AI for healthcare operators, built by an M.D.

For physician groups, dental and specialty practices, surgery centers and the organizations behind them. We find the constraint that is actually gating the practice, build AI to fix it (in-house where that's cheaper, the right vendor where it's better), and stay accountable for whether it worked.

The evidence is from your own field

The model was never the problem.

In a randomized trial published in JAMA Network Open in 2024, physicians given a top model did not outperform physicians working without it, even though the model on its own outperformed both groups. In a 2025 follow-up in NEJM AI, twenty hours of AI training did not protect physicians from following bad AI advice.

The model was already better than the experts. What failed was the handoff between a capable clinician and a capable tool, and better models widen that gap rather than closing it, because the more often a tool is right, the faster people stop checking it.

Those trials measured clinical diagnosis. Our work is on the operating side of a practice, so the transfer is an analogy rather than evidence. The structure still holds, though. Training did not fix it, so the fix has to be structural: the verification built into the workflow itself, where it happens whether or not anyone remembers to run it.

Where most practices are right now

Four situations that keep showing up.

You bought a tool, most often a scribe. Notes got a little faster. The phones, the front desk and the denials queue did not change, because the tool was never aimed at them.
Your staff are probably already using AI on their own. Most practices have no written AI policy and no approved tool, so nobody can say what patient information is going where.
Everyone has a guess about where the practice loses the most time and money, whether it is denials, prior authorization, the phones or intake. Nobody has put a number on it.
Your staff lose their week to payers instead of patients. Checking eligibility in one portal, submitting a prior authorization in another, tracking status in a third, then documenting it all in a record system that connects to none of them.

What links all four is a constraint nobody has named yet. Finding it comes before building anything.

In MGMA polls of medical group leaders, 56% reported neither a formal AI policy nor one in development (January 2026), denials and appeals were named the biggest revenue cycle leak at 48% (January 2026), limited tracking made it hard to set a productivity baseline for AI at all (May 2026), and 61% said their staff work in seven or more payer portals every week (March 2026). In an AMA survey of 1,000 physicians, practices averaged 39 prior authorizations per physician per week, about 13 hours of physician and staff time (published April 2025).

Where we'd look first

A few places practices lose money quietly.

Every practice is different, and the diagnosis decides where we start. Here's a sampling of areas we keep finding worth a look:

Payments that come back smaller than billed. Some payers pay a visit at a lower level than it was billed without flagging the change, so the remittance reads as paid. Comparing billed and paid codes on your own remittance files catches it the day the payment lands.
Contract rates nobody checks. Some contracts tie rates to a payer's own fee schedule rather than Medicare, and scheduled increases don't always get loaded. Holding every contract in one place, and checking what you were paid against what you agreed to, shows where the gap is.
AI tools bought before anyone vetted them. A short, vendor-neutral review before you sign (how the tool works, who maintains it, what the contract promises if it underperforms) costs far less than unwinding a tool later.
AI output nobody measures. A scribe or billing tool that's right most of the time still needs someone confirming that notes support the level billed. A sampled review with an error rate turns "it seems fine" into a number you can defend.
Denials that start at the front desk. Coverage and coordination-of-benefits checks before the visit stop a share of denials before a claim is ever filed.

If the thing slowing your practice down isn't on this list, that's common. Finding it is the first step of every engagement.

The five checks above, each with a test your staff can run this quarter on numbers the practice already holds, are written up in What should a medical practice check first with AI heading into 2027?

Patient data

Settled first, not last.

Anything that could touch protected health information starts with three questions before a line of the system is written: where the data lives, who can see it, and which agreements cover it. Where a workflow does not need patient data at all, it does not get any.

How an engagement runs

Three steps, in this order.

Constraint diagnosis, and a first win.

We find the decision that is actually gating the practice, and start with a check on your own files (often the remittances), so there's a dollar figure on the table within days. The most expensive mistake in this work is automating something that was never the bottleneck.

Architecture and build.

Designed before it is written, with patient data settled first and the verification sitting where a person still carries the liability.

Operation.

We run it with you and keep tuning it. Every workflow ships with its checks, a named owner on your side who approves changes, and a history of every change, all in a repository you own. Systems nobody owns are systems that quietly stop being true.

Engagements run as a fractional Head of AI seat or as a scoped build, depending on how much of the operation is in play.

Written down

Four questions worth answering first.

Can AI handle prior authorization for a medical practice? Partly, and the part it handles is the part that eats the hours. Where software does the work, and where a person still decides.

What is a fractional Head of AI, and when does a practice need one? A part-time senior operator who owns two calls: where AI gets pointed, and whether what it produces can be trusted before anyone acts on it.

Where does a medical practice find the money for AI in 2027? Mostly from money it already pays: billing and denial rework, prior authorization hours, front-office turnover, and tools bought to patch the EHR.

What should a medical practice check first with AI heading into 2027? Five checks on your own data, each with a test your staff can run this quarter, before any new AI gets planned.

Why an M.D. builds this.

I trained as a physician and never practiced, but I never stopped practicing differential diagnosis. Rule out what can kill before you treat what might heal. That is the same discipline as underwriting, and it is the one I bring to every system I build. It also means the clinical side of your operation is not a foreign language to me.

The rest of the judgment came from somewhere less comfortable: years running a business where my own capital sat on every decision, and where being wrong cost me money. Before that, I spent 10+ years directing engineering teams, since 2015, specifying the work and sending it back when it came back wrong. More on where that came from.

Most fractional Heads of AI write the strategy and hand it over. I own the strategy, build the system, and stay accountable for whether it worked.

Proof, on our own business first

Every number we publish comes from a business we own, with our own capital exposed to the result. We ran the method on ourselves before we offered it to anybody else.

Zero hands-on

Underwriting that took 45 to 60 minutes by hand now runs unattended in under 30, and landed at $87,992 against a hand-built $88,000 on a test deal

$6.5M+

Funded and realized in land deals, our own capital at title, 41% operating margins

$13,754

A year in total overhead, down from about $32.7K, after a full-time operations role was rebuilt as routines in roughly 48 hours

The full before and after, with the source behind each number, is on the Proof page.

Start with the practice name.

Tell us which practice or group, and before the first call we will look at what is public (your website, how a new patient books, what your reviews say) and bring a short read on where the constraint probably sits. If there is no fit, we will say so and point you toward someone better suited.

Tell us what's breaking