A concierge medical and wellness practice in South Florida came to us to fix their marketing. Medical weight loss, hormone optimization, peptide therapy, IV drips, device-based body contouring and facial treatments — a full menu, experienced clinicians, a wall of genuine five-star reviews, and patients who stayed for years once they were in the door.
The problem was the door. And it wasn’t a marketing problem.
We ended up doing four things: rebuilding the website, wiring intake into the EHR the practice actually runs on, putting an AI receptionist on the phone, and fixing local search. What that work exposed is a pattern we now expect in almost every clinic we look at. Four leaks, all of them upstream of anything an ad budget can solve.
Leak 1 — The call that comes in at 7:40pm
A clinic’s phone rings hardest at the exact moments nobody can answer it. The front desk is checking in a patient, the injector is in the room, the office closed forty minutes ago, or it’s Saturday.
Wellness and aesthetic inquiries are impulse-adjacent in a way that primary care isn’t. Someone reads about GLP-1 therapy at 10pm, decides tonight is the night they finally do something, and calls three clinics. The one that answers gets the consult. The other two get a voicemail they’ll return at 10am — to someone who already booked.
What we built: an AI voice receptionist that answers on the first ring, at any hour. It knows the service menu, the hours, the location, and which treatments require a consultation before anything is scheduled. It captures a name, a callback number, and what the caller is asking about; it can book into an open slot; and it sends the front desk a summary so the morning starts with context instead of a callback list. During business hours it transfers a live caller who wants a human. Outside them, it takes a complete message rather than a beep.
The design decision that mattered most was what it refuses to do.
Design principle — an AI receptionist at a medical practice is a scheduling and intake assistant, not a clinician. “Is semaglutide safe with my blood pressure medication?” is not a question it should ever answer, however confidently it could generate a sentence. That boundary is enforced in the system prompt and backed by an escalation path to a human, not left to the model’s judgment.
That constraint isn’t a limitation we apologize for. It’s the reason the practice was willing to put it on the phone at all.
Leak 2 — The booking form that isn’t the booking system
Nearly every clinic site has a “Request an appointment” form. Almost none of them are connected to anything.
The form emails the office. The office runs on an EHR — in this case athenaOne — which is where the real schedule, the real chart, and the real patient record live. So somebody reads the email, opens the EHR, and retypes what the patient already typed. Then they call to confirm the time the patient thought they’d already chosen.
Every retype is a delay, and every delay is a chance for the patient to book somewhere else. Worse, it creates two versions of the truth: an inbox full of half-leads and a schedule that doesn’t know about them. Nobody can answer “how many people asked to book last month?” because the answer is split across two systems and one person’s memory.
What we built: intake that writes into the system of record instead of into an inbox. The website’s job is to collect the minimum needed to get someone onto the schedule, hand it to the EHR, and confirm back to the patient in the same session — so the appointment they think they made is the appointment that exists.
Being honest about the shape of this work: the hard part is never the form. It’s the negotiation between what the EHR’s interfaces permit, what the practice’s workflow actually is once you watch it for a week, and how little data you can get away with moving. We started from the last of those three and worked backward. Which leads directly to leak four.
Leak 3 — The service page that doesn’t exist
Clinics organize their websites the way the clinic thinks: Wellness Services and Aesthetic Services, each a dropdown holding eight treatments. Patients search the way patients think — the name of the drug they read about, the brand name of the device they saw on Instagram, the treatment plus their neighborhood, or “how much does it cost.”
When eight treatments share one page, that page ranks for none of them and converts poorly for all of them. The visitor who searched a specific treatment lands on a menu and has to start the work over.
What we built: a page per treatment intent. What it is, who it’s a good fit for, who it isn’t, what happens at the first visit, what it costs to get started, and what it will not do. That last part converts better than the marketing copy it replaced — people arriving at a medical decision are looking for a reason to trust the practice, and admitting a treatment’s limits is one.
Provider credentials went onto those pages rather than being buried in About Us. Search engines reward it, and it’s what patients are actually checking: who is doing this to me, and what are they licensed to do. Then the local layer — one properly built location page, consistent name/address/phone everywhere it appears, and the review profile treated as an asset rather than an accident. For a clinic that draws from a fifteen-mile radius, that work outperforms publishing volume by a wide margin.
Leak 4 — The line you don’t cross
Every automation you attach to a medical practice eventually runs into protected health information. There’s a temptation to build the impressive version: a chatbot that discusses symptoms, an intake form that collects history, a marketing stack that knows what each lead is being treated for.
The safest system is the one that never holds the data in the first place.
- Website forms collect no patient information. Name, contact, treatment interest. The message field says so explicitly, because people will otherwise volunteer their medical history to a contact form.
- The receptionist doesn’t take symptoms, diagnoses, or medication lists. It takes who you are, how to reach you, and what you’re interested in. Anything clinical routes to a human.
- Anything that belongs in a chart goes to the EHR, which is built, audited, and covered for exactly that.
This is a scoping decision made with the practice, not legal advice — every clinic needs its own counsel on where its obligations start. But the engineering principle generalizes: a marketing stack should not be a place where PHI quietly accumulates. Ad platforms, form-submission emails, analytics tools, and CRM records are all places data goes to live for years, and none of them are where a patient’s health information should end up.
Deciding not to collect something is the cheapest compliance work available.
What actually shipped
The website, the intake path into the EHR, the AI receptionist, and local search — designed together, by one team accountable for all four.
That last part is the quiet reason it worked. When the site is one vendor, the phone is another, and the booking system is the EHR reseller’s problem, the leaks between them belong to nobody. Every clinic we’ve looked at loses more in those seams than on any single page.
Four questions to ask about your own clinic
You can run this audit yourself this week:
- Call your own clinic at 8pm on a Saturday. Listen to what a new patient hears. Then decide whether that’s the first impression you want to pay for with ads.
- Book an appointment on your website as a stranger. Count how many times your information gets retyped by a human before it lands on the schedule.
- Search your three most profitable treatments plus your city. Is there a page for each one — and are you on it?
- Find where patient information lands in your marketing stack. Then ask whether it needs to be there at all.
None of those are marketing questions. That’s the point: growth in a clinic is usually an operations problem wearing a marketing costume.
We build the front desk — website, booking, phone, and local search — for med spas and wellness clinics, and the AI receptionist that answers when the room is busy. If your calendar has gaps your ad spend can’t explain, tell us about it and we’ll tell you honestly which of the four leaks you have.