The Real Cost of AI for Clinics: Knowing What to Build, and How

For a long time, software was something you bought.

You picked a platform, paid the monthly fee, and bent your clinic to fit whatever the vendor decided you needed. Half the features you never touched. The one report you actually wanted didn't exist. That was the deal, and everyone took it, because building your own meant hiring engineers and spending six figures you didn't have.

AI changed that, though not in the way the hype says. It's worth being precise about what actually changed and what didn't.

Start with what didn't. You are not going to replace your practice management system, the Jane or Cliniko or IntakeQ you run the clinic on. Good luck standing up a booking and records platform that stays online, processes payments, survives an audit, and protects patient data, then keeping it that way at two in the morning when it goes down. That is not the opportunity. Anyone selling you a rebuild of your core platform is selling you risk you were smart to offload in the first place. Your practice management system is the commodity layer. Records, scheduling, payments, storage, security, uptime. You rent that forever, and you should.

What changed is everything that sits on top of it.

Your platform becomes the system of record, the clean source of truth. The clinic-specific layer gets built on top, the view of your business no vendor will ever ship because it exists only inside your clinic. Traditional SaaS can't follow you up there. Its whole business is selling one product to ten thousand clinics. Customized at scale is the thing it structurally cannot do.

Here's where founders get the math wrong. AI made the code cheap. It did not make the software cheap.

Custom doesn't replace your practice management subscription. It sits on top of it. You keep paying for the system you run on, and you invest in the layer above it, so this is an addition to the budget, and additions have to earn their place. The upfront number is real. Maintenance after that runs lower, but lower is not zero, and it arrives in lumps: when the export format changes, when your needs move, when something drifts. The honest shape is a trade. A predictable monthly fee forever, in exchange for money upfront and smaller costs here and there. Which is exactly why saving money is the wrong reason to do any of this. Do it for ROI. When I built an economic model for my own clinic, it surfaced about $43,000 in recoverable revenue at zero incremental cost. That's the frame every tool gets held to now: what did it return.

Now the part the hype skips.

When building software was expensive, the barrier was code. You needed people who could write it. AI didn't remove that barrier so much as lift it up and show you the two sitting underneath. The first is knowing what your clinic actually needs, the right question to build against. The second is knowing how to build it without hurting yourself.

Take a simple patient follow-up tool. AI will write it, and it will work on your laptop. What AI will not do is stop and tell you where the patient data goes while that tool runs. Send identifiable data off to a server without the right setup and you've created an exposure you didn't know existed. The decision to keep the whole thing running locally, so patient data never leaves the machine, is what keeps you safe, and AI won't make that call unless you already knew to ask for it. Same with configuring the data properly. Same with knowing what your export does and doesn't contain. AI collapsed the typing. It exposed the expertise that was always underneath the typing.

That's the thing founders find out the hard way. "Just build it" turns out to be a stack of decisions you didn't know were there. The learning curve doesn't announce itself until you're standing in the middle of it.

UX is the same story. Paying attention to the person using the tool is only the entry fee. The skill is designing a flow a slammed front desk will actually adopt, then reworking it three or four times until they do, because they won't the first time. That's product development, and it's a craft with reps behind it. The failure mode here is quiet. A tool that works fine and nobody opens. That one is worse, because you paid for it and got nothing back.

None of this makes it an enterprise game. Deployed right, it runs the other way. For the first time, a small clinic can run software as sharp and as tailored as a big one's. The tools that used to belong to whoever could afford a dev team now belong to whoever is willing to make the investment. For the independent operator, this is the most level the playing field has ever been.

So should you build it yourself? Sometimes, genuinely, yes. If it's low stakes, touches no patient data, and costs you nothing when it's clunky, a quick internal calculator, a one-time look at a number, build it and learn something. The calculus flips the moment the tool is load-bearing, or it touches PHI, or other people have to adopt it. That's where the invisible work lives. That's where doing it yourself stops being thrift and starts getting expensive.

Which is the real case for a partner. Your time is part of it. Every hour spent wiring this together is an hour you're not treating patients or leading your team, and the hallmark of a good founder is delegation, knowing what to hand off. The deeper reason is that the work left over is skilled work that looks smaller than it is, and a few of the ways it bites you cost real money to learn firsthand. What you want is someone who knows your vertical cold, who understands a clinic and not just code, who jives with your team, cares about your mission, and is genuinely good to work with.

All of it takes a mindshift. Software stops being a monthly bill you tolerate, or tech bought for tech's sake, and becomes an investment you make on purpose, judged on what it returns. The clinics that make that shift run more responsive, more profitable operations, and the gap compounds year over year. AI made that possible for the first time. Building it well is still the work. Eventually that's what separates the winners, not who had access to AI, because everyone will, but who knew what to build with it, and had the sense to build it right.

Written by Luke Bujarski. Founder, LUFT