Insights
Short notes for practice owners, administrators and billing managers, written to be read on a phone between patients. Each one is a LinkedIn post, with its sources.
Follow TH Intelligence on LinkedIn for new ones.
video · October 22, 2026
Not every problem in your practice needs AI.
Before a chart review, medical assistants look up history notes, recent labs and which quality measures are already covered. I built an RPA bot (robotic process automation: software that clicks through screens the way a person does). It opens each upcoming patient in the EHR, reads that history, and matches it to the measures. The hard part wasn't AI. It was matching every patient to the right chart: 1,405 of 1,409 matched, at about 160 patients an hour. Use AI where a task needs judgment. For the clicking, an RPA bot is often the better tool.
video · October 22, 2026
Most AI coding tools fail one simple test. Can they show their evidence?
I built a risk-adjustment gap engine for our practice. It flags conditions that haven't been billed this year from plan claims, our billing and the EHR problem list. The first version matched on approximate names, and most of its problem-list suggestions were wrong. What fixed it: exact codes, the evidence behind every suggestion, and a practice manager who checked real rows by hand and turned each false alarm into a rule. If a vendor is selling you an AI coding tool, ask to see the evidence behind its suggestions.
video · October 12, 2026
Your AI scribe needs one thing before it starts listening.
Patients in California have filed proposed class actions alleging that health systems recorded visits with an AI scribe without their consent. If your practice uses an ambient scribe, or is about to, the consent step matters as much as the tool: get the patient's affirmative consent, add AI to your notice of privacy practices, and have a process for when a patient says no. Not legal advice; check with your counsel.
Start with 20 minutes.
Bring one process that eats your team’s day. We’ll talk through what to measure first, and whether automation or AI fits it at all.