One Roof, Two Businesses: Why Dermatology AI Has to Be Built for Both Sides

A practice administrator at a dermatology office in Aventura is looking at the morning roster. Same provider, ten patients on the books before noon.

A full body skin check at 9:00, established Medicare patient. A biopsy follow-up at 9:15, results conversation in Spanish. A Botox consult at 9:30, new patient, cash pay. A Dupixent injection at 9:45 for a long-running atopic dermatitis case. A Mohs follow-up at 10:00. Laser hair removal at 10:15, third visit of a six-treatment package. A psoriasis follow-up at 10:30 with a biologic decision pending. A cosmetic consult at 10:45, returning patient considering fillers. Pediatric eczema at 11:00. CoolSculpting consult at 11:15.

One provider. One front desk. One phone line. One Electronic Health Record (EHR).

Two completely different businesses.

This is the structural reality of every dermatology practice in South Florida that runs both a medical and an aesthetic line. The two sides share staff, software, and a roof. They share almost nothing else. And almost every AI tool sold into dermatology today is built for one side and treats the other as an afterthought.

That mismatch is where most dermatology AI implementations quietly fail.

Why this is two businesses, not one

The medical side runs on insurance. Pre-authorizations, denials, appeals, biopsy logistics, pathology workflows, screening recall, chronic disease management. Margins are tight. Volume is high. The patients are usually established. The job is operational.

The aesthetic side runs on cash. Consultation conversion, membership programs, retention sequences, online reputation, promotional cadence tied to specific procedures. Margins are higher. Volume per patient is lower but lifetime value is heavier. The patients are often shopping. The job is sales engineering wrapped in clinical excellence.

A medical dermatology patient is rarely comparison shopping in any meaningful way. They came because their primary care physician sent them, because the spot on their arm changed shape, because the eczema flared again. They picked the practice because it was in network, or close to home, or both. The job of the AI is to keep that patient flowing through the operational pipeline without leaks.

An aesthetic dermatology patient is comparison shopping by default. They saw the practice on Instagram, on Google, on a friend's recommendation. They are evaluating other practices in parallel. The price point is high enough that the decision is considered. The job of the AI is to move them from inquiry to consultation to first procedure with as little friction as possible, and then to keep them coming back year after year.

A platform built for one of these jobs will fail at the other.

What the medical side actually needs

The medical side of a South Florida dermatology practice has three operational workflows where AI produces real and measurable lift. None of them are chatbots.

Biologic prior authorization is the heaviest. Dupixent for atopic dermatitis runs roughly $40,000 a year per patient. Skyrizi and Cosentyx for psoriasis sit in the same range. Insurance plans deny these aggressively on first submission. Each appeal eats two to four hours of staff time. Most practices have one person who handles biologic auths full time, and that person is the bottleneck for every patient who needs one of these drugs. An agentic system that drafts the appeal letter, pulls the supporting clinical documentation from the chart, monitors the response timeline, and escalates to staff when a case stalls can move that work from hours to minutes per case. The practice is not buying a chatbot. It is buying back the staff time of a full-time position and expanding how many patients can actually access the medications they need.

Biopsy follow-up is the second. Patient comes in for a suspicious lesion. Biopsy is taken. Sample goes to pathology. Results come back in three to seven days, sometimes longer. Patient gets a call, a portal message, sometimes a letter. The system fails when the call goes unanswered, the portal message goes unread, and the letter sits in a drawer. Practice administrators know what their notification gap rate is, even if they do not measure it formally. AI agents that orchestrate multi-channel follow-up, escalate when no channel is working, and document every contact attempt in the chart are doing the job a human coordinator would do, except they do not get sick and they do not quit.

Skin cancer screening recall is the third. South Florida has among the highest rates of basal cell, squamous cell, and melanoma in the country, driven by year-round sun and a patient population that includes both retired snowbirds and a large outdoor-working community. Patients on annual or six-month screening protocols do not show up unless somebody actively brings them back. Mass mailings work for nobody. Generic SMS reminders work poorly in a population where many patients communicate primarily through WhatsApp. AI agents that adapt channel and language to patient preference, schedule against actual provider availability rather than a static template, and handle the rebook conversation when a patient calls back are doing operational work that human staff can do but rarely have time to do consistently well.

What the aesthetic side actually needs

The aesthetic side runs on different math. Higher margins, fully cash, much harder to acquire customers, much easier to lose them. Three workflows where AI is genuinely useful, none of which look like the medical side.

Consultation conversion comes first. A cosmetic consult is the front door of every aesthetic practice. The conversion rate from consult to first procedure varies between thirty and sixty percent depending on the practice, the procedure mix, and the consult experience itself. AI agents that engage the prospect before the appointment, surface the relevant pre-procedure information in their preferred language, set realistic expectations on cost and recovery, and handle the appointment confirmation logic move conversion noticeably. Unlike the medical side, where the patient has already committed to the visit, the aesthetic patient is shopping. The AI here is a sales engineering layer with clinical accuracy.

Membership and retention is the second. Most successful aesthetic practices in the tri-county area now run some version of a membership program. Monthly fee, included treatments, perks for additional services. Administering these manually is brutal. Tracking who has used what, who is approaching their treatment cap, who is up for renewal, who has quietly stopped showing up. AI agents that maintain membership state, surface at-risk members before they churn, and handle the renewal communication sequence solve a real and recurring administrative burden that most practices currently absorb as a hidden cost.

Online reputation is the third. Aesthetic dermatology is one of the most review-driven specialties in healthcare. New patients almost universally check Google, Yelp, and RealSelf before booking a cosmetic consult. The practices that win this systematically request reviews from satisfied patients at the moment of satisfaction, route negative feedback into a service recovery flow before it becomes a public review, and respond to every public review in a tone that matches the brand. AI agents that orchestrate the review request timing, route the response logic, and draft the responses for staff approval are doing work that practice managers know is important and almost never have time to do consistently.

Why generic AI fails in dermatology

Most healthcare AI platforms are built for the medical side. They speak EHR. They speak insurance. They speak HIPAA. They do not speak cash-pay psychology, membership state, review economy, or aesthetic patient acquisition. When a dermatology practice implements one of these platforms across the whole operation, the aesthetic side either gets left out or gets retrofitted with workflow patterns that were designed for a different business model.

Most marketing AI platforms are built for the aesthetic side. They speak conversion. They speak reviews. They speak membership. They do not speak prior auth, biologics, biopsy follow-up, screening recall, or multilingual chronic care management. When a practice tries to make a marketing AI tool serve the medical side, the medical side either gets ignored or gets retrofitted with workflow patterns that were designed for a different business model.

Buying both is worse than buying one. Two platforms means two data silos, two onboarding sequences, two vendor relationships, two contracts. The patient data does not flow between them. A patient who came in for a cosmetic consult and got identified as having a suspicious lesion is invisible to the medical side until somebody manually creates a referral. A medical patient who would be a strong cosmetic candidate is invisible to the aesthetic side until somebody manually flags it. The integration that should happen automatically does not happen at all.

The architecture has to be designed once, with both business models in scope, and the workflows for each side configured on top of a shared patient and operational layer. That is consulting work, not vendor work.

A working architecture has three layers.

Patient identity and clinical context sits at the bottom. Who is this person, what conditions do they have, what is their language preference, what is their channel preference, what is their insurance state, what is their membership state, what is their cosmetic interest profile. This layer has to be accurate, current, and accessible to every workflow above it.

Workflow orchestration sits in the middle. The medical workflows and the aesthetic workflows live here. They share the patient layer below but they are configured separately because they have different rules, different KPIs, and different escalation logic.

Institutional memory sits across both. What did this practice already decide about this patient. What treatments have they already had. What promotions have they already responded to. What objections have they raised in the past. What language did the staff use that worked. This is the layer that stops a system from re-explaining context every time a workflow runs, and it is the layer that almost no off-the-shelf platform actually delivers at the level of fidelity that produces real productivity gains.

Our team built and operates LawMem.ai, a memory-as-a-service product for legal AI agents. We are extending the same architecture, HIPAA-aligned, into healthcare. The point is not the product itself. The point is that we have built this layer from the foundation and we know where these systems actually break in production.

The South Florida specifics

Three local realities change every architecture choice for a dermatology practice in the tri-county area.

The first is multilingual reality. A practice in Aventura, Doral, Brickell, or Coral Gables operates in Spanish at least half the time, often more. A practice in Pembroke Pines or Plantation may handle Spanish, Haitian Creole, and Portuguese in the same week. Generic AI tools that handle Spanish as a translation layer fail on tone, fail on register, and fail on the multi-language switching that real intake conversations actually involve. Multilingual has to be a first-class architectural concern in the design, not a feature toggle.

The second is snowbird seasonality. The practice that runs at one volume from May to October runs at a different volume from November to April. The recall logic, the membership communication cadence, and the cosmetic consultation flow all have to adapt to that swing. The system has to know when to push harder and when to ease off.

The third is hurricane disruption. South Florida practices lose days to storms every year. The recall queue, the appointment logic, and the staff communication sequences all have to handle that reality. AI agents that do not account for storm cancellations create more work than they save when a storm hits.

What to ask before signing anything

Five questions for any vendor pitching dermatology AI, on the second call, before the pilot.

First, on the dual business model. How does your platform handle the difference between insurance-driven medical workflows and cash-pay aesthetic workflows? If the answer treats them as the same thing with different settings, the vendor has not actually thought about dermatology.

Second, on patient context flow. If a cosmetic patient is identified as having a clinical concern during their consult, how does that move into the medical side of the practice in your system? If the answer involves manual data entry, the integration is not real.

Third, on multilingual. How does your system handle a Miami-Dade intake conversation that switches between Spanish and English mid-sentence? Demo it on a real example, not a sanitized one. If the demo refuses, that is the answer.

Fourth, on biologic auth. Show a sample appeal letter your system would draft for a Dupixent denial. Pick the patient profile yourself. Is it good enough that your senior staff would sign it.

Fifth, on data. Where does the patient data live. Does the model train on practice data either during this engagement or in aggregate. What does the contract say about retention, deletion, and the BAA terms.

The vendor that answers all five cleanly is rare. The vendor that deflects on any of them should be running a deeper pilot, not a larger rollout.

The window

Dermatology AI is going to consolidate. The platforms that survive will be the ones that recognize the two-business reality of the specialty and build for it. The platforms that do not will quietly fade. Practices that move now will have an architectural advantage they will not have in two years.

The practices that get this right are not the ones that buy the loudest tool. They are the ones that treat AI adoption as an architecture decision, scoped to the actual two-sided shape of their business, with help from advisors who have seen these systems break and rebuild in production.

The window is open. It will not stay open forever.