This resonates deeply, especially the distinction between routing every signal through the physician versus reserving physician cognition for the decisions that actually require it.
Feels like many of the current economic and workflow failures stem from conflating those two things.
This framing is very close to the real issue: primary care is not just understaffed; it is under-architected.
More clinicians, more grit, or more documentation automation may help temporarily, but the deeper failure is that patients, clinicians, staff, referrals, follow-ups, and unresolved context still move through systems that were never designed to preserve shared operational awareness.
AI can help, but only if it becomes part of a better care operating model rather than another layer placed on top of the same broken handoffs.
There aren’t many that would disagree with re-architecting the solution. The problem is medicine’s payment system also needs to be rearchitected. Unlike free-market systems where changes can be made logically, medicine is mired in a 3rd party-payment structure that gums up every step. Laws at the state and national level need to be changed in order for non-physicians to complete a process for the physician in order for the transaction to occur.
So, how do we change the Social Security Act? The Accountable Care Act and the hundreds of other acts signed into law across the nation to allow this to happen?
The decision to step away and move to Direct Primary Care or Concierge Care isn’t the right solution and isn’t scaleable as we’re well over a half-million physicians short to even contemplate that. It does solve many of the problems with primary care but only for the few.
What will it take to overhaul the 3rd party payment system?
Well said David. There is really only one way in history to disrupt a system like you highlight above…find a way to disrupt them where it actually hurts. In their pocketbooks.
When we went through the quality/six-sigma/lean process in manufacturing in the early 90s, we had a rough equivalent to this problem. In the form of fix/repair/rework routines on the shop floor. 1000s of little defects that needed engineers to ok a fix. We kept arguing we have no time for root cause. The defects only appeared on 5% of the time. We measured the cost of repair at that moment. We left out the need of infrastructure to fix the defect. And failed to recognize that while we thought it was 5%, it was the 5% we saw. We ignored fixing the defects that got past us.
Your layout of the problem gave me a bit of PTSD from the days of fighting defects. Every time I message my provider, with what I know is a back office issue, I wonder, "Is staff going to pick this up or is the doctor going to disposition?"
I can see the doctors advising Epic "No, we need to look at each message to be sure there is not a care issue". While never really understandign the system opportunity.
In aircrat, we just made whole classes of issues disappear. We, for example, figured out how to make more precise parts with low variability. That then allowed us to avoid $700M on new tooling. We had avoided the solutions because originally we thought the cost was 5% of the time we saw the defects. We had missed that it was on the ready labor, defects we missed, etc. Once we understood we could never capture the full cost, but it was huge, it changed our approach. And our cost-benefit calculations.
The real cost in your inbox is the new dialysis center. 80% of the hospital's rooms. The memory extend care unit. The decade to train a doctor. No one wants to change it because it removes their immediate revenue.
SpaceX's reusable rocket sells fewer rockets. It delivers more satellites to space. A healthcare solution that fixes PCP, delivers more health span. Maybe JP Morgan should be the compounder who fixes PCP.
Are the Epic shareholders happy that Epic automates a crappy process? Is the ACA happy they forced all providers to adopt EMRs?
Fixing food ends a lot of the inbox.
What issues should never even land at the PCP system doorstep? If we had no AI, how would you solve this?
Carter, you handed me the better frame in three paragraphs. The manufacturing analogy is the missing layer in the post.
I measured the cost of the inbox the way the shop floor measured the cost of repair. At the moment of touch. I left out the infrastructure cost, the defects we never see, and the ready labor. The 5% we measure is the 5% we let ourselves measure. The other 95% is the dialysis center, the cardiology fellowship, the malpractice premium, the burned-out internist who left clinic at 45, and the patient who stopped messaging because last time it took six days to get a yes.
Pass-the-buck is the visible cost. The invisible cost is the entire downstream healthcare economy that exists because pass-the-buck failed upstream.
The dialysis center line is the one that should be on the slide. Every unmanaged metabolic message in the inbox eventually becomes a bed. The legacy hospital P&L depends on those beds. The Epic incentive depends on those beds. The specialist incentive depends on those beds. No one inside the System B revenue model is asked to fix the upstream, because the upstream is the product line.
Epic shareholders are perfectly happy. Epic automated the workflow that protects the downstream business model. The ACA mandated EMR adoption without mandating that EMRs be designed for resolution rather than documentation. The result is software that perfects the wrong workflow at scale.
This is the part of System C that does not market well to incumbents. SpaceX sold fewer rockets and delivered more satellites. A working PCP architecture sells fewer admissions and delivers more health span. The buyer is not the hospital. The buyer is whoever holds the actuarial risk. Self-insured employers. Medicare Advantage payers. A JP Morgan-style compounder that buys the upstream and clips the downstream avoidance.
Your two questions.
What should never land at the PCP system doorstep?
Refills. Prior auths. Forms. Logistics. Insurance back-and-forth. Pharmacy back-and-forth. Records requests. Normal results. Standing care-gap closures. Population health nudges. Specialist scheduling. Any portal message whose root cause is “the system did not do the thing the system was supposed to do.” Roughly 70 to 80 percent of inbox volume in a legacy clinic.
What should land at the physician’s doorstep is what only a physician can resolve. Novel clinical presentations. Multi-condition decision points. Hard diagnoses. Care planning conversations. The judgment call no protocol covers.
If we had no AI, how would you solve this?
The same way. The architecture is the achievement. AI is the substrate that makes it economically feasible at primary care economics.
Direct primary care solved most of this without AI by paying for a different staffing ratio. Kaiser at its best solved most of this without AI by building defined escalation matrices and standing orders. Concierge medicine solved most of this without AI by charging $5,000 a year. The architecture is well understood by anyone who has worked inside a clinic running the resolution-layer model.
What AI changes is the unit economics. AI takes the cost of right-layer resolution from concierge-tier PMPM to primary-care-tier PMPM. That is the move. Not invention. Affordability.
The food question is the deepest one. Fixing food ends a lot of the inbox. It also ends a lot of the dialysis center. Which is exactly why the legacy system has no incentive to fix it.
We should probably write a longer post about this, and cross post
I was wandering around the idea of looking at many for the billing codes, incidence rate, and related infrastructure to represent the hidden cost of each code. Essentally, if you treat, but never diagnose, what costs are you "Buying" into the patient. Then use that rough model to help systems people re-engineer. In quality, we gave up on that because every time we did, the costs change was massive.
We are thinking this way in the "Healthy Harvest" exercise we started - "how could 50B into ag, eliminate 500B a year in HC." I am thinking we push it on GitHub. Build a community of models around the issue to help people in the system see the through line.
This resonates deeply, especially the distinction between routing every signal through the physician versus reserving physician cognition for the decisions that actually require it.
Feels like many of the current economic and workflow failures stem from conflating those two things.
Big time. And the reason it has always been that way is the financial model. Everything flows through the billing provider.
This framing is very close to the real issue: primary care is not just understaffed; it is under-architected.
More clinicians, more grit, or more documentation automation may help temporarily, but the deeper failure is that patients, clinicians, staff, referrals, follow-ups, and unresolved context still move through systems that were never designed to preserve shared operational awareness.
AI can help, but only if it becomes part of a better care operating model rather than another layer placed on top of the same broken handoffs.
There aren’t many that would disagree with re-architecting the solution. The problem is medicine’s payment system also needs to be rearchitected. Unlike free-market systems where changes can be made logically, medicine is mired in a 3rd party-payment structure that gums up every step. Laws at the state and national level need to be changed in order for non-physicians to complete a process for the physician in order for the transaction to occur.
So, how do we change the Social Security Act? The Accountable Care Act and the hundreds of other acts signed into law across the nation to allow this to happen?
The decision to step away and move to Direct Primary Care or Concierge Care isn’t the right solution and isn’t scaleable as we’re well over a half-million physicians short to even contemplate that. It does solve many of the problems with primary care but only for the few.
What will it take to overhaul the 3rd party payment system?
Well said David. There is really only one way in history to disrupt a system like you highlight above…find a way to disrupt them where it actually hurts. In their pocketbooks.
When we went through the quality/six-sigma/lean process in manufacturing in the early 90s, we had a rough equivalent to this problem. In the form of fix/repair/rework routines on the shop floor. 1000s of little defects that needed engineers to ok a fix. We kept arguing we have no time for root cause. The defects only appeared on 5% of the time. We measured the cost of repair at that moment. We left out the need of infrastructure to fix the defect. And failed to recognize that while we thought it was 5%, it was the 5% we saw. We ignored fixing the defects that got past us.
Your layout of the problem gave me a bit of PTSD from the days of fighting defects. Every time I message my provider, with what I know is a back office issue, I wonder, "Is staff going to pick this up or is the doctor going to disposition?"
I can see the doctors advising Epic "No, we need to look at each message to be sure there is not a care issue". While never really understandign the system opportunity.
In aircrat, we just made whole classes of issues disappear. We, for example, figured out how to make more precise parts with low variability. That then allowed us to avoid $700M on new tooling. We had avoided the solutions because originally we thought the cost was 5% of the time we saw the defects. We had missed that it was on the ready labor, defects we missed, etc. Once we understood we could never capture the full cost, but it was huge, it changed our approach. And our cost-benefit calculations.
The real cost in your inbox is the new dialysis center. 80% of the hospital's rooms. The memory extend care unit. The decade to train a doctor. No one wants to change it because it removes their immediate revenue.
SpaceX's reusable rocket sells fewer rockets. It delivers more satellites to space. A healthcare solution that fixes PCP, delivers more health span. Maybe JP Morgan should be the compounder who fixes PCP.
Are the Epic shareholders happy that Epic automates a crappy process? Is the ACA happy they forced all providers to adopt EMRs?
Fixing food ends a lot of the inbox.
What issues should never even land at the PCP system doorstep? If we had no AI, how would you solve this?
Carter, you handed me the better frame in three paragraphs. The manufacturing analogy is the missing layer in the post.
I measured the cost of the inbox the way the shop floor measured the cost of repair. At the moment of touch. I left out the infrastructure cost, the defects we never see, and the ready labor. The 5% we measure is the 5% we let ourselves measure. The other 95% is the dialysis center, the cardiology fellowship, the malpractice premium, the burned-out internist who left clinic at 45, and the patient who stopped messaging because last time it took six days to get a yes.
Pass-the-buck is the visible cost. The invisible cost is the entire downstream healthcare economy that exists because pass-the-buck failed upstream.
The dialysis center line is the one that should be on the slide. Every unmanaged metabolic message in the inbox eventually becomes a bed. The legacy hospital P&L depends on those beds. The Epic incentive depends on those beds. The specialist incentive depends on those beds. No one inside the System B revenue model is asked to fix the upstream, because the upstream is the product line.
Epic shareholders are perfectly happy. Epic automated the workflow that protects the downstream business model. The ACA mandated EMR adoption without mandating that EMRs be designed for resolution rather than documentation. The result is software that perfects the wrong workflow at scale.
This is the part of System C that does not market well to incumbents. SpaceX sold fewer rockets and delivered more satellites. A working PCP architecture sells fewer admissions and delivers more health span. The buyer is not the hospital. The buyer is whoever holds the actuarial risk. Self-insured employers. Medicare Advantage payers. A JP Morgan-style compounder that buys the upstream and clips the downstream avoidance.
Your two questions.
What should never land at the PCP system doorstep?
Refills. Prior auths. Forms. Logistics. Insurance back-and-forth. Pharmacy back-and-forth. Records requests. Normal results. Standing care-gap closures. Population health nudges. Specialist scheduling. Any portal message whose root cause is “the system did not do the thing the system was supposed to do.” Roughly 70 to 80 percent of inbox volume in a legacy clinic.
What should land at the physician’s doorstep is what only a physician can resolve. Novel clinical presentations. Multi-condition decision points. Hard diagnoses. Care planning conversations. The judgment call no protocol covers.
If we had no AI, how would you solve this?
The same way. The architecture is the achievement. AI is the substrate that makes it economically feasible at primary care economics.
Direct primary care solved most of this without AI by paying for a different staffing ratio. Kaiser at its best solved most of this without AI by building defined escalation matrices and standing orders. Concierge medicine solved most of this without AI by charging $5,000 a year. The architecture is well understood by anyone who has worked inside a clinic running the resolution-layer model.
What AI changes is the unit economics. AI takes the cost of right-layer resolution from concierge-tier PMPM to primary-care-tier PMPM. That is the move. Not invention. Affordability.
The food question is the deepest one. Fixing food ends a lot of the inbox. It also ends a lot of the dialysis center. Which is exactly why the legacy system has no incentive to fix it.
We should probably write a longer post about this, and cross post
I was wandering around the idea of looking at many for the billing codes, incidence rate, and related infrastructure to represent the hidden cost of each code. Essentally, if you treat, but never diagnose, what costs are you "Buying" into the patient. Then use that rough model to help systems people re-engineer. In quality, we gave up on that because every time we did, the costs change was massive.
We are thinking this way in the "Healthy Harvest" exercise we started - "how could 50B into ag, eliminate 500B a year in HC." I am thinking we push it on GitHub. Build a community of models around the issue to help people in the system see the through line.
Love this idea!