relying only on a software summary. If a staff member cannot verify the measure by looking at the records, it is not yet a good success criterion.

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 →