Insights
Notes on AI, documentation, and staying independent.
Healthcare has no shortage of AI enthusiasm and no shortage of AI backlash. What it lacks is writing that takes both seriously. These pieces work from published evidence — including the studies that found less than the headlines promised — and try to be specific about what a small practice should actually expect.
All pieces
Twelve essays on where AI genuinely changes the work of a practice — and where the evidence says it doesn't, yet.
The fax is still the front door
Hospitals sent care summaries by fax as often in 2025 as they did in 2018. Half of specialty referrals never complete. The work after the page lands is the part AI can actually do.
Read → BillingThe claim should already exist when the visit ends
Most billing AI is bolted onto a reconstructed claim. The useful version never leaves the chart: codes attested in the visit, an 837 assembled from that record.
Read → Prior authThe denial is prevented in the room
An X12 278 has no field for the note. Prior authorization is won by documenting the requirement while the patient is still in front of you.
Read → DocumentationThe documentation burden, measured
Everyone agrees the EHR eats the day. Far fewer people have looked at what the logs actually say — or at what ambient AI has been measured to give back.
Read → SafetyWhat “human in the loop” actually means
It's the most over-used phrase in clinical AI and one of the few with a precise legal meaning. Here is the standard, and the research that explains why it exists.
Read → CodingWhy E/M levels are hard, and what AI should not do about it
Expert coders disagree with each other on more than half of office encounters. That is the context every claim about “AI coding accuracy” should be read in.
Read → Revenue cycleMost denials are decided before the visit starts
Practice leaders name denials the biggest leak in the revenue cycle. The causes cluster at registration, eligibility and authorisation — none of which happen in billing.
Read → Prior authThirteen hours a week, and what is finally changing
Forty authorisations per physician per week, mostly by phone. New federal rules and a set of insurer promises land between now and 2027. Here is what is real.
Read → ProcurementNine questions to ask before you give a vendor your patients' data
Most AI procurement conversations never get past accuracy claims. These are the questions where the answers are checkable and the consequences are yours.
Read → OperationsThe front desk is the most automatable hour of the day
Check-in is where the claim is built, the schedule holds or collapses, and the patient forms their opinion of the practice. It is also mostly retyping.
Read → IndependenceEighteen percent
That is the share of American physicians still working in physician-owned settings. The arithmetic that produced it is not mysterious, and not all of it is fixed.
Read → Product thinkingWhat an AI-native EMR looks like on a Tuesday
Not the demo version. A walk through one ordinary visit, and the places where the design either earns its keep or quietly stops being used.
Read →Reading about it is one thing
ChartVoyant runs a full live demo on synthetic patients. Talk a visit through and judge the documentation, the coding, and the approval gates for yourself.