Every time I bring up AI in a medical practice, someone says the same thing.
“Patients don’t want to talk to a robot.”
I actually agree. I just think most practices are solving the wrong problem.
When an owner tells me they’re doing AI, they usually mean a chatbot on the website or something writing their captions. Patient-facing, visible, easy to point at in a meeting.
Meanwhile the leak is sitting at the front desk, and it has been for years.
One thing I see over and over
Nobody leaves a practice because the appointment reminder was automated. They leave because they were told someone would call them back about the plan they discussed, and no one did.
That’s never a decision anyone made. It’s what happens when the person handling follow-up is also handling phones, check-ins, promo texts, and the inventory count. Follow-up is the only task on that list with no immediate consequence, so it’s the one that gets dropped. I have sat with front desk associates who genuinely believed follow-up was happening, because they had every intention of doing it that afternoon.
Six months later the owner is looking at soft consult conversion and asking whether pricing is too high. In most of the practices I’ve been in, price wasn’t the reason.
The Three Buckets
Before you buy anything, list every recurring task in your practice and put each one in a bucket. It takes an afternoon, and it’s the whole framework.
Bucket one. Automate now. No patient health information involved.
Appointment confirmations and reminders. Waitlist backfill. Recall and rebooking. Missed-call text-back. First response and qualification on a new lead. Promo outreach to a defined segment. Review requests. Pre-consult intake collection. Post-treatment check-ins. Inventory reorder triggers. Training and credentialing reminders.
Bucket two. Automate only under a signed business associate agreement.
Anything that records, transmits, summarizes, or writes back clinical detail. The in-room scribe lives here, along with anything that touches the chart.
This is the line practices cross without noticing. A provider dictating consult notes into ChatGPT has created a compliance exposure with no upside. Consumer tools don’t offer a BAA, and there’s no version of that risk that pays you back.
Bucket three. Never automate.
The consult. The treatment plan conversation. Clinical judgment. The call to a patient who’s worried about how something is healing. Any pricing conversation where the honest answer depends on the plan.
If you remember nothing else from this, remember the buckets. Most bad AI purchases in this industry are bucket two or three tools sold as bucket one.
Where I’d start
If I walked into a practice tomorrow and had to install this, this is the order I’d do it in.
Missed-call text-back, first. It’s the cheapest thing on the list and the leak is the most obvious. You’ll know pretty quickly if it’s working. That matters more than people expect, because your first rollout either earns you credibility with a skeptical team or costs you all of it, and the second rollout is much harder without it. Podium’s Avery and Weave both do this well.
Recall and rebooking, second. This is dormant revenue and it asks nothing of your team. Most practices are already paying for this inside Boulevard, Zenoti, or Phorest and have never turned it on. When I ask, the answer is usually that whoever set the system up left.
Consult follow-up, third. This is the one I’d spend the most time on, because it’s the one that usually improves consult conversion. It’s third and not first because the messages have to be right. A generic “just checking in” three days after a consult performs about the same as sending nothing.
Lead response and intake, fourth. GoHighLevel or MarketStorm for routing and nurture. Opaline for pre-consult intake and qualification, so your provider walks into the room already knowing what the patient came in for instead of spending the first six minutes finding out.
The scribe, last. On purpose.
The scribe
This is the change providers push back on hardest and end up appreciating fastest.
Patients notice when you’re looking at the screen instead of looking at them. That’s happening during the ten minutes that decide whether the plan gets accepted. And the note written afterward from memory, between rooms, is worse than what someone actually listening would have captured. Every provider I’ve worked with underestimates how much they’re reconstructing at the end of the day.
Start with your EMR. Ask whether they offer a HIPAA compliant AI scribe before you look at anything standalone. Most of the major platforms have released one or have it coming, and using theirs means the note lands in the chart without a second integration to maintain or a separate BAA to negotiate.
If your EMR has nothing, the rule doesn’t change. Signed BAA, reviewed by someone qualified to review it, before a single consult gets recorded. Consent handled the way your state requires. If a vendor calls itself HIPAA compliant but won’t execute a BAA, you’re done talking.
I put this last because it requires providers to change how they work and it requires a legal review. It will fail if it lands on a team already tired of the last three rollouts.
The part vendors skip
None of these tools build a system on their own.
Someone in your practice has to define what a confirmation says, what happens when a patient doesn’t respond, what qualifies a lead, and what a good follow-up note actually contains. That’s weeks of work. It’s also why most implementations stall in month two, right after the excitement of the demo wears off and before anyone has written the messages.
Do it once and it runs without you. But it does not work in the other order, no matter what the demo suggested.
What I’d measure
Take a baseline this week, before you install anything.
No-show rate. Speed to first response on a new lead, in minutes, not in whether it felt fast. Percentage of consults with a documented follow-up contact within 72 hours, which is usually the worst number in the practice and the one nobody is tracking. Front desk hours spent with patients versus on administrative work. Consult conversion, with pricing held flat, so you know what actually moved it.
You’ll want somewhere to see all of this without rebuilding a spreadsheet every month. Illume is what I’d use.
If those numbers don’t move in a quarter, the tool isn’t the problem. The definitions behind it were never finished.
AI shouldn’t replace the parts of your practice that make patients trust you.
It should remove the work that keeps your team from being present.
Automate the follow-through. Keep the judgment.
theaudreyaesthetic.com