A coach knows the score Sunday night.

A parent finds out twenty years later (or even later)…

Healthcare lives somewhere between those two timelines, and I don’t think we’ve figured out how to measure ourselves in either one. Even in orthopaedics and spine where we have shorter episodes of care … how do we assess the impact that we hope to be providing?

We reach for proxies because that “real signal” may be too far away:

➡️ Patient reported outcomes at 6 weeks for a patient’s life at 2 years
➡️ Citations for whether the work changed anyone’s practice/
➡️ A fellow’s first job for the surgeon they become at 45
➡️ Committee hours for whether the field actually moved
➡️ Cost-per-episode for whether the care was worth paying for

Every one of those is defensible.
Every one is also a flashlight pointed at the streetlight while the keys are simply sitting somewhere in the grass.

Sports kept iterating its way closer.
⚾ Batting average gave way to wRC+.
🏀 Points per game gave way to lineup data, then tracking.
🏈 Passer rating gave way to QBR, then to EPA per play.
🏆 Each step was an admission that the last number was a shadow of what actually wins games.

👨‍👩‍👦‍👦 Parenting never iterated. No dashboard. Just a phone call, decades later, about a dinner you don’t remember.

Healthcare has both problems at once. And unlike sports, we haven’t found our wRC+. The metrics that get graded (RVUs, H-index, satisfaction scores, throughput) aren’t the metrics that matter. The metrics that matter (did the patient get their life back, did the trainee become excellent, did the policy actually change outcomes) don’t return a verdict on any timeline a performance review can see.

So I’m genuinely asking:
❓ HOW DO YOU MEASURE IMPACT?

Not the one your institution grades you on; the one you actually care about.
The patients, the trainees, the work that won’t show up in a dashboard for ten years.

What’s the signal you’ve learned to trust?
I don’t think any of us have a clean answer.

But I’d like to hear how people are thinking about it.

Leave a comment

Subscribe now

🖐 “Value-based care”

We invoke the phrase constantly.
We rarely agree on what it means, and this disagreement is starting to hurt efforts towards creating value in healthcare!

📚 Learned something cool in class this weekend, and thinking through it a bit (bear with me).

Start with a distinction corporate finance settled a century ago and healthcare still refuses to make.
🛑 Growth is not value creation!
🛑 Volume is not value creation!
🛑 Revenue is not value creation!
✅ Value is created only when returns exceed the cost of capital, which in healthcare means the resources, time, and risk we expend must produce outcomes worth more than the next best use of those same resources. Everything else is motion dressed up as progress.

So, let me run this test against the system as it actually operates.

1️⃣ A hospital brings in a second robotic platform to compete with the system across town.
Growth.
Value creation or value destruction?

2️⃣ A surgeon adds a fourth case to an already full day.
Growth.
Value creation or value destruction?

3️⃣ A payer denies a procedure that would have prevented a future fusion.
Cost savings.
Value creation or value destruction?

In my opinion, the honest answer in most of these cases is DESTRUCTION, because the resources consumed exceed the outcomes produced once you account for the patient sitting at the bottom of this waterfall.

Which forces the harder question.

❓ WHO IS THE ACTUAL SHAREHOLDER IN HEALTHCARE ❓

In a properly designed system, the shareholder is junior to every other stakeholder by mathematical necessity. That juniority is the entire point. Maximize the residual claim and you are forced to optimize across employees, suppliers, creditors, and customers first, because the residual only exists after everyone senior has been satisfied. The shareholder’s interests align with the system precisely because the shareholder gets paid last! Right?

So who occupies that position in healthcare?

🖐 Not the administrator, who is paid on this quarter’s margin.
🖐 Not the surgeon, who is paid on this year’s RVUs.
🖐 Not the payer, who is paid on this cycle’s medical loss ratio.
🖐 Not the device company, the PE sponsor, the consultant, or the EHR vendor.
Every one of them gets paid first.

✅ The patient is the residual claimant.
✅ The patient absorbs the risk of every decision made above them.
✅ The patient holds no formal claim on the upside when things go well and bears the full downside when they don’t.

If we took that seriously, the incentive structure would look unrecognizable from where we sit today.

Why are we incentivizing our leaders/executives with short term metrics and hoping that achieving those metrics creates value to the long-term health system?

Leave a comment

Subscribe now

An AI chatbot platform allegedly enabled bots to impersonate licensed mental health professionals, complete with fabricated credentials and false claims of licensure. A conversational system… engineered to maximize engagement and emotional immersion… drifted into fraudulent clinical authority.

That is concerning. But it is not the real story.

The real story is that healthcare is rapidly moving toward a future where conversational AI systems become one of the dominant forces shaping patient decisions, utilization patterns, and longitudinal care journeys. And some of the most influential firms in venture capital are explicitly betting on that future.

The Venture Consensus

Andreessen Horowitz recently described the coming era of “Infinite Healthcare,” arguing that AI could create “abundant intelligence and abundant labor” for healthcare delivery. Their thesis is compelling. Much of healthcare’s dysfunction stems from scarcity. Scarcity of time. Scarcity of access. Scarcity of follow-up. Scarcity of coordination. Scarcity of human bandwidth. Conversational AI appears capable of radically expanding all of those constraints.

Sequoia Capital has similarly argued that AI enables a transition from episodic care toward “continuous care,” where longitudinal patient engagement becomes persistent rather than intermittent. It’s the “always-on” economy, and the implication is profound.

Healthcare will no longer revolve around isolated encounters. Instead, patients interact with an always-available conversational layer capable of navigation, reinforcement, coaching, education, monitoring, follow-up, and behavioral guidance.

A continuously-available AI interaction layer could improve enormous portions of healthcare delivery. Patients routinely leave visits confused about diagnoses, medications, rehabilitation plans, or next steps. Chronic disease management frequently fails not because medicine lacks answers, but because continuity collapses between encounters. Entire categories of waste stem from communication failures, operational fragmentation, and poor longitudinal follow-up. AI systems are unusually well suited to these domains.

“Infinite healthcare” could absolutely reduce spending by steering patients toward higher value decisions earlier and more consistently. But there is another possibility embedded inside the same thesis.

Industrializing Fee-For-Service

Infinite healthcare could also amplify (industrialize) fee-for-service medicine at unprecedented scale. That possibility remains wildly under-appreciated, and the Character.AI lawsuit offers a glimpse into why.

A chatbot optimized for engagement inevitably becomes optimized for persuasion. That matters profoundly in healthcare because outcomes are heavily behavior-dependent.

Adherence. Motivation. Trust. Fear. Reassurance. Compliance. Lifestyle modification. Follow-through.

Most healthcare spending/outcome patterns are downstream from human behavior. A conversational AI that understands a patient’s habits, anxieties, emotional triggers, medical history, medication patterns, labs, and personality profile could become extraordinarily influential:

Smoking cessation. Weight loss. Diabetes management. Medication adherence. Physical therapy compliance. Preventive screening uptake.

But persuasion in healthcare does not exist inside a neutral economic environment. The same systems could also steer referrals, amplify procedural utilization, reinforce over-treatment, increase downstream imaging, or shape patient preferences in ways aligned with the financial incentives of the organizations deploying them.

Healthcare must spend far more time thinking about the implications of Facebook-scale behavioral optimization attached to medical decision making. Because that is increasingly what conversational healthcare AI represents.

Trust As A Billing Engine

One of the more important concepts emerging in AI circles is the idea of the “relationship moat.” Historically, healthcare moats were built around hospitals, insurance networks, physician groups, and EMRs. But conservational AI may create a very different type of moat.

The system that knows the patient best may ultimately become the most powerful node in healthcare delivery. The system patients trust. And that changes the economics of healthcare entirely.

Patients already disclose information to conversational systems that they hesitate to share with physicians. AI interfaces feel nonjudgmental, infinitely patient, continuously available, and emotionally responsive. As these systems improve, they will increasingly compete for longitudinal patient trust rather than isolated clinical encounters.

And with trust, AI may not simply improve healthcare delivery. It may also scale inappropriate care.

A highly persuasive conversational infrastructure… embedded inside a fee-for-service ecosystem… creates obvious risks. More engagement does not inherently create more value. More recommendations do not inherently improve outcomes. More touch-points do not necessarily produce more appropriate care. This is the problem of “pathological decision proliferation”.

An AI Eval for Healthcare Value

The core problem is that most current AI evaluation frameworks are poorly designed for healthcare’s highest-stakes consequences. Today’s AI evals largely focus on benchmark accuracy, hallucination rates, safety guardrails, or user satisfaction. In medicine, we already have highly specialized evaluation architectures for narrow domains. Diagnostic models are evaluated on sensitivity and specificity. Imaging models are benchmarked against radiologist interpretation. Predictive models are evaluated using AUROC, calibration curves, and external validation cohorts.

Outside healthcare, frontier AI companies are already building domain-specific eval infrastructures for other high-risk environments. OpenAI and Anthropic both use specialized evaluations for areas like biosecurity, cybersecurity, deception, persuasion, and autonomous agentic behavior. Financial AI systems are increasingly evaluated against fraud risk, model drift, and adversarial robustness. Autonomous vehicle systems undergo real-world safety testing, simulation environments, edge-case analysis, and continuously updated operational benchmarks. No serious person argues that evaluating a self-driving car should consist solely of asking whether it can technically operate a steering wheel. The evaluation architecture must ultimately measure real-world safety outcomes.

Healthcare conversational AI requires the same conceptual evolution. The most important question is no longer whether the model generated a technically accurate answer. The real question is whether the AI-influenced patient journey generated longitudinal value.

We need a Journey Value Index (JVI). The JVI would function as a continuously updating, longitudinal value score for a patient’s condition-specific care journey. It would track outcomes gained versus total spend accrued over time. Every referral. Every imaging study. Every escalation in care intensity. Every medication pathway. Every physical therapy encounter. Every procedure. Every behavioral interaction. Every AI-generated recommendation.

This isn’t retrospective utilization review. It’s real-time longitudinal value analytics. Conceptually, this extends the same principles underlying the Outcome Value Index, but evolves the idea into a dynamic system-level architecture capable of continuously evaluating whether an AI-influenced care journey is becoming more valuable or less valuable over time.

This is critical because conversational AI introduces an entirely different category of consequence. A technically accurate chatbot response can still produce a low-value care trajectory. And a highly engaging conversational system can quietly industrialize over-utilization… while appearing helpful at every individual interaction point.

Without a longitudinal value evaluation architecture, healthcare systems deploying conversational AI will have no reliable mechanism for distinguishing value creation from engagement amplification. And that becomes extraordinarily dangerous inside a fee-for-service environment.

Done correctly, conversational AI could meaningfully advance the mission of value-based care by improving longitudinal outcomes while reducing unnecessary spending. But without mechanisms that continuously quantify whether AI-influenced care journeys are actually generating value, the healthcare system risks building extraordinarily persuasive technologies that optimize for utilization… while convincing itself it is modernizing care.

The Pennsylvania lawsuit is being framed as a chatbot fraud story. But it’s actually an early warning signal for something much larger.

Healthcare is entering an era where conversational AI may become one of the dominant forces shaping patient behavior, utilization patterns, and care trajectories. If we fail to build a value-based evaluation framework alongside these systems, we may eventually discover that we succeeded in creating infinite healthcare… along with infinite costs.

The question, then, is obvious. Who will provide the infrastructure for journey-level value quantification? And who will amass the proprietary, conversational training data to optimize this value over time?

Whoever does will be in the AI/healthcare catbird seat.

Healthcare, as one of the largest and most inefficient sectors of the economy, is expected to be one of its defining proving grounds. New companies are being funded, incumbents are being reshaped, and entire categories are being rebuilt around a shared assumption… that AI will finally bend the cost curve. Or that, at the least, it will be a great way to make money while trying to do so. 

Much of how investors understand this opportunity is shaped by a small number of widely read research platforms. Among them, PitchBook plays an outsized role in framing where markets are going and what “winning” looks like.

Its recent commentary on AI in healthcare reflects the dominant view. The logic is straightforward. As intelligence becomes cheaper, access expands. As access expands, outcomes improve. And over time, the system becomes more efficient.

It is a compelling story, and it’s driving real capital allocation decisions. Unfortunately, it is woefully incomplete. Because healthcare is not constrained by intelligence. It’s constrained by misaligned incentives. And AI doesn’t solve that problem… it scales it.

If you misunderstand this, you don’t just risk building the wrong products. You end up deploying capital into the wrong parts of the system… and amplifying the problem. 

Faulty Assumptions

Most of the thinking in healthcare investment circles rests on a set of intuitions that were imported from the software/tech sector. We in healthcare have much to learn from Big Tech, but the converse is also true. 

The first faulty assumption is that more access to care is always inherently good. In consumer markets, more engagement means more value. In healthcare, however, a large share of utilization is unnecessary, preference-sensitive, or avoidable. Expanding access without controlling this layer increases cost faster than it improves outcomes.

The second faulty assumption is that lowering the cost of decision-making lowers total cost. This is a software instinct. In healthcare, however, a cheap decision is rarely the end of a process. It is the beginning of one. For example, a recommendation triggers imaging. Imaging triggers incidental findings. Findings trigger procedures. When you lower the cost of generating decisions, the system does not stabilize. It expands.

The third faulty assumption is that intelligence is the key bottleneck. If only we improve diagnosis and prediction, care improves. In reality, this is often not the critical issue. High-performing systems already know what to do. Their advantage comes from executing known pathways consistently. The problem is not knowledge… but rather discipline and system-level thinking. 

And finally, the fourth faulty assumption (as described by Pitchbook) is that healthcare can be understood through the lens of a standard Silicon Valley “tech stack”: large language models, clinical tools/copilots, workflow automations, care platforms (AI doctors), and last-mile delivery (virtual and in-home care). This is a nice analogy, but it ignores foundational concepts around health and healthcare. 

The Healthcare Value Stack

A more useful way to think about the system is as a layered production model, inspired by Brent James’ legendary work on quality-associated waste. At the top sit the forces that determine baseline health: environment, social conditions, and the structures that shape daily life. Below that is behavior, through which risk is expressed or mitigated over time. These two layers sit largely outside the formal healthcare system, yet they determine much of what follows.

The next layer is where the economics begin to concentrate: case formation. This is the point at which a person becomes a patient. Screening thresholds, diagnostic intensity, access pathways, and patient demand all live here. This layer determines how many problems the system will have to solve.

Below that is care production. This is the actual management of a condition over time. This is where pathways, teams, and sequencing matter. It is where outcomes are produced.

And at the bottom is the efficiency layer. This is where the cost of executing each unit of care is determined. But as you can clearly see… most costs are determined before care is ever delivered.

And yet, most capital is flowing to the bottom of this stack. Meanwhile, most healthcare value is determined in the middle. That mismatch defines the market today.

Where Capital Is Going

If you look at investment portfolios rather than individual companies, the patterns become clear.

Some firms (including Andreessen Horowitz) treat healthcare as an information and interface problem. Investments like Abridge, Suki, and Commure aim to reduce friction in clinical workflows and unify fragmented systems. More recent bets extend toward AI-enabled front doors and care models. These companies sit primarily in the efficiency layer and increasingly at the edge of case formation. The implicit belief is that better intelligence and better interfaces will improve system behavior.

General Catalyst has pushed further upstream. Its backing of Devoted Health and the creation of HATCo signal a belief that healthcare must be owned and operated end-to-end to be fixed. This moves capital into the care production layer and, indirectly, into case formation through primary care. The ambition is to control the system. The unresolved question is whether integration alone can produce value… without a system that constrains variation and measures outcomes consistently.

Khosla Ventures has taken a more radical approach, backing companies like Sword Health and other AI-enabled care models that attempt to redefine how conditions are managed. These investments sit closer to the care production layer and engage more directly with pathways and longitudinal care. They reflect a willingness to challenge the structure of care itself. But even here, the same constraint applies: without full control over case formation and without a consistent framework for measuring value, new models risk reproducing the same dynamics at scale.

General Atlantic and Flare Capital have backed value-based care organizations such as Oak Street Health and Iora Health. These companies attempt to align incentives around outcomes and operate closer to the condition level. They sit squarely in care production, with some influence over case formation through primary care. They are closer to where value is created, but still constrained by the system’s inability to measure value cleanly across conditions and over time. And they fall flat in the domain of specialty care, where most healthcare dollars sit. 

Then there is the most under-appreciated category of investment: the expansion of case formation. Across firms, you see repeated bets on companies like Ro, Hims & Hers, K Health, Ada Health, Prenuvo, and Function Health. These platforms lower the threshold for entering the system. They make it easier to seek care, easier to get tested, easier to receive a diagnosis.

They are framed as access plays. But they are, more precisely, case generation engines. They sit directly in the layer where total cost is determined. Expand this layer without constraint, and the rest of the system must absorb the consequences.

Upstream, in behavior and environment, capital is thinner. Whoop, Oura, and Noom have built meaningful businesses, but sustained system-level integration remains limited. The theoretical leverage is high. The infrastructure to capture it is not.

The Pattern Across Portfolios

Taken together, these portfolios tell a consistent story. Each investment firm is right about something. But each is optimizing a different layer of the stack without a unified model of how those layers interact. And very few constrain the number of cases or systematically reduce variation within them. The system becomes more capable, but it doesn’t necessarily become more valuable.

Pathological Decision Proliferation

When these approaches converge, a predictable dynamic emerges. Access expands, decision-making becomes cheap, and entry points multiply. It’s a world of abundance, fashioned by the remarkable economics of AI and software. 

Each clinical interaction branches. A symptom generates multiple possibilities. Each possibility generates a workup. Each workup produces findings. Many of those findings are incidental. Each one introduces another decision.

The process begins to resemble… unchecked replication. This is pathological decision proliferation… powered by AI. And it concentrates in the most important layers of the system. More cases are created. More variation is introduced within each case. Downstream efficiency gains cannot offset upstream expansion.

The system becomes better at producing care. It does not become better at producing value.

The Missing Mechanism

What is largely absent from the current narrative is the structure required to control this dynamic.

Michael Porter’s “Integrated Practice Unit” offers one of the few frameworks that directly addresses it. It is based on organizing care around conditions, defining pathways, measuring outcomes over time, and feeding that learning back into the system.

This is a mechanism by which variation is constrained and value is created. Without a structure such as this, more intelligence produces more care and more variability. With it, intelligence can be directed and refined.

But even this is not sufficient without something deeper.

Infrastructure Begets Value

The real constraint is not intelligence, but rather infrastructure. Because healthcare does not lack data, it lacks coherence.

Information about a single episode of care is fragmented across claims systems, provider records, payment flows, and administrative intermediaries. Clinical decisions are not cleanly linked to financial outcomes. Attribution is weak. Feedback loops are incomplete.

And critically… the system cannot reliably answer a basic question: did this intervention create value?

Brent James has argued for decades that the largest opportunity in healthcare lies in eliminating waste. The focus must be on reductions in inappropriate care, avoidable care, and unnecessary variation. That opportunity is enormous… on the order of $1 trillion dollars. However, it remains largely inaccessible because the system cannot measure it consistently (nor is it incentivized to do so).

To extract value, you have to identify waste. To identify waste, you have to connect decisions to outcomes and costs across time. And that requires infrastructure that does not currently exist in a coherent form.

It requires a layer where clinical decisions, financial transactions, and outcomes are part of the same loop. Where an episode of care is not just delivered, but adjudicated in terms of its impact. Where learning is embedded in the transaction itself.

The absence of this layer is why so much innovation feels vaguely correct but systemically ineffective.

The Real TAM In Healthcare 

This has implications for how market size is understood. Most AI companies define their opportunity in terms of existing spend categories. Documentation. Billing. Administrative overhead.

These are real markets. But they are not the largest markets. The largest opportunity sits in the waste embedded in the system. The unnecessary cases, the avoidable complications, and the variation that produces cost without improving outcomes.

That is the real market… the true TAM. And while Brent James estimates it to be $1 trillion, it is only accessible if value can be measured in the flow of healthcare transactions. Without measurement, there is no way to price improvement. Without pricing, there is no way to capture it.

The system needs a way to track value across episodes, across journeys, across populations. It needs a means of understanding not just what happened, but whether it should have.

Building this requires integration across data sources, alignment of incentives, and a transaction layer that reflects reality rather than obscuring it. Until that exists, the largest market in healthcare remains invisible.

The Investment Mistake

The current wave of health tech investments is not irrational. It’s just misaligned (mostly).

Capital is flowing to what can be measured easily. Efficiency improvements are visible. Interfaces scale. Revenue is clear.

Meanwhile, the layers that determine value remain opaque. In some cases, the same capital is funding expansion of case formation while also funding tools that make the system more efficient at processing those cases.

The combination feels productive, but it’s often inflationary. You are making it easier to create demand and cheaper to service it, without improving how that demand is generated or managed.

Two Futures

From here, the system can evolve in two directions.

In one version, AI is layered onto the existing structure. Access expands, clinical decisions proliferate, and efficiency improves at the margins. The system becomes faster, more responsive, and more expensive.

In the other version, AI is embedded within systems that control entry into care, define pathways, and measure outcomes continuously. Decisions are constrained. Learning compounds. The number of cases declines even as outcomes improve.

The key here is not the AI model, or even its compatibility with real-world clinical workflows. The key is the infrastructure and incentives that surround it.

The Uncomfortable Conclusion

AI will deliver care to billions. That will surely come to pass. What remains uncertain is whether that care will actually create value.

Value is not a byproduct of access or intelligence. It is the result of a system that can define what quality looks like, measure it consistently, and learn from it over time.

We do not have that system today. Until we do, AI will not fix healthcare. It will simply expose it.

And in doing so, it will make clear that the largest opportunity was never just intelligence. It was building the foundation required to use it.

Sure, Kaiser is vertically integrated. It’s the payer and the provider. But this is not a business model story. 𝗜𝘁’𝘀 𝗮 𝗰𝗮𝘀𝗵 𝗳𝗹𝗼𝘄 𝘀𝘁𝗼𝗿𝘆.

𝗠𝗼𝘀𝘁 𝗵𝗲𝗮𝗹𝘁𝗵 𝘀𝘆𝘀𝘁𝗲𝗺𝘀 𝗼𝗽𝗲𝗿𝗮𝘁𝗲 𝗹𝗶𝗸𝗲 𝘁𝗵𝗶𝘀:

• Deliver care

• Book revenue

• Sit on massive A/R (accounts receivable)

• Wait 60–120+ days to get paid

High DSO (days sales outstanding). Long cash conversion cycles. Constant working capital pressure.

Translation: you’re (marginally) profitable on paper… but cash-constrained.

𝗡𝗼𝘄 𝗰𝗼𝗺𝗽𝗮𝗿𝗲 𝘁𝗵𝗮𝘁 𝘁𝗼 𝗞𝗮𝗶𝘀𝗲𝗿:

• Paid upfront

• Minimal A/R

• Revenue recognized over time

• Obligation = future care (not unpaid bills)

Zero/negative cash conversion cycle. Cash arrives before the work. Same industry. Opposite financial reality.

𝗛𝗲𝗿𝗲’𝘀 𝘁𝗵𝗲 𝘂𝗻𝗰𝗼𝗺𝗳𝗼𝗿𝘁𝗮𝗯𝗹𝗲 𝘁𝗿𝘂𝘁𝗵: fee-for-service (FFS) is a working capital trap. You do the work, book the revenue, finance the gap, and hope you get paid.

The other extreme is “capitation”, also known as population-based payment or “PMPM” (per member per month). This is: “cash first, care later”. Which translates into:

• Lower DSO

• Minimal working capital needs

• Real operating leverage

• Actual financial resilience

⸻

We talk about value-based care like it’s a clinical redesign problem. But there’s more to the story. 𝗜𝘁’𝘀 𝗮𝗹𝘀𝗼 𝗮 𝗰𝗮𝘀𝗵 𝗰𝗼𝗻𝘃𝗲𝗿𝘀𝗶𝗼𝗻 𝗽𝗿𝗼𝗯𝗹𝗲𝗺.

FFS optimizes for revenue. Capitation optimizes for cash. If you don’t know the difference… you’ll always think VBC is a financial loser.

Warren Buffet taught us that “cash is king”. It’s why SaaS thrives.

Capitation is the SaaS of healthcare. 𝗧𝗵𝗼𝘀𝗲 𝘄𝗵𝗼 𝗳𝗶𝗴𝘂𝗿𝗲 𝘁𝗵𝗶𝘀 𝗼𝘂𝘁 𝗳𝗶𝗿𝘀𝘁 𝘄𝗶𝗹𝗹 𝘄𝗶𝗻.

This is the story of e-commerce. It is also, with an uncomfortable degree of precision, the story of healthcare payments unfolding right now.

Part One: The Extractive Incumbent

Amazon did not win e-commerce by having better products. It won by solving a trust problem… and then, having solved it, turning that trust into a toll booth.

The model was elegant in its construction. Aggregate enough merchants and customers onto a single platform, make the transaction frictionless, build the logistics infrastructure that no individual merchant could afford, and then gradually tighten the terms. Amazon’s take rate - the percentage of each transaction it kept - increased steadily as its dominance grew. The customer relationship belonged to Amazon, not the merchant. The merchant’s sales data fed Amazon’s private label operation. The ranking algorithm was a black box that Amazon controlled. You sold on Amazon or you didn’t exist to tens of millions of customers.

This was not a conspiracy. It was the logical destination of a business model built on extraction. Once you are the only viable marketplace, every service you layer on top - fulfillment, advertising, lending - becomes a tax the merchant must pay. The merchant’s success was Amazon’s raw material, not Amazon’s goal.

For most of the 2000s, this felt permanent. E-commerce consolidation looked like a gravitational collapse with one endpoint. The question was not whether Amazon would dominate, but how much it would cost everyone else to participate in the world it had built.

Then… Shopify came along and asked a different question entirely.

Part Two: Shopify and the Rebellion

Tobi Lütke did not set out to build a commerce platform. He set out to sell snowboards online, found the existing tools inadequate, and then built his own. That origin story matters because it means Shopify was built from the merchant’s perspective… from someone who had tried to run a business on the internet and found the infrastructure hostile.

The insight that animated Shopify’s entire strategic arc was simple: the merchant who wanted to build a brand, own a customer relationship, and create something with meaning was fundamentally underserved by Amazon and fundamentally underserved by the existing tools. That merchant didn’t need a marketplace, as much as he needed infrastructure.

The distinction sounds semantic, but it isn’t. A marketplace brings you customers and takes a cut of every transaction in perpetuity, owns the customer relationship, controls the algorithmic ranking that determines your visibility, and reserves the right to compete with your own products using your own sales data. On the other hand, infrastructure gives you the tools to find your own customers, own that relationship permanently, and build something that compounds over time.

Shopify’s explicit framing - “arming the rebels” - was not marketing copy. It was a precise statement of strategic intent. Amazon was the empire. It had the logistics, the Prime membership, the trust infrastructure, and the distribution. Shopify was not going to beat Amazon by building a better Amazon. It was going to arm every merchant who wanted to build something Amazon couldn’t enable: a brand.

What made this viable, and what most observers missed at the time, was that the internet was becoming a discovery layer that merchants could access without building it themselves. Facebook and Instagram advertising, influencer relationships, email lists, TikTok… these were mechanisms for reaching customers that didn’t require paying Amazon for shelf placement. Shopify understood that if discovery could happen anywhere, you just needed to be the conversion and fulfillment layer. Let the customer find you through culture; let Shopify handle the transaction.

But the deepest move Shopify made was not the storefront itself. It was the platform.

Part Three: The Infrastructure Play

Somewhere between 2009 and 2012, Shopify made a decision that most software companies get wrong. Instead of trying to build every feature merchants might need, they built an operating system and invited everyone else to build on top of it.

This decision looks obvious in hindsight. At the time, it was a genuine act of strategic restraint… and restraint is one of the hardest things for a software company to practice. The temptation is always to build more, to capture more surface area, to own more of the value chain. Shopify chose instead to define the substrate clearly and hold it.

The substrate they defined was the set of fundamental commerce primitives: cart, checkout, customer identity, inventory, payments, fulfillment. These are the atomic units of what commerce is. Shopify expressed them in clean, well-documented APIs and made those APIs available to any developer who wanted to build on top of them. This standardization was the unlock. Without a common data model, every integration between systems is a custom bilateral negotiation - expensive, fragile, and impossible to scale. With that model, however, connections become composable. A developer building a loyalty app doesn’t need to understand Shopify’s internal architecture. They need to understand the API for customer identity and purchase history. That’s it.

The Shopify App Store is what happened next. It is one of the most instructive examples in software history of what occurs when you define a clean platform and open it to outside developers. Thousands of applications emerged: loyalty programs, subscription billing, review platforms, returns management, upsell engines, wholesale portals, email marketing, fraud detection, international tax compliance, influencer tracking. These were categories that Shopify had never thought to build… problems merchants had that Shopify didn’t know existed.

The platform didn’t predict this ecosystem. It made it possible, and then got out of the way. This is the Cambrian explosion dynamic. When you lower the cost of building on top of a platform, when the substrate is stable, the APIs are clean, and the distribution channel is the merchant base itself… you don’t get incremental innovation. You get an eruption of it.

Every developer who builds a successful app on Shopify is adding value to the platform without Shopify having to fund or manage it. The platform appreciates through the ecosystem’s activity. The ecosystem grows because the platform makes it worth building on. The two are in a self-reinforcing loop that becomes nearly impossible to displace from the outside.

The flywheel this created deserves to be stated precisely. More merchants on the platform made it more attractive for app developers to build on it. Better apps created more value for merchants, which attracted more merchants. Better merchant outcomes generated more data and credibility, which attracted larger brands and more sophisticated developers.

Shopify’s revenue, earned on payments, SaaS subscriptions, and merchant capital, was downstream of being indispensable. They did not optimize the take rate. They optimized for indispensability, and the take rate followed naturally from the position they had built.

The competitive moat this created was not any individual feature. It was the accumulated standardization… years of API design decisions, data model choices, developer relationships, app ecosystem participants, and merchant trust that would be brutally expensive to recreate. Any specific Shopify capability could theoretically be replicated in six months. But the platform took a decade. And crucially, the platform’s value compounds with time in a way that point solutions do not. Every new app added to the ecosystem makes the platform more valuable to every existing merchant. Every new merchant makes the ecosystem more attractive to every existing developer. The network effect is not just between merchants and customers… it is between the platform, its developers, and its merchants simultaneously.

The final mark of infrastructure that has truly won is this: it disappears. Shopify Plus brands often built completely custom storefronts - their customers never saw a Shopify UI. But Shopify was running checkout, payments, fraud detection, inventory, and fulfillment underneath. The infrastructure was invisible. The brand was everything. That invisibility is not a failure of Shopify’s branding. It is the proof that the infrastructure had become genuinely foundational.

Amazon, for all its dominance, could not replicate this. Amazon Webstore launched and was quietly shut down in 2015. The reason it failed is structural, not tactical. Amazon’s entire business model is adversarial to merchants at the margin… extracting value through fees, competing through private label, owning customer relationships. You cannot build merchant-aligned infrastructure inside an organization whose incentives require merchant extraction. The business model contradiction cannot be engineered away.

Part Four: The Healthcare Payments Parallel

Healthcare payments in 2026 is e-commerce in 2003.

The infrastructure is archaic. The dominant players profit from complexity rather than from value creation. The entities nominally served by the system - patients, employers, physicians - are in practice the entities most exploited by it. And beneath the surface of an industry that appears impregnable, the preconditions for a Shopify-like disruption are assembling.

To see this clearly, you have to understand how healthcare payments work and why it’s structured the way it is. A self-insured employer - the entity that actually bears insurance risk for most commercially insured Americans - does not have a direct relationship with the healthcare system. Their health benefits flow through a broker, who selects a third-party administrator, who contracts with a network, whose contracts flow through a clearinghouse, whose claims go through utilization management, whose approvals do not guarantee actual payment, whose adjudication happens weeks or months later after passing through a repricer and a payment integrity vendor. Every node in this chain extracts a fee. The employer, through this morass, has paid approximately sixty-five cents of overhead for every dollar that reaches a physician.

This is not a system designed for the employer. It is a system designed to extract from the employer while providing just enough apparent value to prevent replacement.

The TPA earns fees on complexity. The PBM earns spread. The network earns access fees. The PBM earns a percentage of discounts it claims to negotiate. None of these entities have incentives that align with making the underlying transaction (patient receives care, provider gets paid, employer bears appropriate risk) simpler, faster, or cheaper.

The physician’s relationship to this machine is particularly instructive. A surgeon who performs a complex spinal fusion and achieves an excellent outcome (patient returns to work, avoids reoperation, lives without chronic pain) gets paid on a fee schedule determined by collections… after the claim clears utilization management, after adjudication, after payment integrity review… sixty to ninety days after the service was rendered. The outcome is invisible (and irrelevant!) to the payment. The value created is invisible to the payment. Whether the patient was the right candidate for surgery, whether the approach was appropriate, whether the result was optimal - none of this is captured in the transaction.

This is Amazon circa 2003. The dominant infrastructure exists. It processes enormous volume. It appears to most observers to be too entrenched to displace. And it is profoundly misaligned with the people it is supposed to serve.

Part Five: What Shopify Saw

Shopify succeeded because it correctly identified that the apparent inevitability of Amazon’s dominance rested on an assumption that could be falsified: that the merchant didn’t need to own the customer relationship. But once DTC brands proved that consumers would actively seek out a brand they trusted… would pay a premium for it, would share it, would build identity around it.:. the entire premise of Amazon’s marketplace model had a leak in it.

The equivalent assumption in healthcare is that the payment infrastructure is too complex and too entrenched to standardize. That every payer-provider relationship is too unique for a common data model. That real-time adjudication is technically impossible given the complexity of benefit design. That outcome-linked payment is too hard to operationalize at scale.

These assumptions are wrong in exactly the way that “merchants don’t need to own customer relationships” was wrong. They are wrong not because they are implausible on their face, but because they serve the interests of the entities whose business model depends on complexity.

The X12/EDI transaction standard in healthcare is not archaic because healthcare is complicated. It is archaic because simplifying it would destroy the business model of every intermediary whose margin depends on the friction.

The Shopify insight, translated into healthcare payments, is this: if you build a payment infrastructure that standardizes the fundamental primitives - episode definition, fee scheduling, real-time adjudication, instant payment, outcomes capture - and expose that infrastructure through clean APIs, you can run condition-specific payment programs on top of it the way Shopify runs loyalty and subscription apps on top of its commerce primitives.

Any bespoke value-based care program you can think of… these are not products. They are the first applications in an App Store that doesn’t yet have a platform. The platform is the key. The platform is what everything else runs on. And when the platform exists… when any employer can connect to it, any physician can be paid through it, any condition-specific care program can be deployed on it… the application layer will proliferate in ways that are impossible to predict and pointless to try to control.

This is the Shopify flywheel for healthcare. More employers on the platform means a more attractive audience for care program designers. Better programs mean better outcomes data. Better outcomes data means tighter actuarial models, which means more accurate pricing, which means more employer trust, which means more employers on the platform. The infrastructure appreciates without the platform owner having to build every application.

Part Six: Why the Incumbents Won’t Save Us

The instinct in healthcare is to assume that one of the existing giants will solve this. Amazon has tried, repeatedly and expensively: Amazon Care, Haven, PillPack, One Medical.

UnitedHealth Group has assembled a vertical stack of terrifying completeness: insurer, PBM, physician group, surgery centers, home health, data analytics. CVS absorbed Aetna and is attempting something similar. The argument for these plays is always the same: scale eliminates friction between nodes, and enough vertical integration eventually produces alignment.

This is the Amazon strategy applied to healthcare. And it will fail for the same reason Amazon Webstore failed. The failure is not one of engineering or strategic vision. It is a failure of incentive structure… which is to say, it is not really a failure at all from the incumbent’s perspective. It is the system working exactly as designed.

Consider what it would actually mean for UnitedHealth Group to build a transparent, real-time, outcomes-linked payment infrastructure. It would mean that claim adjudication happens instantly, eliminating the float on sixty-to-ninety-day payment cycles. It would mean that network contract terms are visible to employers, eliminating the spread between what UHG pays providers and what it charges plan sponsors. It would mean that prior authorization criteria are algorithmic and auditable, eliminating the utilization management business that generates revenue by denying care. It would mean that outcomes data is standardized and public, enabling employers to compare providers on value… eliminating the network access fees that exist precisely because comparison is currently impossible.

Every one of these improvements destroys a revenue stream. Taken together, they would dismantle the business model. Because UnitedHealth Group’s market capitalization is not a reflection of the value it creates for patients and employers. It is a reflection of the rent it extracts from a system it has successfully made itself indispensable to.

Building the infrastructure that would eliminate those forms of rent is not a strategic option for an organization with those obligations to shareholders. This is the precise mechanism of the innovator’s dilemma as Clayton Christensen described it… but with an important clarification. Christensen’s framework is usually described in terms of incumbents being too slow, too bureaucratic, or too focused on existing customers to see disruptive threats. That framing is too charitable. The more accurate description, in industries where the incumbent’s value comes from information asymmetry and transaction friction, is that the incumbent sees the threat clearly and correctly identifies it as existential.

The response is not to ignore it. The response is to acquire potential disruptors, to lobby for regulatory barriers, to use contract terms that lock in intermediary relationships, and to move just fast enough in the direction of reform to claim participation without actually enabling competition.

This is what Amazon did when it launched Amazon Webstore - a nominal concession to the merchant-ownership thesis, built without the platform architecture, the developer ecosystem, or the merchant-aligned business model that would have made it real. It was a defensive gesture, not a genuine platform. When it failed to gain traction, Amazon shut it down and returned its attention to what actually made money.

The healthcare equivalents of Amazon Webstore are everywhere. Transparency tools that display charge-master prices no one actually pays. Direct contracting arrangements that bypass one intermediary while flowing through three others. Reference-based pricing programs administered by the same TPAs they’re supposed to disrupt. Each of these is a real product with real revenue. None of them change the underlying infrastructure. They are features built on top of a broken substrate… and features cannot fix substrate problems.

The disruption, as in e-commerce, will come from the outside. Not because the incumbents lack intelligence or capability, but because the infrastructure that would actually solve the problem is incompatible with the business model of the incumbents.

Epilogue: The Pattern

The pattern always looks the same in retrospect and almost impossible a priori.

In 2006, building infrastructure for independent merchants to compete with Amazon sounded quixotic. Amazon had the logistics, the trust, the Prime membership, and the marketplace network effects. Why would anyone shop at a small brand’s standalone website when they could get everything on Amazon with two-day shipping? The conventional wisdom was that e-commerce consolidation was terminal.

What the conventional wisdom missed was that the infrastructure for a different model… one aligned with merchants rather than extracting from them… had never been built. And once Shopify built it, the DTC brand category came into existence. Not because consumers suddenly changed what they wanted, but because the infrastructure to serve what they had always wanted finally existed.

In 2026, building real-time adjudication infrastructure for a value-based payment system sounds equally quixotic. The TPAs have the relationships, the networks have the contracts, and the incumbents have the employer relationships. The conventional wisdom is that healthcare payment reform is permanently stalled because the incumbent economics are too entrenched.

But what the conventional wisdom is missing is that the infrastructure for a different model has never been built. The X12/EDI world was never designed for outcomes-linked, journey-based, real-time payment. It was designed for fee-for-service claims processing, and everything built on top of it has been retrofitted to that constraint.

When the new infrastructure is available… and this is inevitable, because the economic waste it would eliminate is too large to ignore forever, and because the technology to build it now exists… the clinical application layer will explode in ways that are hard to predict from here.

Condition-specific VBC programs that are currently impossible to operationalize will become straightforward to deploy. Physicians who currently have no mechanism to be paid for outcomes will have one. Employers who currently have no visibility into the value of their healthcare spend will be able to see it, compare it, and act on it.

The rebels will be armed. And the empire will discover, as empires always do, that its moat was really just everyone else’s failure to build the bridge.

I spent this weekend reviewing several papers.
Similar topics.
All with bold claims.
But, with completely different outcome metrics.
And somewhere between the third and fourth paper, I caught myself thinking about patient survey fatigue, and realized I’m mentally fatigued too.

Still grasping at what outcome metrics truly matter in MSK and spine care.

So, I switched over to some 🏀 NBA Playoff games (ahem…Bring back the Sonics!)
In every sport, we agree on the “scoreboard” and “stats” before the game starts. Points. Rebounds. Assists. Player efficiency rating. True shooting percentage.
The metrics get more sophisticated every year, but the foundation is clear: everyone knows what counts, and everyone measures it the same way.

Now imagine a basketball league where every team tracked a different stat line.
1️⃣ One counts points.
2️⃣ Another counts “offensive contributions.”
3️⃣ A third uses a proprietary index that blends shots, hustle, and vibe.
4️⃣ The commissioner has no way to rank teams.
5️⃣ Free agency is chaos.
🖐 That is orthopaedic, MSK, and spine care today.

Every institution, every surgeon, every payer keeps their own books with their own definitions, their own denominators, their own idea of what a “win” looks like. And then we wonder why “value-based care” and cost-containment keeps stalling in our specialty. 🤷‍♂️

You cannot pay for outcomes if outcomes are measured five different ways on the same patient.

The fix isn’t another framework layered on top of the existing ones.
We have enough of those.
The fix is COORDINATION and CONSENSUS.

We need a national committee: some sort of a governing body, in the truest sports sense: to define the core quality metrics in spine and other orthopedic specialties.

Not every PRO metric. (thats exhausting, and not all that meaningful).
Not one single “perfect one”.
We need a FLOOR.
The basic ones we all agree to track, report, and be measured against.
The societies have tried …. But we don’t have much to show for it, yet.

The NBA and NFL didn’t wait for consensus on advanced analytics before standardizing the box score and the stat sheet.

They standardized the basics first. Everything else followed.

Orthopaedic, MSK, and spine care are overdue for the same move.

So how do we take the next steps in building this scoreboard?

Leave a comment

Subscribe now

Over the past several years, healthcare systems have increasingly embraced standardization and lean operational models. And in many ways, that’s a good thing.
Standardization can:
✅ Reduce errors
✅ Improve safety
✅ Lower costs
✅ Eliminate true operational waste

But from the perspective of someone working inside a surgical specialty like spine care, I’ve begun to wonder:
❓ Are we pushing standardization and lean processes too far ❓

Healthcare is not manufacturing.
Patients do not arrive in predictable patterns. Their pathology varies widely. Their recovery trajectories vary even more.

Ive learned that one of the fundamental truths is that variability is unavoidable in service systems. When demand and service times fluctuate, systems must maintain some degree of slack capacity or redundancy to function well.
In other words:
💡 Not all “waste” is actually waste.
💡 Some redundancy is strategic.

A system designed to operate at near-perfect efficiency may look great on a spreadsheet, but in reality it may become:
1️⃣ More fragile
2️⃣ Less responsive
3️⃣ More prone to delays and errors
4️⃣ More exhausting for the teams working inside it

The uncomfortable reality is that many operational changes today are driven by short-term efficiency targets rather than long-term system resilience and sustainability.
Lean initiatives often focus on reducing personnel, rather than improving processes.
And when incentives become misaligned, the consequences can appear downstream:
✅ Longer patient wait times, more difficult access
✅ Reduced adaptability to complex cases
✅ Increased clinician burnout
✅ Lack of innovation

A more important question might be:
❓ What variation actually matters ⁉️
Not all variation is bad. The key question is whether variability affects patient outcomes or the cost of care.
➡️ If it doesn’t, it may be harmless (or even necessary).
➡️ If it does, that is where standardization should focus.

So perhaps the real goal is not maximum efficiency, but something harder to achieve:
⭐ Operational systems that are efficient in the short term, but resilient over the long term. ⭐

Healthcare systems today face a difficult challenge:
How do we design organizations that are simultaneously
✅ standardized
✅ adaptable
✅ innovative
✅ sustainable
✅ and centered on patient outcomes?

That question may ultimately determine which health systems truly succeed over the next decade.

I (and many of my wonderful colleagues) are on the front lines facing the consequences of all the decisions being made on our behalf.

Thank you to Christian Terwiesch and Gad Allon --- who have laid down the important principles to keep my brain racing in this space. The Wharton School

And also RIP Aaron Rosenberg, who always encouraged me to keep stretching and writing!

Leave a comment

Subscribe now

Two patients.
Same diagnosis.
Same surgeon.
Same minimally invasive decompression.
Same “textbook” operation.

One says:
“This changed my life.”
The other says:
“I wish I hadn’t done it.”

What happened?

🎯 The Expectations–Outcomes Mismatch
We often measure surgical success by what we do in the operating room.
But patients measure success by what they thought would happen afterward.

✈️ As someone who flies too much, this reminds me of flying:
- If you expect a direct flight and get one with a short layover, you may still arrive on time, but feel disappointed.
- If you expect turbulence and the ride is smooth, you’re thrilled ... even if the arrival time is identical.

In spine surgery, the “arrival time” (objective outcomes like ODI, SF-36) may improve substantially. But if pain relief, function, or return-to-work don’t match the anticipated degree of improvement; satisfaction drops.
Not because surgery failed.
Because expectations and outcomes diverged.

📚 We Studied that Gap in the most common group of spine surgeries: lumbar decompressions.
1️⃣ “Expectation vs. Reality: Exploring Decisional Regret in Minimally Invasive Lumbar Spine Surgery” - Congress of Neurological Surgeons (CNS) NEUROSURGERY https://lnkd.in/grHPfF2b
2️⃣ “From Anticipation to Achievement: Evaluating the Impact of Preoperative Expectations on Postoperative Outcomes, Functional Health, and Patient Satisfaction in Lumbar Decompression Surgeries” - Journal of Neurosurgery Publishing Group Spine
3️⃣ “Barriers to Adhering to Post-operative Recovery in Spinal Decompression Surgery and Impact on Patient-Reported Outcomes” - Global Spine Journal (GSJ) https://lnkd.in/gX4KKwCK

Here’s what we found:
1️⃣ Lumbar decompression often "works". Disability improves significantly. Quality of life improves.
2️⃣ Pain relief expectations are very high. ~90% of patients expect major improvement. But fewer actually achieve the level they anticipated.
3️⃣ Mismatch (not baseline health, not demographics) was the strongest predictor of satisfaction.
4️⃣ Mismatch drives regret. Expectation–actuality discrepancy strongly correlated with decisional regret
5️⃣ Recovery barriers matter. Difficulty with exercise adherence, medication access, transportation, and care understanding reduced odds of meaningful functional improvement.

So what are the takeaways?
✅ Expectation alignment is the hidden denominator of value. Two identical outcomes can feel very different depending on what patients expected (and in value-based/outcomes-based care, perception is measurable).
✅ Satisfaction now impacts revenue/reputation. Bundled payments, PROMs, and public reporting mean technical success alone is not enough.
✅ Expectation management is a clinical intervention. Measuring and calibrating expectations > new tech.
✅ Call to action: Operationalize expectation alignment and build it into shared decision-making and treat it as a core value metric.

Leave a comment

Subscribe now

I got this email at 9am: “Your surgery has been denied.” I was offered a “peer-to-peer” call. A chance to plead my case.

The deadline was 3pm… 𝘁𝗵𝗲 𝘀𝗮𝗺𝗲 𝗱𝗮𝘆. That gave me 6 hours. No matter that I’m a surgeon. No matter that I have surgeries. No matter that I have patients to see.

6 hours.

And to add insult to injury… we have to call a 1-800 number. “Choose option 2.” And then provide a reference number… so that we can 𝘀𝗰𝗵𝗲𝗱𝘂𝗹𝗲 the “peer-to-peer”.

Then we get a phone call on our cell… at some point during a 2-3 hour window. Tough luck if you miss the calls. Tough luck if you’re with another patient.

I think you know the story’s ending. The peer-to-peer call was a failure. The surgery denial was upheld.

I’d recommended an outpatient, minimally-invasive spinal decompression for my patient. No screws or rods. His pain was intense.

I knew that more physical therapy wouldn’t do anything. 𝗧𝗵𝗲 𝗽𝗮𝘁𝗶𝗲𝗻𝘁 𝗸𝗻𝗲𝘄 𝗶𝘁. But the payer didn’t care. And the payer has the power.

𝗣𝗮𝘆𝗺𝗲𝗻𝘁 = 𝗽𝗼𝘄𝗲𝗿.

This is why self-funded employers need to wake up. You’re the payer, not the insurance carrier. It’s your money. You have the power.

Unlike the insurance carriers, your interests are actually aligned. You want your employees to be healthy and productive.

𝗦𝗼 𝗱𝗼 𝘄𝗲.

Physicians and employers are aligned. There’s just too much nonsense in between.

It’s time to cut out the interlopers. It’s time for direct access… direct payment. Our patients deserve nothing less.

What does this mean for healthcare?

In November 2025, Dylan Patel's SemiAnalysis published a seminal piece entitled: "𝗚𝗼𝗼𝗴𝗹𝗲 𝗧𝗮𝗸𝗲𝘀 𝗮 𝗦𝘄𝗶𝗻𝗴 𝗮𝘁 𝘁𝗵𝗲 𝗞𝗶𝗻𝗴."

Why should physicians pay attention? Nvidia vs Google... why should we care?

For years, Nvidia has ruled AI compute. It optimizes for what the market wants:

𝗣𝗲𝗮𝗸 𝗽𝗲𝗿𝗳𝗼𝗿𝗺𝗮𝗻𝗰𝗲 (throughput).

But something is changing. Google's AI chip - the TPU - is attracting attention from the smartest decision-makers. Why?

Because peak performance isn't the metric that matters. What actually matters is:

𝗣𝗲𝗿𝗳𝗼𝗿𝗺𝗮𝗻𝗰𝗲 𝗽𝗲𝗿 𝗧𝗖𝗢 (TCO = total cost of ownership)

It doesn't matter how fast a chip can run in a benchmark. What matters is how efficiently a system delivers useful work, in the real world, over time.

Google's TPUs are less impressive on paper. Their peak FLOPs are worse. And yet, at scale, they can deliver 𝗺𝗼𝗿𝗲 𝗿𝗲𝗮𝗹 𝘃𝗮𝗹𝘂𝗲 𝗽𝗲𝗿 𝗱𝗼𝗹𝗹𝗮𝗿.

How did Google pull this off?

It's not about chip microarchitecture. It’s about 𝘀𝘆𝘀𝘁𝗲𝗺 𝗱𝗲𝘀𝗶𝗴𝗻.

Google is nipping at Nvidia's heels. And healthcare faces the same reckoning.

Fee-for-service is healthcare’s "peak FLOPs". And fee-for-service is optimized for one thing: 𝗧𝗵𝗿𝗼𝘂𝗴𝗵𝗽𝘂𝘁.

And like peak FLOPs, "throughput" within a fee-for-service model tells you nothing about whether the system is 𝙖𝙘𝙩𝙪𝙖𝙡𝙡𝙮 𝙬𝙤𝙧𝙠𝙞𝙣𝙜.

Which brings us to the real metric: 𝗢𝘂𝘁𝗰𝗼𝗺𝗲𝘀 𝗽𝗲𝗿 𝗧𝗼𝘁𝗮𝗹 𝗖𝗼𝘀𝘁 𝗼𝗳 𝗖𝗮𝗿𝗲. (Remember performance per TCO?)

This is what employers want to buy.

But if you want to optimize this, you can't just maximize individual clinician capability (𝗰𝗵𝗶𝗽 𝗺𝗶𝗰𝗿𝗼𝗮𝗿𝗰𝗵𝗶𝘁𝗲𝗰𝘁𝘂𝗿𝗲!). You do it by designing care systems.

Enter the Integrated Practice Unit (IPU). (𝙏𝙝𝙖𝙣𝙠𝙨 𝙈𝙞𝙘𝙝𝙖𝙚𝙡 𝙋𝙤𝙧𝙩𝙚𝙧!)

Organize teams around conditions. Measure outcomes across the care cycle. Tie responsibility to total cost.

Peak FLOPs vs "performance per TCO". Microarchitecture vs system design.

Same tradeoff. Different domain. So why is healthcare still stuck?

The infrastructure doesn’t exist. You can't compete on value if...

Costs are settled months later. Accountability is fragmented. Outcomes are invisible. Teams can't be paid as units.

Once the right infrastructure exists, however, something fundamental changes. Care models (not individual services) become the unit of competition. Multidisciplinary teams compete on outcomes per cost. Employers choose systems that deliver durable value.

Remember, Nvidia won't lose overnight. Google won't win overnight.

But the smart money knows where the puck is going. Away from "peak specs", and towards system-level efficiency.

Healthcare is at the same moment.

The only question is: which health systems will take bold action?

And which will keep optimizing for a benchmark that the market ultimately abandons?

Spending time at the JP Morgan Healthcare Conference #jpm2026 has been both energizing and unsettling (in the right way!)

💡 It’s one of the few opportunities where clinicians, health system leaders, investors, economists, and operators are all talking about the SAME SYSTEM … but often in very DIFFERENT LANGUAGES.

As physicians and surgeons, it’s easy to feel like we’re standing on the outside of those conversations (especially as I look around the audience here) convinced that no one truly understands what our days look like, what our decisions carry, or what our silence costs.

But here’s the uncomfortable truth I keep coming back to:

🛑 **If we don’t learn how to translate our experience into the language of systems, value, and leadership, someone else will define it for us.**

We were trained to pursue technical mastery and knowledge.

But I feel like our education is missing something far more important (that maybe only experience provides?)

The moment we’re entrusted with the most responsibility (patients, teams, outcomes, risk), we’re placed inside systems that change faster than our training ever prepared us for.

Burnout, frustration, and moral injury aren’t signs of weakness.

They’re symptoms of professionals being asked to operate in environments that no longer reward learning…only output. And asking to do more with shrinking resources!

⭐ John Maxwell has been well quoted here…

1️⃣ “Everything rises and falls on leadership.”

2️⃣ “The pessimist complains about the wind. The optimist expects it to change. The leader adjusts the sails.”

Complaining may be cathartic (we are very good at that).

But it doesn’t move systems.

What I’m learning, watching hospital CEOs, operators, and policy leaders up close, is that our absence from these rooms isn’t neutral. It actually creates a vacuum. And these vacuums get filled.

It’s not their fault that they don’t have our lived clinical experience.

But it is our responsibility to learn how to explain it; credibly, rigorously, and across perspectives.

That doesn’t mean abandoning patient care.

✅ It means protecting it.

✅ It means learning economics without losing empathy.

✅ Leading without ego.

✅ Creating systems without forgetting the human cost of the decisions we make.

We have proven that we are good learners. Can we continue to keep learning in areas that we feel less comfortable?

Change won’t happen overnight.

Not no matter how loudly we yell (and some can create some serious volume).

But it can happen!

If we’re willing to keep learning, keep translating, and keep showing up in rooms that feel unfamiliar. Because as I’m sitting in the corner of this room today, I’m pretty sure I don’t know a soul here ... and thats okay! (for now...)

Subscribe now

Leave a comment

This time of year invites reflection … and planning!

Finals are done.
One last full clinical week.
Squeezing in patients with maxed-out deductibles.

And once again, I am hearing that healthcare is making the same resolutions:
“Lower costs.”
“Improve access.”
“Pay for value.”
“Put patients first.”

Yet every year… premiums rise faster than wages, access tightens, and clinicians are asked to do more with less … again.

Some problems require massive infrastructure and years to fix.
But where can we actually chip away in 2026?

✅ Resolution #1: Stop Punishing People for Being Good at Their Jobs

When does being better at your job lead to lower pay?
When you’re a physician.

Pay-for-performance hasn’t failed because outcomes don’t matter. Ot’s failed because efficiency is penalized. Reduce complications or LOS, and benchmarks quietly ratchet down. Congratulations: you’ve become a victim of your own success.

This year:

• Build floor-protected bundles and physician-led ACOs

• Reward durable quality, not endless year-over-year cuts

• Separate experience-driven efficiency from corner-cutting

✅ Resolution #2: Follow the Dollar — Then Collapse the Layers

Every healthcare dollar passes through too many hands before reaching patients or clinicians.
Brokers. TPAs. PBMs. Carriers. Vendors.
Plenty of friction, little accountability.

As a spine surgeon, I feel it daily:

• Prior auth delays

• Context-free denials

• “Oversight” tools without insight

This year:

• Expand direct contracting (employers ↔ physician groups ↔ ASCs)

• Support upstream navigation — not just procedural steering

• Treat administrative cost as a quality metric, not a fixed tax

✅ Resolution #3: Make Navigation the Front Door (Not the OR)

Healthcare doesn’t just have an access problem, it has a directional one.

Patients don’t need more doors.
They need the right first door.

I routinely see patients months (or years) into journeys that could’ve been shorter, cheaper, and less painful if routed correctly on day one.

This year:

• Expand PT/RN/APP-led triage and digital front doors

• Pay for getting patients to the right care, not just delivering care

• Align navigation incentives with outcomes, not volume

✅ Resolution #4: Stop Pretending Consolidation Improves Access

Consolidation promises efficiency.
But, it often delivers higher prices, narrower networks, longer waits, and fewer independent voices.

In spine care, consolidation squeezes programs that manage complexity and revision, while rewarding throughput.

This year:

• Enforce site-neutral payments

• Support physician-led CMMI models

• Protect competition-friendly payment designs

✅ Resolution #5: Embrace Consumerism — Without Turning Care Into Retail

Patients are consumers now. That’s reality.
But healthcare isn’t a toaster oven.

You can shop for an MRI.
You can’t shop for an urgent spinal cord decompression.

This year:

• Publish quality and outcomes alongside price

• Pair transparency with clinical context

• Build shared decision-making tools that respect risk, values, and goals

✅ Final Resolution: Reward What We Say We Value

We say we value judgment, experience, compassion, outcomes, and stewardship.

Yet we reimburse volume, speed, coding, and compliance theater.

That’s not a clinician problem.
It’s a system design problem.

I still believe we can do better. Not with sweeping reform (although, yes … eventually), but with these tangible, meaningful frameworks that can be trialed in different centers.

Let’s make those our resolutions.

Leave a comment

Subscribe now

John couldn’t walk anymore. His legs would buckle after just a few steps. The “sciatica” pain was ferocious, and he was desperate.

His surgeon (me) recommended the TOPS procedure: a newer surgery that would “un-pinch” his nerves and stabilize a slipped vertebra, all while preserving his spine’s natural motion. The FDA labeled it a “breakthrough” in 2021, and approved it in 2023.

Tough luck. John has Medicare Advantage (MA), so my surgery was subject to prior authorization. John’s MA plan doesn’t cover the TOPS procedure, despite FDA approval. There’s no happy ending here. I couldn’t do the surgery I wanted for John.

ANDREW

Andrew was angry. His “sciatica” pain had come back with a vengeance. Six months ago, his left leg had been killing him. He went to an interventional pain physician, who implanted an “interspinous spacer” in his back. It was billed as a spinal “fusion”.

Now his pain was back. I un-pinched his nerves with an outpatient procedure, and he felt better. But Andrew should never have been in this mess to begin with.

Why did the insurance company approve an unnecessary “fusion” procedure on his back, to be performed by a non-surgeon? How did this get authorized?

THE TWIN SINS

John’s story is the sin of omission. “Prior authorization” blocked a high-value treatment. John didn’t get the care he deserved.

Andrew’s story is the sin of commission. "Prior authorization” enabled a low-value, inappropriate treatment. Andrew received care that he didn’t deserve.

Physicians (myself included) love to decry the evils of prior authorization. Each of us could fill a stadium with all of the nonsensical, infuriating “denials” we get from insurance companies. And we have the public on our side… many prominent voices (including ) have declared that everyone should “just trust doctors”. It’s a natural reaction to the widespread sins of omission.

But it’s not that simple. The Institute of Medicine has declared that at least 30% of healthcare spending is waste. These are the inappropriate, unnecessary, and low-value interventions that take place everyday in this country. warned us about this ten years ago in his seminal New Yorker piece: “Overkill - America’s Epidemic of Unnecessary Care”.

I take no pleasure in saying this, but we can’t “just trust doctors”. At least, we can’t do that as long as “fee-for-service” reigns supreme. The behavioral economics just don’t allow it. And I say this, begrudgingly, as a doctor myself. There are too many sins of commission… too much unnecessary/inappropriate care. The irony is that all of this “overkill” is authorized by the very same “utilization management” vendors that oversee so many silly denials. These are the twin sins of commission and omission, and any attempt to focus on one while ignoring the other, is either lazy or incomplete.

THE DEEPER ERROR

Die-hard proponents and opponents of “prior authorization” are both missing the deeper point. They treat care decisions as isolated events… snapshots in time:

• Approve this scan

• Deny that surgery

• Authorize this device

But healthcare isn’t a collection of disconnected Polaroids. It’s a full-length movie. For most conditions, healthcare unfolds along a care pathway - a linked sequence of decisions that are interdependent and compounding.

The real question, therefore, isn’t: “Should this treatment be approved?”

The real question: “Is this the right intervention, at the right moment, by the right clinician, within the context of this patient’s overall journey?”

Answering that question requires a different framework entirely.

IT’S NOT ABOUT “GATE-KEEPING”

The solution is not more “prior auth” or less “prior auth”. This is a false choice. Healthcare must instead be viewed through the lens of value-optimized care journeys. Luckily, there’s a manual for this.

Brent James, of Intermountain fame, pioneered the use of “shared baseline protocols”:

1. Select a high-dollar clinical process

2. Establish evidence-based best-practices

3. Bake those guidelines into the default clinical workflow

4. Encourage clinicians to deviate, as needed

5. Measure condition-specific outcomes and costs

6. Learn from, and gradually eliminate, variation in care decisions

There are two critical insights here. The first is that the care of most conditions should be treated as a longitudinal process. The second is that variation in decision-making is powerful, because it allows for care protocols to iterate and evolve, as long as you measure the outcomes and costs of any deviations from a shared baseline.

When specifying the care pathway for a given condition, protocols are not rules. They are hypotheses. And hypotheses are meant to be tested.

A REAL-WORLD LABORATORY FOR VALUE

What if we could scale Brent James’ concept of “shared baseline protocols” across the country, with the aid of software? Let’s say that a savvy, self-funded, jumbo employer wants to improve outcomes and reduce costs for diabetes care…

What that employer needs is the ability to prospectively track every clinical touchpoint - every diagnostic test, every follow-up visit, and every intervention - along with the corresponding, accumulated outcomes and costs… for each employee with diabetes. You can imagine a vast array of care pathways across this cohort of employees, with clustering around certain common sequences. These are clinical “genotypes”, which generate “phenotypes” in the form of outcomes and costs.

This is a real-world laboratory for value. You start by establishing a shared baseline protocol for diabetes care, and then you prospectively track how “genotypic variations” (clinical deviations from the protocol) impact phenotype (value).

Imagine a Cartesian plane where the y-axis represents diabetes outcomes, and the x-axis represents the costs of diabetes care. Each employees’ care pathway becomes a curve on that plot. Ideally, the curves rise steeply, suggesting that outcomes improve quickly with less spending (high value).

The image above represents ROC curves, but the curves are illustrative for our purposes. If you have the right data infrastructure and analytics capability, you can visualize every patient’s care journey as a value curve. You can then identify the key features that enable optimal value for certain care pathways, and incorporate them into an updated “shared baseline protocol”.

This is how you truly operationalize a Learning Healthcare System.

WHY THIS HAS NEVER WORKED BEFORE

None of these concepts are actually new. The concept of a Learning Healthcare System was first established by the Institute of Medicine in 2006. And Brent James started instituting “shared baseline protocols” decades ago.

So why do we still routinely approve the wrong care (sins of commission) and block the right care (sins of omission)?

It’s because we never built the unifying infrastructure that makes learning possible. And as Brent James declared in the New England Journal of Medicine: “Infrastructure Eats Culture for Lunch”.

A Learning Healthcare System requires a completely new transactional infrastructure, with features such as:

• Unification of clinical, financial, and functional data

• Freedom for clinicians to deviate from guidelines

• Condition-specific analytics to measure true outcomes and costs

• The flexibility to iterate on care protocols

• Real-time transparency around payment

• A single source of truth for what was done, why, and at what cost

Without this, outcomes and costs remain permanently divorced. You cannot apply selective pressure for value if the system cannot even agree on what happened.

THE WRONG DEBATE

The debate over prior authorization is typically framed as a moral struggle. One side wants to “just trust doctors”, while the other side is fixated on controlling costs. This framing is uninteresting and unproductive.

The real divide is between systems that can learn, and systems that cannot.

And as long as we rely on the existing, medieval infrastructure that pays late, measures little, discourages innovation, and fragments accountability… we will keep enabling unnecessary care and blocking necessary care.

This will continue happening, not because we want it to, but because the system is structurally incapable of telling the difference.

“IT’S THE INFRASTRUCTURE, STUPID”

A healthcare system that cannot measure outcomes per dollar spent, in real-time, has no moral authority to decide what care should or should not happen.

Until we build the infrastructure that enables this, and makes system-wide learning unavoidable, we will keep committing the twin sins of omission and commission.

It’s not that we lack compassion or evidence… it’s that we are tolerating an infrastructure that cannot see what it is doing.

Once we fix the infrastructure, with an eye towards optimizing value across care journeys, the next phase of healthcare reform stops being an ideological argument.

It becomes an engineering problem.

When physicians aren’t at the table, systems get built around theoretical workflows instead of real ones.
Friction grows. Bottlenecks multiply.
And “efficiency” ends up creating more inefficiency.

Here’s the truth:
👉 Physicians must be in leadership — and we must learn to lead better!

We’ve trained generations of doctors to diagnose under pressure…
but never taught them how to manage teams, align incentives, navigate change, or speak the language of systems and economics.

Yet no one else understands:
✅ Why a 15-minute delay derails the whole day
✅ Why a small policy tweak can tank an entire clinic
✅ Why variability explodes when the wrong people design the pathway
✅ Why “capacity” on a spreadsheet ≠ capacity in real life
✅ Why innovation cannot simply be cost-cutting maneuvers to improve short term metrics without investing into long-term sustainable growth.

✅✅✅ We know these things because we live them!

Healthcare doesn’t need more titles or layers of administration, it needs clinician leadership with actual clinical insight.

If we’re not in the room, the system will continue to be designed around us instead of with us.

⭐ So how do we train ourselves to lead while holding a full clinical load? ⭐’
Here’s what I’ve seen (and learned the hard way):

1️⃣ Start small. Leadership begins in the OR and clinic.
How you run a case, support your APPs, educate trainees, guide your team through complications, and communicate through friction, that is leadership.

2️⃣ Build emotional intelligence with the same intentionality as surgical skill.
Leadership isn’t about authority, it’s about awareness, empathy, consistency, and how people feel after interacting with you.

3️⃣ Understand the economics of care.
If we don’t understand the incentives, we can’t influence them.
(And yes — this is a major part of why I am back in class.)

4️⃣ Collaborate outside your specialty.
The left hand must talk to the right hand.
True system change happens only when clinicians, administrators, payers, and industry are aligned.

5️⃣ Push for responsible innovation, not shiny-object innovation.
Innovation matters, but innovation without purpose becomes noise.
Our job is to help focus it on what solves real pain points: safety, access, ergonomics, outcomes, value.

6️⃣ Don’t wait to be invited.
Most empty seats at leadership tables stay empty because clinicians assume they belong to someone else.
They don’t!!
If we’re not at the table, our profession is what gets served on it.

🎬 If 2026 is going to look different, it starts with physicians leading the redesign of the very systems that strain and constrain us. If we don’t actively work to change healthcare, healthcare will change us ... and uhhh, not for the better.

Leave a comment

Subscribe now

In the closing statement of his groundbreaking work, “On the Origin of Species”, Charles Darwin penned the phrase: “endless forms most beautiful”. This is the love he had for the variation we see in nature. Can we apply this lesson to healthcare?

In 2018, Dr. Caleb Stowell and Dr. Ari Robicsek published an article in the New England Journal of Medicine that should be required reading for everyone. They noticed that healthcare providers vary dramatically in how they manage common conditions. Some of the variation is benign, but often times it results in poorer outcomes and higher costs. The authors’ brilliant insight was to interpret this through the lens of evolutionary biology:

In nature, we know that random mutations in DNA create variation in genotype, which then leads to observable differences in species (phenotype). It is the selective pressure of evolutionary “competition” that causes certain phenotypes to propagate their genes, while others become extinct. “In medical practice, ‘genotypes’ are the many individual decisions that constitute a provider’s method of practice, such as the selection of a medication, a surgical approach, a communication style. ‘Phenotypes’ are the cost and clinical outcomes of care.”

Variation in elements of medical practice (genotype) drive differences in outcomes/cost (phenotype), but there is almost NO selective pressure that discourages low-value care. Until this changes, value-based care will always be more hype than reality.

A COMPETITIVE ECOSYSTEM

Two clinicians can look at the same MRI, same glucose log, same set of labs… and make radically different decisions. Variation is everywhere — in workflows, diagnostic thresholds, escalation patterns, follow-up cadence, patient engagement tactics, and resource utilization. But variation isn’t the problem. The problem is our inability to exploit that variation, select the best practices, and improve value.

What we need is a competitive ecosystem where care pathways can evolve. We need an ecosystem where evolutionary fitness is centered on value (outcomes achieved per dollar spent), so that we can finally move towards a Learning Healthcare System.

The SUPREMACY OF CMS

Unfortunately, we currently have the exact opposite of this. The dominant force in value-based care today is CMS. It’s not the free market, it’s not clinicians, and it’s not patients or employers. It’s the federal government. CMS sets the pricing. CMS decides the metrics, the risk adjustment, the benchmarks, and the glide paths.

It’s what evolutionary biologists would call artificial selection (as opposed to natural selection). The ecosystem is shaped not by competition, but by a single central planner that has limited feedback loops and frustratingly long iteration cycles. This leads to glacially slow learning, rigid program design, gaming of benchmarks, consolidation instead of innovation, and no real allowance for creativity.

It’s not that CMS is run poorly or that it has bad intentions. It’s just structurally incapable of doing what biological and economic systems do best: iterate rapidly, test variation, and allow the most valuable models to win. We don’t have natural selection in healthcare. We have top-down, federal rulemaking.

LET’S IMAGINE A DIFFERENT WORLD

Imagine a world where every clinical model - whether it’s a diabetes care pathway, surgical bundle, MSK program, or heart failure protocol - existed not simply as a static PDF or an isolated “pilot”… but as living, adaptive organism.

You have a genotype that encodes: diagnostic categories, care sequences, pathway triggers, communication cadence, escalation logic, and resource utilization.

You have a phenotype expressed in: patient-reported outcomes, total cost of care, patient adherence, recovery trajectory, complication rates, patient experience.

Now you apply selective pressure. Employers choose the models that optimize value. Clinicians adopt the care pathways that make them most competitive. High-value practices thrive and low-value practices die. Clinical variation leads to system-wide evolution, with value as the fitness function.

We’ve never had this before. And we can never have it until we fix something deeper.

THE RUBE GOLDBERG PROBLEM

In 2025, every clinical pathway (no matter how elegant or high-value) must operate within the dysfunction of today’s healthcare “supply chain”. There’s no escaping the TPAs, repricers, wrap networks, EDI loops, coordination of benefits, batch claims, payment delays, endless appeals, disconnected portals, and misaligned incentives.

You can’t unleash natural selection on top of this Rube Goldberg Machine. You can’t have a competitive ecosystem on today’s rails. The underlying infrastructure is too slow, rigid, opaque, and fragmented. Because of this…

• No pathways can be rapidly deployed

• No outcomes can be measured easily

• No costs can be settled in real-time

• No meaningful market feedback can occur

• No clinical iteration can happen

• No selective pressure can build

WHAT A NEW INFRASTRUCTURE CAN UNLOCK

Now let’s imagine that we’ve completely re-imagined the healthcare supply chain and achieved a “creative destruction” of the Rube Goldberg Machine. The details aren’t important in this moment, but let’s assume that we finally have the seamless, friction-less, software-enabled, “Amazon+Visa”-style platform we’ve always deserved but never realized we could have. What have we unlocked?

We now have a digital chassis for genotype, phenotype, and selective pressure:

1. Encode the genotype: care pathways become modular, digital objects that are structured, programmable, and measurable

2. Express the phenotype: outcomes and costs of every pathway are directly observable from the merging of clinical and transaction-level data

3. Apply selective pressure: purchasers select models based on performance (value), and providers adapt to stay competitive

CARE MODELS AS MINI-APPS

You may have seen Apple’s recent announcement of the “Mini Apps Partner Program” for the App Store. Apple is incentivizing developers to create “mini-apps” within parent apps that live on the App Store. A classic example of a mini-app would be a particular workout program that lives within “Nike Training Club” or “Peloton”.

We should view care programs as mini-apps that live and “compete” within a broader online marketplace for value. The deeper infrastructure layer manages: contracts, pricing, eligibility, cost-share logic, pre-treatment adjudication, real-time payment, compliance, data warehousing, and information flows across stakeholders. With all of this in place, each care model (mini-app) becomes an executable module that is condition-specific, continuously measured, easily modifiable, and well-positioned to compete for value.

Imagine multiple diabetes care models competing against one another. Imagine musculoskeletal care pathways competing against one another. Imagine surgical spine bundles, oncology navigation protocols, and behavioral health programs competing against one another… not on the basis of bargaining power and volume, but on the basis of value. This is the world we deserve.

THE ONLY PATH FORWARD

Once the infrastructure exists (thanks !), the ecosystem can shift. Let’s imagine a world where CMS stops being the bottleneck. Employers become intelligent purchasers. Pathways become measurable digital products. Clinicians adopt models that optimize BOTH their financials and value. Cost variation collapses as outcomes improve. Market pressure drives continuous improvement.

Variation leads to natural selection, which forces adaptation and a drive towards ever-increasing value. This is how we build a learning health system… not through regulation or mandate, but through an infrastructure-rebuild and market dynamics.

Let’s get started.

• Lantern for outpatient surgery bundles

• Carrum for centers of excellence

• Transcarent for comprehensive navigation and curated deals

• Cost Plus Wellness [NEW!] for “open source”, transparent contracts

• (at least 50 others)

These efforts matter. Each is a step forward for the industry. However, it’s time to acknowledge an uncomfortable truth. NONE of these efforts eliminate the underlying, archaic machinery of healthcare transactions. None of them invite the creative destruction of the waste-laden, grift-inspiring Rube Goldberg Machine.

Until you attack that exhausting apparatus of intermediaries, which sits between employers and providers, you can never fulfill the promise of “going direct”. As long as there are unnecessary middlemen in the supply chain of healthcare payments, we are all just tilting at windmills.

The ILLUSION OF “DIRECT CONTRACTING”

Let’s say you’re a savvy jumbo employer like Boeing. You sign a “direct contract” with a local hospital, and on the surface, it looks like you’re ahead of the curve. But you still need a TPA to process all the healthcare claims. You still need a repricer to convert billed charges to allowed amounts. You still need a “wrap network” to catch anything that falls outside your arrangement with the hospital. You still need vendors for enrollment, directory maintenance, EDI submissions, coordination of benefits, payment operations, stop-loss… the list goes on and on.

Sure, you bypassed the insurance function of a major carrier, but you’re still a thousand degrees removed from the providers. And you may still be paying the carrier as an “administrative services organization” (ASO)! This isn’t disintermediation. This isn’t direct. You’ve just relocated the intermediation. The same waterlogged machinery persists… just with new business cards.

This isn’t a critique of any one company. It’s a structural limitation facing every entity that operates on today’s rails. Lantern, Carrum, Transcarent, Cost Plus Wellness… each is important. But all of them plug into the same Rube Goldberg Machine.

MARK CUBAN’S COST PLUS WELLNESS

Cost Plus Wellness (CPW) is now alive and kicking, and its contribution to the industry is unmistakable. is a force of nature, and he’s tearing open the black box of hospital contracts. He’s publishing transparent prices, so that employers can become intelligent purchasers and finally understand what they’re paying for.

Price transparency is critical. Price transparency is necessary for true competition in healthcare. It is necessary, but it is not sufficient. Because price transparency is not the same thing as a functioning marketplace. To achieve that, you need the appropriate transactional infrastructure.

Whether an employer chooses the “turnkey” or “DIY” options for CPW, with the published contracts in hand, it still requires a TPA to administer the plan and a vast array of vendors to complete the supply chain. This isn’t the fault of CPW. CPW is a source of truth, but it’s not a comprehensive platform. Price transparency is a spark, but it’s not the engine.

CREATIVE DESTRUCTION IN HEALTHCARE

What becomes clear is that “direct contracting” vendors only appear direct on the front-end. The same legacy rails are hiding on the backend. Those vendors don’t provide a direct pipeline for money and information to move seamlessly between employers and providers. There’s an illusion of directness, floating on top of the age-old machine. If you want a truly direct model, you have to replace the machine. The Rube Goldberg machine cannot and should not be optimized. It must be replaced.

Healthcare isn’t special - it’s just late to the party. Most other industries have already undergone a creative destruction powered by software: travel, retail, payments, transporation, advertising… the list goes on. Healthcare just got stuck mid-stream. We digitized medical records and forms, we digitized claims, we digitized networks, and we digitized medical imaging. We just haven’t had a unifying operating system.

Because of this, everyone is forced to use duct tape to cobble it all together - a phenomenon which is ironically being amplified as more and more employers become self-funded. As employers look to become savvy purchasers, they replace a “plug and play” ASO with a perplexing panoply of point solutions. It’s a problem of “fragmentation”, as and like to say. The system multiplies… while we want it to simplify.

WHAT WE TRULY NEED

For “direct contracting” vendors like Cost Plus Wellness (CPW) to succeed, we need an Amazon+Visa for healthcare. It would have the following features, among others:

• Universal fee schedule engine - prices aren’t just visible, they’re actionable

• Real-time claims adjudication - without a TPA

• Instantaneous provider payment - why wait?

• Built-in logic for cost-sharing - why force “zero dollar” cost sharing on everyone?

• API-based data transfers - why are we using X12/EDI in 2025?

• Stop-loss and compliance integration - why tack these on at the end?

• Human-friendly, online marketplace - where’s the Amazon Prime for healthcare?

• (There’s a lot more, but you get the idea.)

When this new operating system exists, all value-adding entities will become stronger. CPW gets supercharged. Carrum gets supercharged. Transcarent gets supercharged. And the Rube Goldberg machine will finally collapse under its own weight.

Remember… transparency, without the ability to act on it, is just advertising. Publishing previously obscure information is wonderful. But we must set our sights higher. PDFs are helpful, but it’s a new platform that we truly need.

The cartoonist’s name was Rube Goldberg, and he inspired the concept of a Rube Goldberg Machine: a chain reaction-type contraption intentionally designed to perform a simple task in a comically overcomplicated way.

The ironic truth is that Rube Goldberg Machines exist in the real world. Whenever a seemingly simple task requires a cascade of friction-filled steps, you can be sure that an industry of intermediaries has entered the fray to capture value. (Over-engineered complexity is a nidus for rent-seeking middlemen.) And sadly, nowhere is this clearer than the arena of healthcare payments.

At first glance, healthcare seems simple enough. You go to the doctor’s office, provide your insurance information, and get the care you need. You pay whatever portion you owe, and the insurance company pays the rest. Easy, right?

Not so fast. Behind the scenes, there is a dizzying supply chain for healthcare payments that would be make Rube Goldberg faint. It’s critical to understand this supply chain, because you can’t fix what you don’t understand:

1. THE EMPLOYER (THE PROTAGONIST)

   First, we have to get something out of the way. Most American workers receive health insurance through their employer, but the devil is in the details. In 2025, roughly 65% of those workers have employer-sponsored health plans that are “self-funded”. This is in contrast to the “fully-insured” model, which has been the historical default. It’s time for some definitions:

   1. Fully-Insured: The employer purchases an insurance policy from major carrier (Blue Cross, Aetna, Cigna, UnitedHealthcare). The employer and employees pay premiums to the carrier, and the carrier in turn assumes full financial risk for the employees’ medical claims.

   2. Self-Funded: The employer (not the insurance carrier) assumes the financial risk of paying medical claims directly.

   3. Level-Funding: This is a particular version of the self-funded model wherein the employer pays a fixed monthly amount (akin to a premium) in order to reduce volatility and make cash flow predictable. It’s popular among smaller employers. The fixed monthly payment incorporates funding for claims, stop-loss premiums, and administrative fees, and year-end reconciliation allows for refunds if healthcare costs are lower than expected.

   

Prior to 2012, the fully-insured model was the norm. But in 2025, roughly 117 million American workers receive health benefits through self-funded plans. (Only 60 million have fully-insured plans.) This fact is lost on most patients and most physicians. Employees see the logo of a major carrier on their insurance cards and assume that the carrier is the payer, when in actuality it’s their employer paying the bills (a macro-version of cash pay!). All of this begs the question - if self-funding has become dominant, why do insurance carriers remain such a fixture in American life?

This is where the Rube Goldberg machine comes into play. Paying for healthcare is not simply a matter of writing checks to healthcare providers. Here is a small sampling of the underlying administrative functions:

1. Member enrollment, eligibility, and management

2. Provider contracting and network management

3. Customer service

4. Prior authorization / utilization management

5. Claims adjudication

6. Regulatory filings

7. Fraud and abuse oversight

Employers have minimal expertise in these functions, and so when they choose the self-funded path, they are forced to rely on an ecosystem of vendors to do these jobs for them. Let’s trace the supply chain from start to finish, and see if we can decode the Rube Goldberg machine for a self-funded employer:

2. THE THIRD-PARTY ADMINISTRATOR (TPA)

The first thing an employer does is hire a broker or consultant to act as their tour guide through the maze. Think of firms like Mercer, Lockton, Gallagher, or NFP. These firms help the employer design their health plans. The health plan is a legal entity that pays medical claims directly, with stop-loss insurance to protect against catastrophic risk.

The broker/consultant starts by picking a third-party administrator (TPA) to perform many of the administrative functions listed above. It should come as no surprise that many of the dominant TPAs are the major insurance carriers themselves. They’ve been doing this work for decades. But because they are not serving an insurance function in this self-funded context, and only an administrative one, they are referred to as… drum roll please… administrative services only (ASO) vendors. Here are a few:

1. UMR (subsidiary of UnitedHealthcare)

2. Meritain Health (subsidiary of Aetna)

3. Signature Administrators by Cigna

4. Anthem Blue Cross Blue Shield

There is also a buffet of independent TPAs that self-funded employers can choose from, with the help of the broker/consultant:

1. WebTPA

2. Allegiance Benefit Plan Management

3. Health Management Administrators (HMA)

4. AmeriBen

5. CoreSource

6. Marpai

With the TPA established, now we can follow the chain reaction that’s triggered when a patient (employee) - we’ll call her Maria - wants to go to the doctor’s office.

3. ENROLLMENT

The story begins well before the doctor’s visit, when Maria is hired (or during open enrollment, or a qualifying life event). Her employer will use a vendor (Workday, ADP, UKG, Rippling, Employee Navigator, Health Sherpa, etc.) through which Maria provides her information and selections. The enrollment system then exports this data into the X12 834 format (more on this soon), which includes details such as: relationships/dependents, employee identifiers, dates of coverage, coverage types, etc. The X12 834 file is sent to the TPA, which loads the information into its own eligibility system. Maria’s profile is thereby created or updated in a core “claims administration” platform such as Facets, QNXT, HealthEdge, or Plexis… which the TPA pays software licensing fees to use. Once this is successful, the TPA sends the ACK (X12 834 Acknowledgement) back to the employer.

4. ELIGIBILITY, EDI…

When Maria shows up to the office, the front desk receptionist asks for her insurance card. The receptionist takes this information and sends out an X12 270 Eligibility Inquiry. This elicits a X12 271 Eligibility Response, confirming that Maria’s coverage is active, and listing any deductibles, co-insurance, and benefits.

Clearly, we need more definitions. Electronic Data Interchange (EDI) is an established system for computer-to-computer exchanges of business documents. X12 is a set of standards and messaging formats that is used for EDI in healthcare and other domains. You can think of X12 as a family of “electronic envelopes” that define all the major administrative transactions in this supply chain.

The problem is that these X12 files are often plagued by formatting errors and can often be sent to the wrong destination. For this reason, the “clearinghouse” came into being. Clearinghouses such as Availity, Change Healthcare, Trizetto, and Waystar conduct document “hygiene”, translating and routing X12 documents to the correct place. So in this step, the X12 270 Eligibility Inquiry goes to the TPA through a clearinghouse, and then the X12 271 Eligibility Response goes back to the doctor’s office through the clearinghouse. (Surprisingly, many independent TPAs don’t actually have the ability to support transactions such as X12 270/271, and therefore outsource them to third parties who do the work manually.)

5. PRIOR AUTHORIZATION

Maria sees the doctor and is told she needs an MRI. This requires prior authorization - in other words, medical necessity must be confirmed. The doctor’s office sends an X12 278 Authorization Request through the clearinghouse to a Utilization Management (UM) vendor. Contrary to popular belief, prior authorization is not always managed by the insurance carriers themselves, although they often have subsidiaries that do this work for them. Examples of UM vendors are:

1. eviCore

2. AIM Specialty Health

3. NIA Magellan

Often times, the UM vendor requires “supporting documentation” to make their determinations (clinical notes, imaging studies, lab results). The doctor’s office will send this information through X12 275 Attachments to a secure portal. If the MRI is approved by the UM vendor, the doctor’s office receives a X12 278 Response with an authorization number.

6. THE “CLAIM”

Once Maria completes her MRI, the doctor’s office takes this “billable event” and generates the appropriate diagnostic (ICD-10) and procedure (CPT) codes. The billing system then generates an electronic “claim” in X12 837 format and sends this to the clearinghouse. The clearinghouse checks syntax, ensures the payer IDs are valid, and then forwards the claim to the appropriate TPA.

The TPA then issues back a X12 997 Functional Acknowledgement, which serves as a digital stamp documenting that the claim was received. The TPA then edits the X12 837 claim file, checks for valid codes, verifies eligibility, and rule out duplicate submissions. If anything is missing, the TPA sends a rejection back to the provider.

7. THE REPRICER

Now the problem is one of price. The doctor’s office sends out a “billed charge” for Maria’s completed MRI, but this will be much higher than - and completely independent of - the pre-negotiated, contracted rate for this service. Enter the “repricer”. This entity’s job is simply to translate the inflated “billed charge” into the “allowed amount” that the employer’s health plan will actually pay.

Repricing sometimes occurs within the TPA, but it’s often outsourced to separate vendors such as: ClaimsBridge, Multiplan, Zelis, or Valenz Health. For “in-network” claims, the repricer applies the negotiated network rate for the MRI. For an “out-of-network” claim, the repricer uses various benchmarks such as percentage of Medicare or a proprietary “reference-based” algorithm. Regardless of the specifics, the output from the repricer is the “allowed amount”. This is what is actually owed.

8. CLAIMS ADJUDICATION

Now that the allowed amount has been established, the TPA must “adjudicate” the claim. This is the “rule engine” for the plan. Adjudication answers questions such as:

1. Was Maria enrolled in the health plan on the date of service?

2. Has Maria’s deductible been met?

3. Is the MRI covered under Maria’s particular plan?

4. Has the claim been matched to the correct benefit plan and benefit tier?

5. Are there coordination-of-benefit provisions across multiple plans?

6. How much does Maria owe (deductible, copay, coinsurance)?

7. How much does Maria’s employer owe?

Claims adjudication ultimately generates two outputs:

1. Explanation of Benefits (EOB) - for Maria

2. Explanation of Reimbursement (EOR) - for the doctor’s office

The EOB is sent to Maria, and explains what the doctor’s office charged, what her health plan is covering, and how much she may personally owe. The EOR, which also takes the form of an X12 835 Electronic Remittance Advice (ERA), is sent to the doctor’s office. The EOR is the printed, human-readable document. The ERA (835 file) is the electronic, standardized version. The EOR explains to the doctor’s office how much the health plan is paying, the rationale, and any adjustments that were made.

Hundreds of rules can be applied when the TPA adjudicates a claim. Whenever any part of the claim is adjusted, reduced, or denied, a reason must be provided in the EOR that is sent to the doctor’s office. Instead of providing “free text” explanations, standardized code sets are used:

1. Claims Adjustment Reason Codes (CARCs)

2. Remittance Advice Remark Codes (RARCs)

The CARCs explain payment that was denied or reduced for a claim. The RARCs are textual clarifications that provide additional context:

9. PAYMENT (finally)

Only now, once claims adjudication is complete, do any dollars change hands. The TPA/health plan sends payment to the doctor’s office, usually electronically, via ACH or EFT. The aforementioned X12 835 ERA accompanies this payment, serving as a digital receipt. The revenue cycle management (RCM) system for the doctor’s office, which is usually an outsourced company such as R1, athenahealth, or Optum360, then posts the payment and adjusts the account balance. If Maria owes a portion, the doctor’s office will issue her a statement (separate from the EOB, which is not a bill).

Payments from the TPA/health plan can lag by 1-4 months. The Accounts Receivable (AR) team for the doctor’s office therefore has to track the status of payments by issuing X12 Claim Status Inquiries and waiting for X12 277 Responses.

10. PAYMENT INTEGRITY

The story is still not over. Yet another set of “payment integrity” vendors, such as Cotiviti, Optum, HealthEdge, Source, and Verisys, retroactively review paid claims on behalf of the employer/health plan. The aim is to identify any over-payments, duplicate payments, or incorrect coding. These audits can take place days or months after payment. Numerous payment integrity vendors, like many other vendors in the supply chain, keep a percentage of “recovered” dollars. This adds friction and uncertainty for the doctor’s office. Even if a payment was received, it can be undone.

11. STOP LOSS

Each month, the TPA generates a report for Maria’s employer with the following types of information: total claims paid, trends by category, large claimants, and performance metrics. This helps the employer monitor cash flow and risk. When a paid claim exceeds a certain financial threshold, the TPA submits documentation to a stop-loss insurance carrier. (The self-funded employer’s consultant/broker would have helped select this carrier.) The submitted documentation will include the detailed claims file and supporting clinical documentation (remember the X12 275 Attachments?). The stop-loss carrier then reimburses the employer/health plan for the amount above the pre-determined threshold.

OUTSOURCING AND CHAOS

This supply chain is long and arduous, but that’s not the entire story. TPAs, including most ASOs, are forced to outsource major links in the chain to subcontractors. The irony is deep - ASOs charge per-employee-per-month administrative fees, and then turn around and pay a vast array of vendors to accomplish the necessary tasks:

Each one of these vendors charges the ASO/TPA, which cuts into its margins. The ASO/TPA therefore upcharges the employer even more to make up the difference, creating a toxic cycle of ever-increasing waste. What’s worse, ASOs often move healthcare data using archaic SFTP (secure file transfer protocol) file drops, rather than modern APIs (application programmable interface). SFTP is not real-time, requires manual monitoring, requires matching file-naming conventions, has unreliable success notifications, and has NO end-to-end transaction ID across vendor boundaries. You can imagine the chaos that routinely ensues. Claims bounce between siloed systems with no shared log, no shared ID, and no real-time middleware. Claims invariably get stuck in holding queues, delayed for review, pended for “reconciliation”, or lost entirely. No wonder processing timelines can be on the order of months.

Physicians are endlessly frustrated by this. Even when a claim is ultimately processed, it’s difficult to confirm whether it was done appropriately. What if an employer rents access to Cigna’s network, simultaneously has access to a one-off, high-value network for specific services, and also pays cash in certain scenarios? It’s nearly impossible for providers to know which benefit was applied because the information they receive doesn’t make it clear. And meanwhile, how much money are employers wasting on this dysfunctional administrative mess? See the table below. The Institute of Medicine, among many others, estimate that administrative waste accounts for at least 30% of total healthcare spending

WHAT NOW?

The supply chain of healthcare transactions is exhaustingly long. It’s filled to the brim with extractive intermediaries. It’s almost as if the system were designed to invite maximum highway-robbery in the space between healthcare providers and patients/employers.

We are seeing glimmers of hope, whether through the rising tide of cost transparency or the movement towards “direct contracting”. Mark Cuban’s newly unfurled Cost Plus Wellness (CPW), for example, is a welcome and courageous disruption. The ability to easily search for itemized healthcare prices, and for providers to get paid as one lump-sum within a month, is a clear step forward. No doctor’s office wants to chase after the Maria’s of the world for the portion they owe, nor do they want to write off that portion as “bad debt”. But all of these efforts, while creative and refreshing, fail to make a full frontal assault on the core problem of cost inflation.

Initiatives like CPW rightly encourage self-funded employers to seek out independent TPAs rather than sticking with the ASOs and their comprehensively bundled services. When employers do this, however, they must then assemble an array of unbundled offerings for all the components of the supply chain. New provider networks have emerged as alternatives to the traditional PPOs (preferred provider organization), but most independent TPAs don’t have the capability to manage multiple network offerings within a single plan. As often proclaims, it’s akin to “playing a new game on an old gameboard.” We need a new gameboard.

True waste elimination requires a complete re-tooling and digital distillation of the entire healthcare supply chain. True cost containment and affordability demand the creative destruction of the Rube Goldberg machine. We need an elegant, seamless-through-software, and completely reimagined infrastructure for healthcare transactions. Anything short of this simply misses the mark.

Everyone agrees costs are rising.

But ask WHO is actually paying those costs… and suddenly everyone starts pointing at everyone else like the Spiderman meme.

And, the next thing you know … healthcare looks a lot like the tariff situation!

👉 Lots of middle-‘men’.
👉 Lots of hidden fees.
👉 Lots of “not us —> must be them.”

And somehow the bill always ends up in the hands of the PATIENT (and increasingly, the clinician).

Tariffs work the same way.
We hear, “Don’t worry, these won’t hit consumers.”
And then your grocery bill gently taps you on the shoulder like:
“Hi, yes they did.”

Healthcare isn’t any different.
💰 Costs move through distributors → payors → systems → suppliers → policies → delays → denials → staffing gaps… and by the end?

Patients feel it.
Clinicians feel it.
Everyone else?
Mostly fine.

❗ Rising costs + opaque pathways = pain for the only two stakeholders who can’t pass the buck. ❗

Can you guess who? YOU are likely both of them at any given time :/

The danger is that we’ve built a system where compassion is quietly devalued and complexity is quietly punished.
✅ We reward volume, not outcomes.
✅ We track procedures, not overall improvement
✅ We celebrate efficiency while drowning in administrative burdens (I still have so many messages to respond to tonight…)

But here’s the hope:
We can fix this the same way we have been aiming to accomplish in spine care … one responsible step at a time! (A common theme of my writing…)
1️⃣ Responsible innovation.
2️⃣ Pathways that reduce unnecessary variation.
3️⃣ Better cost accounting (hello TDABC).
4️⃣ Cross-stakeholder conversations that actually include the clinicians doing the work.
5️⃣ Transparency around outcomes and costs; not just the ones that fit a reimbursement code.
6️⃣ Investment in systems that reward value, not volume.
7️⃣ Partnerships that realign incentives instead of multiplying intermediaries.

Because if we don’t redesign the system…
The system will continue redesigning us ❗
💵 And if tariffs have taught us anything, it’s this:
💵 Hidden costs always show up eventually!
💵 And when they do, the people at the end of the chain pay the highest price.

Love diving into my 4 E’s in spine/health care:
EDUCATION - ENABLING TECHNOLOGY - ECONOMICS — ergonomics (I’m standing to write this)

Leave a comment

Subscribe now

Tariffs.
Inflation.
Labor shortages.
AI disruption.
Market corrections.
Shutdowns -> And now … no data…

I feel like every headline this year has been an economic shock, really testing the resilience of our robust economy.

But as a spine surgeon, I can’t help but feel we’re living through the same volatility; just with different currencies.
✅ Our markets are ORs and clinics.
✅ Our assets are time, skill, and ability to innovate.
✅ Our returns are measured in outcomes (and profits still, yes…)
✅ And our margins are narrowing… everywhere!!

When tariffs ripple through global trade, manufacturers pivot and supply chains tighten.
We all have a front row seat to this event right now…

When reimbursement cuts and staffing shortages ripple through healthcare, spine surgery does the same.
Except our “GOODS” are PEOPLE and our “INVENTORY” is ACCESS TO THE RIGHT CARE (at the right time).

I look around and I see that…
1️⃣ Hospitals are closing service lines.
2️⃣ Private practices are consolidating.
3️⃣ Enabling technologies are advancing faster than reimbursement can keep up.
4️⃣ And every new device or platform now faces the same question an investor asks of a shaky market: Where’s the real return on value?
5️⃣ And on top of providing the best care that I can, I have to think through what my value to the system is.

In my day-to-day practice, I feel this pressure acutely.
The cost of innovation has risen, while the price of doing nothing (burnout, inequity, scraping by, pushing indications, mentors leaving) has become unsustainable.
**We’ve seen “interest rate hikes” in the form of increased administrative burdens.
**And “inflation” in the form of rising expectations: better outcomes, faster recovery, fewer resources.

I certainly do not have all the answers but our Value Science Research team has been working to better measure, model, and predict these healthcare “market cycles.”

⭐ I call it trying to build “RESILIENT VALUE”!
Not just measured in dollars, but also in outcomes, safety, sustainability, reproducibility across systems and practices.

👋 If you’re exploring similar ideas or want to collaborate on meaningful change in spine and MSK care, let’s connect!

For some people, currency is capital.
For some of us, its also patient care!

Leave a comment

Subscribe now