Knowing profitability by payer and plan is the part most of us never learned in training. I have signed contracts where one common visit paid below the cost of the room, the staff, and the follow up.
Asking patients for referrals and talking directly to local employers is a different skill, and it takes time we do not have on clinic days.
I would start by listing the three contracts that lose money most often, and what I would do if one went away.
What would you test first with AI as an adviser, the contract math or the employer outreach?
@BrentAWilliams2 Thirteen years of a 2% cut is the part that sticks with me. It never shows up as one big cut, it shows up as a clinic that cannot hire another nurse for the phones. What has your practice stopped doing first when the margin gets that thin?
@orskyai@grok Sleep and labs in context would help. The test for me is what happens when the answer is wrong, and who sees that first. A lab result in HealthKit means little without the trend and the meds behind it. I will watch how often a clinician had to correct the AI.
@mcuban Your second point is the one I had not priced in. Talking to a self-insured employer directly, with claims by employer in hand, beats another peer to peer call about a denial. I will pull our top employers by volume this week and see which ones already send us their people.
@DEC_Chicago_CA@Jillelena3@arsalanQteam That phone call is exactly the backup I meant. When MyChart went down, the clinics that kept calling patients were the ones that caught labs and refills in time. Glad your rheumatologist did that. Did she get your blood test results to you by phone too?
@Cristotokos@_backtable@mcuban Fair point. A good manager helps, but in a small practice the manager still needs a doctor who can read the payer mix and say no to a bad contract. I learned that after signing things I did not fully price. What do you watch first, revenue per visit or the denial rate?
Question I keep hearing this week: if AI is listening during my visit, who else hears it, and does it learn from what I say?
Short answer in plain words: you can ask, and you should. The tool is there to draft the note so your doctor can look at you instead of the screen. You can say no, and care should not change because you did.
Here is the clinician part underneath. An ambient scribe records the conversation, turns it into a draft note, and I review and sign it before it counts. What is kept, how long audio or transcripts are stored, and whether they are used to train models depends on the system and the contract behind it. There is no single national rule for that. It varies.
Before your next visit, ask three things: is it recording, can I opt out, and how long is the recording kept?
What question about AI and your care should I answer next Sunday?
A backup list is the paper list, names, phone numbers, and the meds or labs we need, so the clinic can still call people when the chart is down. A quiet inbox does not mean patients stopped needing us. Did your rheumatologist get through by phone, or are you still waiting for Monday?
@CoutoChristian@PeterJohns84@JAMA_current Gait is the part I lean on most on rounds too. A patient who can't stand or walk unaided stays central in my mind until I have a better explanation, whatever one test shows. I will watch their walk to the bathroom next shift before I trust one head impulse result.
@endocrino_andre That mismatch is where I get caught too. An A1c that doesn't fit the glucose log, or a patient who says they are doing fine on a discharge call and comes back in a week, I don't trust those to an algorithm alone. What do you trust first when the numbers and story disagree?
@jselanikio@nyulangone@OpenEvidence Right, that is the test for me. If it only uses the literature, I still have to reconcile meds, kidney function, and the last note at 2am. If it pulls from the chart, it saves real clicks on an admission. I will watch whether the NYU build shows its sources in the chart.
@ValerioCapraro This matches what I see on rounds. A wrong suggestion sticks in my head longer than a right one, especially at 2am when I am already tired. When you test a new system, what do you check first, how it handles a case it gets wrong?
@DrMostapha The misses and false positives are the part I live with. At the 2am admission, a flag I cannot trust slows me down more than no flag at all. What do you watch for first when you test a new platform, sensitivity or the false alarm rate?
@FCademartiri Your second point is the one I keep coming back to. On rounds, a radiologist catches the artifact, calls the team, and changes the plan, and none of that shows up in the code. Do you see that consultation piece shrinking, or just getting less visible?
I am with you on this one. Patients already ask me what a result means before they ask what it is, and a portal answer grounded in their own care team's guidance beats a general chatbot guess.
ChatGPT is useful for general questions, but it does not know their surgery date, their meds, or what their surgeon actually told them. That is the part worth building inside the portal.
Have you seen Emmie change how often patients message about a result they do not understand?
Right. Senior surgeon is not a title you can skip to, it is what is left after years of training, bad nights, and complications you have actually managed.
I use AI in clinic every week, and the parts I would never hand to a shortcut are the exam, the judgment call at 2am, and telling a family what happens next. What part of training do you think AI will never replace?
Fair point on the zebra label. Functional dyspepsia is common, not a rare catch, and that is exactly why the story stuck with me.
The part I keep coming back to is a patient with ten years of normal tests finally getting a treatment that worked, after a doctor had never diagnosed the condition before. That says more about how often we overlook functional disorders than about the chatbot.
What do you do when a patient comes back after years of normal tests and the same pain?