My previous post covered the site-of-service premium and what the CY2026 OPPS final rule does to it — the repricing of drug administration fees at grandfathered off-campus provider-based departments that compresses facility fee revenue on the non-drug service side of the hospital income statement. This post adds a second, unrelated compression on the same line: the 340B conversion factor offset, which arrived through a completely different statutory channel and hits a broader set of hospitals. Together, they are restructuring the revenue architecture of hospital outpatient care in ways neither was designed to create. This post explains how the offset works, who it hits, and why the CY2027 rate decision is worth modeling now.

What Is the 340B Program?

The 340B Drug Pricing Program was created by Congress in 1992 under Section 340B of the Public Health Service Act. Its design is a cross-subsidy: drug manufacturers are required to sell covered outpatient drugs to qualifying “covered entities” at ceiling prices — a statutory maximum set at average manufacturer price (AMP) minus a unit rebate amount, under a formula that differs for branded and generic drugs. Covered entities purchase at that discounted price and bill Medicare at the standard Part B rate of ASP+6%. The spread between acquisition cost and reimbursement — the 340B margin — flows to the covered entity. The subsidy is not a federal appropriation; it is effectively a transfer from manufacturers to safety-net providers, with Medicare as the billing vehicle.

Qualifying covered entity categories include disproportionate share hospitals, children’s hospitals, critical access hospitals, federally qualified health centers, rural referral centers, and freestanding cancer hospitals. The program was designed to let these providers stretch limited resources further — serving low-income and uninsured patients at a lower net drug cost than the open market would allow.

The program’s scale has grown well beyond what Congress contemplated in 1992. Total covered entity drug purchases under 340B reached $81.4 billion in 2024, with disproportionate share hospitals alone accounting for $64.1 billion. Specialty medicines — primarily oncology and immunology — drove most of that growth, consistent with national drug spending trends. But the growth is not only about drug prices. As hospitals acquire physician practices and enroll them as provider-based departments — the same dynamic described in my previous post — those sites can qualify under the acquiring hospital’s covered entity status, extending 340B purchasing eligibility to outpatient settings that may serve a materially different patient population than the safety-net mission intended. The result is a program whose financial footprint has grown substantially faster than its target population, and that gap is the source of most of the sustained policy friction.

The 340B margin is meaningful wherever high-cost outpatient drugs are administered — oncology, rheumatology, ophthalmology, neurology. The economic logic follows the drug cost, not the disease category: four of the top ten 340B drugs by 2024 purchase volume are oncology agents, but the other six span HIV, multiple sclerosis, and immunology. I work in oncology, and it represents the largest single chunk of the program — pembrolizumab alone was $8.16 billion in 340B purchases in 2024, more than any other drug and not particularly close.

Medicare’s Failed Attempted Fix

Starting in calendar year 2018, CMS reduced the Medicare Part B payment rate for 340B-acquired drugs from ASP+6% to ASP-22.5%. The rationale was direct: 340B hospitals acquire drugs at prices well below market, their actual acquisition cost is substantially below what ASP+6% represents, and the spread constitutes windfall revenue rather than necessary program support. CMS argued it had authority under the OPPS statute to set payment rates for separately payable drugs using data other than ASP.

The hospital industry challenged the cut. In American Hospital Association v. Becerra, decided June 2022, the Supreme Court reversed CMS unanimously. The statutory problem was precise: the OPPS statute permits CMS to deviate from the ASP+6% default only under two conditions — either pay ASP+6%, or conduct a survey of hospitals’ actual acquisition costs and set rates based on that survey. CMS had done neither. It had estimated acquisition costs without the required survey and imposed a cut on that basis. Nine justices agreed that was impermissible — not a close call.

Five years of below-statutory payments left a $10.6 billion hole. Of that, $1.6 billion had already been recovered through claims reprocessing after the policy was vacated in late 2022 — leaving a net underpayment of $9.0 billion that CMS was obligated to remedy.

The Remedy Structure: The Conversion Factor Offset

Congress had not appropriated funds for this purpose. CMS had to design a remedy within its existing OPPS authority and recover the cost in a way that preserved budget neutrality. The CMS 340B Final Remedy Rule (CMS-1793-F), issued November 2, 2023, established two tracks.

• The remedy payment: CMS instructed Medicare Administrative Contractors to disburse lump sum payments to approximately 1,700 affected 340B covered entity hospitals starting January 8, 2024, with all payments expected no later than February 7, 2024. No interest was included — CMS determined it lacked statutory authority to pay it. Non-340B hospitals received no remedy payment, because they were never subject to the ASP-22.5% cut.

• The cost recovery mechanism — the conversion factor offset: To fund the remedy without exceeding OPPS budget authority, CMS reduces the OPPS conversion factor for non-drug items and services by 0.5% per year. The conversion factor is the multiplier that translates APC relative weights into dollar payment amounts for every non-drug, non-device OPPS service. The 0.5% annual reduction applies to all OPPS hospitals enrolled before January 1, 2018, and continues until the cumulative reduction equals $7.8 billion — a timeline that runs approximately through 2041 at the current rate.

The Asymmetry Worth Underscoring

The cost recovery mechanism is broad by design: 0.5% applies to every pre-2018 OPPS hospital’s non-drug service revenue regardless of 340B status. A 340B hospital absorbs that annual reduction but also received the lump sum remedy; whether the net is positive or negative depends on each hospital’s drug volume relative to its non-drug OPPS revenue mix. A non-340B hospital absorbs the offset with no remedy — a straight reduction to non-drug service revenue as the cost of remedying a wrong it did not participate in. This potential compounds with the site-neutral repricing from my previous post, which lands on the same income statement line through an entirely different statutory channel.

CY2027: The Rate, the Survey, and What Comes Next

The rate

CMS held the CF offset at 0.5% for CY2026. In the proposed rule, CMS had floated 2%; the final rule deferred that increase but stated explicitly that 2% would likely begin in CY2027. The difference between the two rates is not incremental. At 0.5% annually, a program reaches 2.5% cumulative compression after five years. At 2%, it reaches 10% in the same window — a planning problem of a different order.

The survey

The CY2026 final rule did something beyond setting the offset rate. Acting on a direct instruction from President Trump’s April 2025 executive order on drug pricing, CMS introduced the OPPS Drug Acquisition Cost Survey — a mandatory data collection requiring covered entities to report net acquisition cost, inclusive of all rebates and discounts, for specified covered outpatient drugs purchased between July 1, 2024 and June 30, 2025. CMS has stated the results will inform CY2027 rulemaking. This is the first OPPS rulemaking cycle to include an actual acquisition cost survey. Every prior year’s rate-setting — including the 2018 cut — proceeded without one.

That distinction matters because of what the Supreme Court established in AHA v. Becerra. The Court did not prohibit CMS from cutting 340B drug payment rates. It said CMS could only deviate from the statutory ASP+6% default after conducting exactly this kind of survey. CMS lacked that predicate in 2018; the Court blocked the cut on that basis alone. The survey now closes the statutory gap. Once CMS has the data, it has both the legal authority and — given the executive order and the explicit CMS commentary in the CY2026 final rule — the stated policy intent to act on it. The CY2026 rule, in other words, did not just set a rate. It built the foundation for a larger one.

Non-response is not a neutral option. CMS indicated it will impute the lowest acquisition cost reported among similar responding hospitals for any non-respondent — making declining to participate an actively adverse choice rather than a way to stay out of the data.

What comes next

The CY2027 proposed rule, typically released in late July, will be the first rule with both the legal authority and the empirical basis to reprice 340B drug payments. The CF offset rate — 0.5% or 2% — is one decision. What CMS does with the survey data on drug payment rates is another, potentially larger one. For programs with significant 340B drug volume, neither is a routine regulatory filing.

A Program in Transformation: The Rebate Model Debate

The acquisition cost survey is one front in a broader debate about whether the 340B program’s fundamental mechanics should be restructured.

The manufacturers claim that the upfront discount model is poorly targeted, prone to duplicate discounts where both 340B and Medicaid best price apply to the same dispensing, and has scaled into a program whose financial benefits are increasingly detached from patient need. They propose a rebate model where covered entities would receive an upfront payment at WAC, submit claims data documenting eligible dispensing, and receive manufacturer rebates based on verified patient-level attribution. The mechanism is designed to make eligibility visible and eliminate the conditions that produce duplicate discounts.

The hospital argue thatthe rebate model shifts cash flow risk onto entities that currently depend on upfront discounts to fund operations, adds administrative friction, and is structurally designed to shrink the program rather than reform it. The American Hospital Association argued in its April 2026 comment letter to HRSA that the switch would do “serious, irreparable harm to 340B hospitals and the patients they serve.”

The Trump administration launched a 340B Rebate Model Pilot Program in 2025 covering a limited set of drugs from eight manufacturers. That pilot was blocked by a federal court in February 2026 in American Hospital Association v. Kennedy, and HRSA restarted the process with a new request for information whose comment period closed April 20, 2026. The litigation has not resolved the underlying question; it has deferred it.

What this adds up to: the 340B program is simultaneously under pressure on three fronts — the CF offset compressing non-drug OPPS revenue, a likely CY2027 attempt to reprice 340B drug payments on the basis of the new survey, and a structural reform debate over the program’s discount mechanics that is actively in litigation. A covered entity with a significant drug program cannot treat any of these as routine.

Good Intentions, Complicated Consequences

Both policies in this series started from defensible premises. The site-of-service premium exists because Medicare built two billing tracks that reward care routing through hospital classification rather than clinical reality — site-neutral payment is the correction. The 340B program exists because safety-net providers need a structural cross-subsidy to serve populations that Medicare and Medicaid rates alone cannot adequately fund. The remedy is legally required because CMS overstepped its statutory authority. None of these are hard to defend in isolation.

The unintended consequences follow from the implementation. A safety-net subsidy scaled into an $81 billion procurement mechanism, with eligibility extended through the same acquisition dynamics that drove the site-of-service premium. A remedy financed by a broad-based offset that falls on every OPPS hospital regardless of whether they benefited. Blunt tools applied to structural design problems that the tools were not built to solve.

A more coherent system design would move toward a single reimbursement price for equivalent services regardless of facility classification, and toward a drug acquisition framework that links pricing to actual cost rather than preserving a margin structure that has grown beyond its original rationale. Neither is achievable through rulemaking alone. Both require Congressional action — the kind that has remained out of reach for decades and is unlikely to arrive soon. In its absence, CMS will keep working within the constraints of its existing authority: incremental, layered, and producing the kind of interacting policy changes that are genuinely difficult to track and price.

These changes arrive staggered, through different statutory channels, landing differently depending on site type, 340B status, and enrollment history. An MA plan on a three-year contract cycle, a VBC program benchmarked against prior-year FFS spend, a CFO building a five-year capital plan — none of them can model the combined effect without reliable, centralized data. That is where CMMI has a role beyond model design: building the data infrastructure and transparency that lets non-government institutions — actuaries, APM designers, operators — make data-informed decisions and meaningfully shape policy. Provider classification crosswalks, policy change logs, acquisition cost survey data made publicly available for independent analysis. The government sets the rules; democratizing access to the underlying data is what makes it possible for everyone else to keep up.

Key sources

• American Hospital Association v. Becerra, 596 U.S. 724 (2022)

• CMS 340B Final Remedy Rule (CMS-1793-F), November 2023

• CMS CY2026 OPPS/ASC Final Rule (CMS-1834-FC), November 2025

• Executive Order, Lowering Drug Prices By Once Again Putting Americans First, April 15, 2025

• Baker Donelson, Important Issues for 340B Providers in the OPPS Final Rule, December 2025

• HRSA, 2024 340B Covered Entity Purchases, December 2025

• Becker’s Hospital Review, Top 10 340B Drugs in 2024, 2025

• HRSA, 340B Rebate Model Pilot Program, updated March 2026

• CMS OPPS Drug Acquisition Cost Survey Draft Template (CMS-10931)

• American Hospital Association, Comment Letter on HRSA 340B Rebate Model RFI, April 20, 2026

This is the first in a series that uses two active Medicare payment changes — site-neutral payment expansion and the 340B drug payment offset — as a working illustration of a broader challenge: the need to follow, understand, and account for legislative and regulatory change in healthcare. Each policy is complex on its own. Together, they interact in ways that can produce adverse outcomes neither was designed to create. Operators of care delivery companies and architects of alternative payment models need to understand the implications and update their thinking as these changes unfold. We walk through the rate mechanics and the 340B program structure first, then show what both mean for episode cost and how they may affect value-based care arrangements. The analysis focuses on Medicare FFS — where the changes originate — with MA implications noted throughout.

Consider a Medicare patient receiving pembrolizumab (an immunotherapy to treat many types of cancer) at an infusion center two miles from the hospital that owns it. The drug is the same. The nurse is the same. The chair is the same. But because the infusion center is enrolled in Medicare as a hospital outpatient department, it bills a facility fee of $337 just for administering the infusion. At a freestanding oncology clinic with no hospital affiliation, there is no separate facility fee — the physician's all-in rate for the same service is $133. The drug cost is identical. That $204 gap exists entirely because of how the site is classified in Medicare's billing system — not what was done, not who did it.

This is the site-of-service premium. It is not a Medicare peculiarity, and it is not limited to drug administration — the same dynamic applies to E&M visits, imaging, laboratory services, and surgical procedures. Wherever a service can be billed under two different facility classifications, the hospital outpatient rate exceeds the physician office rate, often by a wide margin. It is a structural feature of how healthcare services are priced in the United States, and it exists across every payer. The incentives embedded in the billing structure reward routing care through higher-paying facility classifications — and the data reflects that. Arnold Ventures, citing CMS claims data, reports that between 2015 and 2021 chemotherapy administration billed in freestanding clinician offices declined by more than 14% while the same services in hospital outpatient departments increased by more than 21%.

The commercial version is larger and harder to fix. Milliman's 2025 benchmarking estimates commercial payers reimburse medical services at approximately 196% of fully loaded Medicare FFS rates nationally, with facility fees a primary driver in outpatient settings. Hospital systems with regional market dominance negotiate from structural strength — the credible threat to go out of network, consolidated physician workforces, and the absence of viable alternatives in many markets. Commercial facility fee premiums frequently exceed Medicare's differential and are sticky in ways regulated rates are not: changing them requires bilateral contract renegotiation on multi-year cycles, under conditions that favor hospital systems.

Even for traditional Medicare, meaningful reform requires Congress. Arnold Ventures estimates comprehensive site-neutral legislation would save Medicare more than $150 billion and reduce beneficiary costs by more than $90 billion — with 84% voter support across party lines. That scale of change is beyond what annual rulemaking can accomplish. But while Congress has not acted comprehensively, CMS has been making incremental progress within its existing authority: clinic visits at excepted off-campus provider-based departments were repriced to physician office-equivalent rates in 2019; drug administration services followed in the CY2026 OPPS final rule; on-campus hospital outpatient departments are now in scope for future rulemaking.

This post focuses on a specific set of drug administration codes under Medicare FFS, where the rate data is transparent and the policy changes are confirmed. The near-term impact on MA and commercial is limited — MA contracts renegotiate on their own cycles, and commercial contracts move even more slowly; neither automatically follows a CMS rule. But the analysis matters beyond traditional Medicare because of CMS’s benchmark role. Medicare FFS rates are the reference point around which MA contracts, Medicaid fee schedules, and commercial negotiations orient. When CMS reprices a service category, it shifts the anchor. Slowly, imperfectly — but it shifts.

Location, Location: How Medicare Pays for Outpatient Services

To understand where the gap comes from, it helps to know that Medicare pays for outpatient services through two completely separate rate systems depending on who owns the facility. We will use CPT 96413 — the most commonly billed drug administration code in outpatient oncology — as the running example throughout.

When a patient receives an infusion at a freestanding physician office — a clinic with no hospital affiliation — Medicare pays a single all-in rate set by the Physician Fee Schedule (PFS). That rate covers the physician's professional work, the nursing staff time, the supplies, and the overhead. There is no separate facility fee. For 96413 in 2026, that all-in rate is $133.

When the same patient receives the same infusion at a hospital outpatient department (HOPD) — whether a main campus infusion suite or an off-campus facility enrolled as part of the hospital — Medicare splits the payment across two separate billing tracks. The hospital bills a facility fee under the Outpatient Prospective Payment System (OPPS), covering the overhead of the encounter: the infusion chair, nursing staff, supplies, and facility operations. For 96413 at a main campus HOPD in 2026, that facility fee is $337. The physician also bills separately for their professional work, at the lower “facility” rate under the Physician Fee Schedule — approximately $26 — because the hospital, not the physician, is covering the overhead. Total Medicare payment for the HOPD encounter: roughly $363.

The gap between the two approaches: $363 total at the HOPD versus $133 at the freestanding office — a difference of roughly $230 for the same clinical service. The facility fee accounts for all of it. Site-neutral payment policy targets this gap specifically: it does not change the physician’s fee, and it does not change what Medicare pays for the drug. It changes only what Medicare pays the hospital for the overhead of administering the service.

The facility fee was designed to compensate hospitals for the genuine overhead of running outpatient departments — capital infrastructure, nursing and pharmacy staffing, compliance costs that freestanding offices don’t carry. That rationale has become harder to rationalize as hospital systems have been acquiring independent physician practices and operating them as outpatient departments and billing at hospital rates for encounters that are operationally unchanged from before the acquisition.

CMS has responded by trying to reflect this reality in how it pays — distinguishing between types of hospital outpatient sites and incrementally adjusting reimbursement to match the actual care delivery setting rather than just the billing classification. The approach is deliberate and layered: not a single across-the-board cut, but a series of targeted changes that treat different site types differently. That nuance is what makes this policy area genuinely complex to follow.

Three types of sites, three rate structures — and a decade of policy changes

Not all outpatient facilities billing under hospital rates are the same. Understanding the current policy requires distinguishing three types of sites, because the rules are different for each.

• Main campus HOPDs are the original hospital outpatient departments — physically on or adjacent to the main hospital building. These have always billed under OPPS and continue to do so. No site-neutral payment change has touched them. A patient receiving chemotherapy at a main campus cancer infusion center is still billed at the full $337 OPPS rate in 2026.

• Off-campus provider-based departments opened after November 2, 2015 are facilities that hospitals enrolled as outpatient departments after Congress drew a line in the ACA. These sites are paid at physician office-equivalent rates — the same $133 — regardless of their hospital affiliation. They never had access to the OPPS premium.

• Off-campus PBDs enrolled before November 2015 — the “grandfathered” or “excepted” sites — are where the active policy changes are happening. These facilities had been billing at full OPPS rates ($337 for 96413) since their enrollment. Congress grandfathered them in 2015 rather than immediately cutting their rates. CMS has been narrowing that premium in stages:

  • 2019: Clinic and office visits at these sites were repriced from the OPPS rate to the physician office rate. Drug administration and infusion services were left at OPPS rates.

  • 2026: Drug administration services at these same grandfathered off-campus sites were repriced to the physician office rate. For 96413, that is a reduction from $337 to $133 per infusion visit, effective January 1, 2026.

Main campus HOPDs remain unaffected. Rural sole community hospitals are exempt from the 2026 change. And these changes apply only to drug administration service codes — imaging, radiation oncology, and other service categories at the same grandfathered sites continue to bill at full OPPS rates.

What Do the Numbers Show?

The table below shows the rate structure for the most commonly billed drug administration codes in an oncology population. The three rate columns correspond to the three site types described above.

A few things in the table are worth noting:

1. The physician’s payment is the same across all settings — the entire rate differential is in the hospital facility fee, not physician compensation.

2. The premium is largest for lower-intensity services: hydration ($33 at a physician office, $217 at a hospital) and IV push ($38 vs. $217) carry higher ratios than chemotherapy itself, because Medicare’s hospital outpatient rate-setting groups these three codes into the same payment bucket regardless of their very different physician office rates.

3. The 2025-to-2026 increase in physician office rates (10–16%, well above the 3.3% annual update) reflects a CMS methodology change that increased non-facility practice expense allocations — the new site-neutral rate at grandfathered off-campus sites is consequently higher in 2026 than it would have been under prior methodology, which modestly offsets the reduction from losing OPPS rates.

Several add-on codes commonly billed alongside these — 96367 (each additional infusion hour), 96375 (sequential injection add-on), 96417 (sequential chemo add-on) — are excluded from the table because they generate no separate payment under hospital outpatient billing; the hospital facility fee covers them within the primary service rate. Under the physician fee schedule they are billed and paid separately.

What Does This Mean for Cost of Care?

Building from code to episode: for a stage IV NSCLC patient on carboplatin, pemetrexed, and pembrolizumab — the standard first-line regimen in the absence of actionable mutations — a 6-month outpatient episode typically involves roughly 7 infusion visits (4 induction cycles plus 3 maintenance pembrolizumab administrations). At the 2026 rate differential of $204 per visit for 96413 alone, that is approximately $1,430 in facility fee reduction across the episode — but only at sites subject to the repricing. On a full Part A + B + D episode that typically runs $60–70k for a stage IV lung cancer patient, $1,430 represents roughly 2% of total episode spending. Not a rounding error, and concentrated entirely on the administration side of the ledger.

That 2%, however, comes with an important qualifier: it applies only at grandfathered off-campus sites, and the realized cost reduction depends on the site mix within a given geography. A patient receiving the same regimen at a main campus HOPD, or at an NCI-designated cancer center on hospital grounds, sees none of it. The impact for any VBC program depends on what share of attributed infusion volume sits at affected sites — and the exceptions and carve-outs make this genuinely difficult to calculate. Main campus HOPDs are untouched. Rural sole community hospitals are exempt. Only a specific subset of grandfathered off-campus facilities is in scope. Most programs have not mapped their attributed infusion volume to this level of site granularity, which means the actual impact on their cost of care is an open question until they do.

At the system level, CMS estimates the repricing reduces OPPS spending by $290 million annually. That breaks into two effects worth separating: $220 million is Medicare program savings — taxpayer money recovered from a program that spends over $800 billion annually. The other $70 million is reduced beneficiary coinsurance — direct out-of-pocket relief for patients currently paying 20% of the OPPS rate at affected sites.

Where Do We Go from Here?

CMS’s incremental approach is consistent with how major Medicare payment changes move — bounded steps, prospective only, with carve-outs to limit acute disruption. That caution is understandable. But it has a cost worth naming: the site-of-service differential is a real and ongoing transfer from Medicare taxpayers, commercial premium payers, and patients paying 20% coinsurance on OPPS rates they have no visibility into. Arnold Ventures puts the full scope at $150 billion in available Medicare savings and $140 billion in commercial savings through comprehensive legislation. The $290 million from the CY2026 repricing is meaningful — and a fraction of what is possible. The direction is right. The pace is a policy choice.

For MA and commercial, the gap between FFS policy and market reality is wider still. MA facility agreements renegotiate on their own multi-year cycles; a rule effective January 1, 2026 does not reset a contract signed in 2024. For commercial, the timeline is longer and hospital systems negotiate from a stronger position. The policy moves the anchor. How far and how fast the rest of the market follows is not determined by the rule.

For VBC architects and operators, this is genuinely hard — and the complexity deserves acknowledgment rather than hand-waving. The policy changes are staggered across service categories and years. Their impact is geographically uneven, because grandfathered off-campus sites are not uniformly distributed across markets. They flow into FFS benchmarks with a lag. And MA contracts sit between the FFS signal and the actual cost of care, absorbing or delaying the impact depending on when they last renegotiated. None of this is insurmountable. But it does mean that measuring the effect on your program requires deliberate infrastructure: site-type classification of attributed providers, contract-level tracking of MA facility rates, and explicit scenario modeling for the benchmark timing gap — not a one-time analysis, but ongoing as the policy continues to evolve.

There is a broader opportunity here worth flagging. CMMI operates multiple oncology and specialty programs, each with its own benchmark methodology, data infrastructure, and measurement approach. A unified infrastructure — shared site-type crosswalks, common provider classification standards, a policy change log that flows directly into benchmark adjustment models — would make it faster to design, implement, and measure payment policy changes across all of them. Right now, each program and each operator builds bespoke systems to answer the same underlying questions. That fragmentation slows the feedback loop between policy intent and measured impact. Solving that problem would help accelerate the former.

Key sources

• CMS CY2026 OPPS/ASC Final Rule (CMS-1834-FC), November 2025

• CMS CY2026 OPPS/ASC Proposed Rule (CMS-1834-P), July 2025

• CMS CY2026 Physician Fee Schedule Final Rule (CMS-1832-F), October 2025

• CMS OPPS Addendum B (CY2026)

• CMS PFS National Payment Amount File (CY2025, CY2026)

• Milliman: Commercial Reimbursement Benchmarking Against Medicare FFS Rates, 2025

• Arnold Ventures: Site-Neutral Payment

My wife and daughter are on vacation this weekend, which means you are getting a second AI post in three days. If you missed the first one — You Can't Automate What You Haven't Defined, you can find it here. This is a good companion to it.

The pressure to adopt AI is everywhere right now. It comes up in every conversation with my friends and former colleagues in technical leadership roles: boards pushing harder, current and prospective investors watching closely, executive leadership asking for results. The pressure is rooted in something real and different in kind from previous technological shifts. While Cloud changed infrastructure and mobile changed distribution, AI is changing the production function of entire industries — what work gets done, by whom, and at what cost — and it is doing it faster than most organizations have ever had to adapt.

The mental model I keep returning to frames AI investment through three arguments — revenue, margin, and market multiple — each of which shapes how your company is ultimately valued. While the success of a company can be measured numerous ways, I focus on valuation deliberately. Just as a willing paying customer is the best proof of product-market fit, valuation is the clearest proof that you have built something the market believes in. When executives, board members, and functional leaders share that frame, it does something practical — it replaces the tension between AI zealots and skeptics with a more useful question: what actually creates value, and how does AI help us get there?

1. Revenue — Build the best product for your customers

Steve Jobs said it plainly:

“If you keep your eye on the profit, you’re going to skimp on the product. But if you focus on making really great products, then the profits will follow.”

The best use of AI is to know your customers more deeply and solve their problems more completely. The companies that get this right are not the ones chasing an AI adoption rate — they are the ones that stay closest to their customer, understand their problems truly and deeply, and use AI to solve them in ways that were not previously possible — better, faster, and tackling problems that were previously too complex or too costly to address at all.

In technology-enabled care delivery, this is not abstract. It means identifying patients with the highest needs before their conditions escalate — and reaching the ones whose trajectories are at inflection points, where timely intervention can change the outcome. Those are the moments where your care team can make a real difference. AI can make it possible to find those moments at scale, consistently, before they pass. It does not just improve how care is delivered — it changes what care can accomplish. And that is where durable value gets created.

The revenue that follows is the most durable kind: customers who pay, renew, and expand because your product is genuinely addressing a need and creating value for the people they serve. That is recurring revenue. But more than that — it is a moat. In technology-enabled care delivery models, that moat is the ability to care for every member at the right dose, at the right time, in a scalable delivery model. A moat built on that kind of deep operational and clinical understanding compounds differently than a feature moat. Understanding — the kind that accumulates over years of being close to your patient population, multiplied by AI’s ability to act on it at scale — is very hard to replicate. That is pricing power. That is the gap worth widening.

The companies that define this space will be the ones that used AI to widen it. Everything else in this framework depends on getting this argument right first.

2. Margin — Build a business that sustains itself

Revenue is the proof that customers value what you build. Margin and cash flow are the proof that you can sustain it. Every CFO understands this distinction (even dreams about it) — and AI makes it more achievable than ever. A cash-flow-positive business has options: raise on your own terms, exit when the timing is right, weather a downturn, or reinvest back into the product. You choose. That freedom is not accidental. It is built.

The reason AI has such a profound impact on margin is that it does not just make your organization more efficient — it fundamentally changes what your organization can produce. Let’s look at an illustrative toy model.

Traditional Cobb-Douglas

In the traditional Cobb-Douglas model (remember ECON 101), technology is a scalar A — a fixed multiplier that raises output proportionally across existing inputs. Better tools raise A slightly. Everyone gets marginally more productive. The ceiling moves, but it is still a ceiling.

AI as third input

AI introduces a third factor of production that breaks that ceiling entirely. The productivity of every person in your organization now scales directly with the AI capital deployed alongside them. Your best people are not replaced. They are multiplied. And unlike a traditional scalar improvement, this transformation works on both sides of your income statement simultaneously — AI expands your revenue potential by making your product better, and restructures your cost base by making your people more productive.

That simultaneity is what makes AI transformative rather than a traditional efficiency gain. A business expanding revenue and margins at the same time reaches sustainable cash flow faster — and from a stronger foundation.

3. Market Multiple — Real, but build to deserve it

Once you have built the business described in Arguments One and Two — durable revenue, expanding margins, sustainable cash flow — something else follows. The market notices.

For technology-enabled services organizations, being AI-forward carries a meaningful valuation premium — earned on two fronts. The first is the genuine transformative potential we described: better products, multiplied people, stronger margins. The second is simpler — the market is excited, capital is abundant, and narratives are moving multiples. Venture capital deployed into AI reached record levels last year, representing the majority of all global venture investment, and the largest technology companies are committing capital to AI infrastructure at a scale not seen since the early buildout of cloud. For any company raising capital or considering an exit, this tailwind is real and it would be a mistake to ignore it.

The important qualification is that a multiple is a multiplier. It amplifies the underlying value of the business it is applied to. A high multiple on strong, recurring, cash-flow-positive revenue is a genuinely powerful outcome. A high multiple on a business without those fundamentals is a favorable-looking number with an unstable foundation. The premium reflects value. It does not create it. And chasing a multiple without building the underlying business can be a costly mistake.

The right instinct is the opposite. Go after revenue by building the best product for your customers. Truly transform your production function with AI capital. Get to cash-flow positive. Do those things well and the multiple will follow — and it will not be a narrative-driven multiple that evaporates when sentiment shifts.

Remember that multiplier premiums are seasonal. The SaaS market between 2020 and 2022 is the clearest recent example. Companies raised at 15, 20, and 30 times revenue during a period of cheap capital and strong narrative tailwinds. When rates rose and the environment shifted, those same businesses were valued at a fraction of their peak. The founders who had prioritized durable revenue and reached cash-flow positive had choices — they could wait, reinvest, or transact on their own terms. Those who had optimized for the multiple found themselves without options, forced to transact under pressure or unable to transact at all. Cash flow bought time. Time bought choice.

The AI premium is real today. Its duration is not guaranteed. Build a business that captures the upside when the multiple is favorable and remains sound when it compresses.

Build something worth multiplying

The three arguments are designed to work together. Revenue without margin is fragile. Margin without revenue is hollow. And a multiple without either is a number that will not hold.

But here is the thing about this moment. The pressure is real, the capital is abundant, and the technology is genuinely transformative. That combination does not come around often. The question is not whether to invest in AI. The question is whether you are investing in a way that builds something durable — or just something that looks good in the current climate.

The best response to that pressure is not to signal AI-forwardness. It is to build the best product you possibly can, for the customers who need it most, as effectively and efficiently as AI will allow.

Build something worth multiplying.

Till next time,

Alphan

The thread that has stayed with me is something Sinofsky said about the nature of the work itself. He observed that algorithmic thinking — breaking work down into discrete, well-defined steps that can be executed consistently — is genuinely hard for the vast majority of people. Most organizations do not operate with that level of process clarity. They have loosely defined ownership, informal handoffs, and a “we’ll figure it out as we see it” approach to edge cases. This works, sort of, when you have experienced people who carry the process in their heads. But it does not scale — not with humans, and certainly not with machines. You cannot automate what you have not defined. And you cannot define what you have never been forced to think through.

That observation crystallized something I have been seeing across tech-enabled services organizations for years. The urgency to adopt AI is real and the opportunity is significant. But the organizations racing to deploy agents on top of undefined processes are going to struggle — not because the technology is not capable, but because they are skipping the foundational step that makes any of it work. That foundational step is the process diagram.

1. An Enforcement Mechanism for Crisp Thinking

The most important function of a process diagram is not documentation. It is the act of forcing people to think through the process with precision — when does something happen, what triggers it, who does it, what do they do it to, and what does the output need to look like for the next step to work. This sounds obvious. In practice, it is genuinely hard and most organizations never do it.

Think about a common operational task in a tech-enabled services organization: delivering an intervention or a campaign to a cohort of members. At face value, this seems straightforward. You identify the population, you deliver the intervention, you track the outcome. But the moment you sit down to actually write the process — to define every state, every decision point, every handoff — the complexity surfaces. Who is included? What are the exclusion criteria, and when are they evaluated? What happens if a member meets inclusion criteria today but an exclusion criterion triggers tomorrow? How do you handle recurrence? What does the handoff to the next step look like, and what data needs to travel with it? What is the definition of “done” for this step?

These are not edge cases. They are the process. And until you have worked through them explicitly, your organization is not really operating a process — it is operating a loose approximation of one that depends on the judgment of a few experienced people to hold it together. That works until it does not: until volume grows, until those people leave, until you try to train someone new, or until you try to hand any part of it to an agent.

The discipline of drawing the process diagram is what forces this thinking to happen. It is the enforcement mechanism — not for compliance, but for clarity. It requires the organization to agree, in explicit terms, on how work actually moves. The artifact that comes out the other end is valuable. But the process of creating it is where the real work happens.

2. A Diagnostic Instrument — and a Measurement Framework

Once you have a process diagram, you gain the ability to diagnose problems rather than just observe symptoms. This is the second function, and it follows directly from the first. You can only diagnose what you have defined.

The most common failure mode in operational organizations is that you know something is not working but you cannot locate the break. Volume is dropping. Quality is inconsistent. Certain outcomes are not materializing. Without a defined process, you are tracing a path that only exists in people’s memories. A well-drawn process diagram changes this. When each step is explicit — inputs, outputs, decision criteria, handoff points, desired outcomes — you can measure against it. Where is volume falling off? Where are delays accumulating? Where are handoffs producing malformed inputs for the next step? The diagram gives you coordinates. You can see where things are dropping off, and you can trace back to understand why.

This diagnostic capability is also the prerequisite for meaningful measurement. I see this pattern frequently: organizations want KPIs and dashboards, they want to know if things are working, but they have not done the process design work first. The result is that the metrics they track are disconnected from the actual flow — they measure activity rather than the quality of outcomes at each step, and they do not have the data elements they would need to actually diagnose a problem. You cannot know what to measure if you have not defined what the process is supposed to produce at each stage. And you cannot ensure that your systems are capturing the right data if you have not mapped the workflow that generates it.

I have spoken about this directly in the context of building data and analytics frameworks — the point being that the framework only works if the underlying workflows and desired outcomes are clearly defined first. The data model follows the process model. When the process is articulated at the level of individual steps, diagnosis becomes tractable: you can decompose a problem, trace the data across the workflow, and identify with precision where something is breaking down and what data you would need to prove it.

3. A Platform for Modular Change

There is a third function that connects the first two to everything happening in AI right now. When your process is defined to the level of individual, well-scoped steps, each step becomes independently modifiable. You can change one component — improve it, replace it, automate it — while understanding the upstream inputs and downstream effects. This is the difference between surgical change and disruptive change, and in a complex operational environment with interdependent workflows, it matters enormously.

Without this decomposition, every modification carries unknown risk. You think you are improving the intake step; you do not realize you have broken how the downstream eligibility check receives its inputs. With a properly drawn state machine as your foundation, the scope of any change is clear. You understand exactly what you are touching, what depends on it, and what the logical flow looks like before and after the change. You can reason about the second-order effects before you make the first-order change.

This modular design discipline is not a new idea — it is a foundational principle of how good systems are built. What the current moment is surfacing is that it applies equally to operational processes as it does to software architecture. The organizations that have invested in this kind of process clarity are the ones that can make changes with confidence, can train people against a defined structure, and — when the time comes — can identify exactly where an agent can take on a step and what the boundaries of its ownership should be.

Now You Are Ready to Talk About AI Agents

The thesis of the AI adoption conversation right now is that agents will transform how organizations operate. I agree with that. But the practical question is not whether to adopt agents — it is whether your organization is ready to do it in a way that actually produces durable, measurable results. And that readiness depends almost entirely on whether you have done the process work first.

Organizations that are trying to deploy agents on top of loosely defined workflows are not going to get there. As Sinofsky pointed out, most organizations have not developed the algorithmic thinking required to define their work precisely enough for a human to execute consistently, let alone a machine. The agent is not the bottleneck. The undefined process is.

Once you have your process diagram, the path to incorporating agents becomes deliberate rather than speculative. You can look at each step and ask two concrete questions: what is the volume here, and what is the friction? High volume and high friction is your starting point. The ROI is arithmetic, not abstract — you know the step, the frequency, the cost in human time and error rate, and what good looks like at the output. You can measure an agent against that baseline from the first day you deploy it.

More importantly, you can introduce agents incrementally, in the same way you would expand the scope of a new team member. You do not give a new hire the entire operation on their first day. You give them a clearly scoped responsibility, define what success looks like for that scope, and measure their performance against it. Over time, as they prove reliable, you expand their ownership into adjacent steps. You maintain oversight of the edge cases until you are confident the agent can handle them. This is just good management — and it applies to agents precisely as it applies to people.

The inverse is also true: just as you cannot develop a person — give them real feedback, set meaningful expectations, help them grow — without a clear definition of their role and responsibilities, you cannot govern an agent without the same clarity. You will not know if it is succeeding. You will not know what to fix when it fails. You will not know when it is safe to expand its scope. The process diagram is the job description. Without it, you are not running an agentic workflow. You are running a black box and hoping for the best.

The Best Practices Have Not Changed

In the midst of all the excitement around AI and the rapid pace of change, it is easy to lose sight of the fundamentals. Careful process design, modular systems, defined ownership, measurement frameworks — these are not legacy ideas that the current moment has made obsolete. They are the foundation. And if anything, they matter more now than they ever have.

The reason is straightforward: agents execute at a speed and scale that amplifies both what is working and what is not. A poorly defined process that a skilled human can navigate through judgment and context becomes a systematic failure mode at machine speed. The edge cases that your most experienced people handle instinctively — and that never get written down — become the failure surface that your agents will hit repeatedly.

The organizations that will successfully adopt and expand AI are not the ones that moved fastest in the short term. They are the ones that invested in process clarity first — that took the time to draw the diagram, define the steps, instrument the workflows, and develop the organizational fluency to think algorithmically about their own operations. That work is harder than buying a tool or deploying a platform. It is also the work that cannot be skipped.

Till next time,

Alphan

What is different about Specialty Care?

Early VBC waves focused on primary care—for good reasons. Primary care offers frequent touchpoints that shape downstream utilization, and PCP panels are large and heterogeneous. Because those panels are effectively random samples of the total book, they tend to behave like the overall population on average. In that context, it made sense to emphasize contracts and ROI stories that target the mean: design an intervention, lower average cost, and use those savings to fund operations and share value.

If we’re serious about bending the cost curve, we have to bring that same discipline into specialty populations. In 2023, direct specialty spending in orthopedics, oncology, cardiology, women’s health, behavioral health, and nephrology accounted for 38% of total medical spending, and the total cost of care for people seeing specialists in those six areas accounted for 68% of commercial and Medicare spending. That is where the curve is steepest, and that is where the work must happen. Policy and capital are already shifting: the CMS Innovation Center has launched more than a dozen specialty‑focused models over the past decade and, in its 2025 strategy, called for expansion across Medicare Advantage and Medicaid with greater participation from independent specialists; private investors put more than $63B into specialty‑enablement companies in 2023. (Source: McKinsey; see Appendix.)

Yet only a small share of specialty care is delivered under risk‑bearing VBC. A major reason: we lack transparent, standard contract structures that move the right risks to the right party and pay specialty organizations appropriately. Copying primary‑care templates rarely works—specialty panels are smaller and their costs are dominated by tail events.

A short actuarial literature

Finance and insurance judge performance relative to risk, not only by expected return or expected cost. In markets that shows up as Sharpe/Sortino ratios (returns over volatility or downside volatility). In (re)insurance and health finance, we codify the same idea with tail‑sensitive measures and risk margins—we ask how bad the bad years get, quantify the capital needed for those years, and put a price on shifting that burden.

The most influential tool here is Expected Shortfall (ES)—also called Tail Value‑at‑Risk (TVaR). Where VaR identifies a cutoff (e.g., the 95th or 99.5th percentile), ES/TVaR averages the outcomes beyond that cutoff—the mean of the worst 5 (or 0.5) percent of years. ES is coherent (it behaves well under diversification) and is preferred when tails drive the story. Modern solvency frameworks operationalize this by translating tail capital into a risk margin with a cost‑of‑capital (CoC) charge (e.g., Solvency II’s published CoC; the Swiss Solvency Test builds capital directly on ES). The same logic carries neatly into payer–vendor contracts.

There are adjacent pricing principles worth knowing. Distortion (Choquet) premiums—like the Wang transform—tilt probabilities to put more weight in the tail and then take a distorted expectation; they’re practical as a sensitivity for non‑traded healthcare risks. Utility‑based approaches (e.g., the exponential premium and Esscher transforms) tilt the loss distribution according to risk aversion and capture more than variance—useful for cross‑checking how a different risk appetite would price the same layer.

The same ideas in health plan CFO/executive language

Think of risk‑adjusted pricing as a straight translation of a budget stress test into a PMPM.

1. Define the exposure you’re moving. Decide what the vendor will hold: either all specialty costs (a full carve‑out) or a risk corridor. In our simulation we use a dynamic corridor where, each year, the vendor holds the per‑member layer from the 5th to the 95th percentile of the specialty distribution. That removes “everyday big losses” while leaving very low and very high costs with the plan.

2. Measure how bad the bad years are. Compute ES/TVaR at a chosen confidence (e.g., 95% or 99%). ES answers: “If next year lands in the worst 5 (or 1) out of 100 years for this layer, what is the average dollar loss?” The difference TVaR − Expected is the tail capital the vendor must be able to carry.

3. Put a price on that capital. Apply a CoC to tail capital to get a risk margin. This mirrors solvency practice and turns “less tail for the payer” into a transparent dollar amount.

4. Build the vendor’s PMPM.

Divide by 12 x covered lives to get PMPM. You’re paying for three things: expected costs in the layer, operating expense, and the capital the vendor must hold to absorb bad‑year losses. That capital relief is why the payer’s results become steadier and risk‑adjusted return improves.

5. Use contract shape to control how much risk you move. A full carve‑out replaces a jagged, tail‑heavy specialty spend with a fixed payment, collapsing variance, skewness, kurtosis, and tail capital. A corridor trims the middle but keeps the super‑tail, so tail capital falls partially, not fully—hence the lower price.

A numerical example: PCP lens vs Specialty Care lens

Consider a population of 100,000 members. Annual costs are log‑normal—realistic for skewed medical spend. Ninety percent belong to a general cohort with a mean of $7,500 PMPY; ten percent belong to a specialty cohort with a mean of $120,000 PMPY and much heavier tails.

A per‑member distribution plot makes the contrast obvious. A PCP carve‑out that manages a random 10% panel behaves like the overall population (a random slice of the book). A specialty carve‑out concentrates the right‑tail of the distribution—exactly the costs that make plan‑level results volatile.

Fig 1: Visualization of the difference of a PCP cohort carve out vs the Specialty Care cohort carve out.

We repeat the simulation 2,000 times as a simple bootstrap. You could compute many of these statistics analytically under independence assumptions, but the bootstrap and visuals resonate with a broader audience.

In the summary table, two things stand out. First, the whole population and PCP 10% cohorts have nearly identical average per‑member costs; the PCP panel just has a larger standard error because it’s smaller. Second, when you remove the specialty cohort, you not only lower the average per‑member cost but also make the tail (p75/p95) far more predictable—exactly the point of moving higher moments off the payer’s books.

Table 1: Summary statistics from the bootstrapping exercise Summary statistics from the bootstrapping exercise.

Value of risk transfer

We then compare the payer’s retained exposure (claims only) under three scenarios: no contract, full carve‑out, and a p5–p95 corridor on the specialty cohort.

• Both risk transfers reduce the plan’s exposure, but the full carve‑out delivers the largest drop in standard deviation and tail capital (TVaR − mean).

• The p5–p95 corridor narrows volatility in the middle of the distribution yet leaves the super‑tail with the plan, so tail capital falls meaningfully but not maximally.

Using TVaR at 95%, a CoC of 8%, and $30 PMPM admin on the full book, the pricing equation cleanly decomposes into expected layer loss, risk margin (CoC × tail capital), and admin. That transparency lets plans decide how much capital relief they want to buy, and lets vendors decide how much risk is appropriate for their balance sheet.

Sidebar: Because specialty VBC is still early, the cost of capital for care‑delivery vendors is often higher than for large payers. That reality makes corridors an attractive on‑ramp: material stabilization for the plan at a price that’s feasible for vendor capital stacks.

Closing Thoughts: Pricing the Risk We’re Actually Trading

If specialty value‑based care is going to scale, we have to price the risk we’re actually trading. Payers don’t just benefit from lower average costs—they buy stability. Every time a specialty vendor assumes downside exposure, the payer’s variance, skewness, and tail risk shrink. That capital relief has enterprise value and deserves to be priced explicitly.

Vendors that shoulder specialty tails should be compensated for the higher‑moment risk they absorb, not just the mean they manage. An actuarial toolkit can be formally built out to commoditize the pricing and transfer of risk: TVaR‑based risk margins to value tail capital, and distortion‑premium sensitivities to reflect different appetites for tail risk.

This article is a conversation starter, not a full model; it intentionally sets aside messy real‑world complications—new therapies, market entry waves, legislation, and data lags. But the direction is clear: if we want specialty VBC to move from pilots to system‑level impact, we need to price risk transfer so the organizations holding it are appropriately compensated for the capital, volatility, and uncertainty they take on.

Appendix

Key sources to link for readers who want to go deeper

1. https://www.mckinsey.com/industries/healthcare/our-insights/specialty-risk-the-next-frontier-of-value-based-care

2. Artzner–Delbaen–Eber–Heath on coherent risk

3. Acerbi & Tasche on the coherence of Expected Shortfall

4. Wang on distortion premiums

5. Wang on two risk measures: VaR and Expected Shortfall

6. Denuit and the Loss Data Analytics text for exponential/Esscher principles

7. EIOPA for the cost‑of‑capital risk‑margin method in Solvency II

8. FINMA on the Swiss Solvency Test and ES‑based capital

Over the years, I have adopted the Lego blocks metaphor as an effective tool to visually represent product roadmaps for complex systems that empower healthcare organizations by helping make abstract concepts and multifaceted interdependencies more concrete. A Lego block diagram is a high-level visual representation of a system, breaking down complex workflows and dependencies into modular, distinct, understandable units.

Consider the simple illustrative example below (to keep me out of trouble!) for data and analytics pipelines for a value-based care organization managing multiple contracts across Commercial and Medicare Advantage populations. This is neither a formal system diagram nor a formal process diagram but rather an intuitive depiction that captures the bigger picture and makes it accessible to a diverse audience. Even though each contract may look different, they will all require the identification of the eligible population and its segmentation for enrollment and activation. Besides the overlapping components, we can also see the pieces that are unique to the Medicare populations. Your overall reaction may be, “This is simple!” Well, that is great! Writing down and defining the pieces is the hard part. Communicating what they are should be easy and intuitive. That is the goal.

Why Legos?

My choice of the Lego block metaphor ties back to the initial design goals of Lego’s creators:

1. Compatibility: Blocks were designed to have a universal connection mechanism, ensuring that any Lego piece could interlock securely with others.

2. Versatility and modularity: The bricks were designed to allow for maximum flexibility and versatility in building. They can be combined in countless ways, enabling children (and adults) to create anything from simple structures to highly complex models.

3. Simplicity: The original design philosophy emphasized simplicity. Lego  bricks are basic geometric shapes, making them intuitive for children to handle and build with. The clean, minimalistic design also supports endless creativity, as builders are not constrained by complex or overly specific parts.

4. Stimulating imagination, learning, and problem solving: The primary objective of Lego was to stimulate creativity and learning through play. The blocks were intended to encourage problem-solving, fine motor skills, and spatial awareness.

These are the goals I wanted to achieve in design and planning conversations while building systems and technology behind health tech companies.

What makes health tech organizations unique?

Unlike many other industries, where a Minimum Viable Product (or Minimum Desirable Product) can often be created with limited core functionality, healthcare systems demand comprehensive solutions that address multiple layers of compliance, safety, and end-to-end workflow requirements. Often, this can lead to an overwhelming feeling of not knowing where to start. The Lego blocks diagrams can help you identify where to start laying your bricks. (🥁) Moreover, as the complexity continues to grow, the Lego blocks approach provides a visual vocabulary that you can reference as you operate and maintain your systems.

The Lego block metaphor initially served as a personal tool for me to understand the complex systems being built by the teams I was leading. Over time, it evolved into a powerful means to facilitate a deeper understanding and alignment across the broader organization. Especially for tech enabled healthcare service organizations, the core leadership team may include folks who have never worked within a technology organization and did not have to make longer term tech investment decisions. Hence, visual representations of your systems and roadmap help engage them as customers of the products and systems you will be building as well as key stakeholders who hold the money purse to get the budget for the investments you may want to make.

Why use Lego blocks over more traditional tools?

Deliberate and Explicit Decision-Making on Resource Allocation and Investments

The Lego blocks approach begins by defining each piece of the system, thereby enforcing a deeper understanding and establishing a common language among cross-functional teams. Yes, this can be achieved in spreadsheets, but I strongly believe that the visual representation provides a significantly more tangible view and better understanding of interdependencies and nuanced layers.

Moreover, Lego blocks also clearly and explicitly indicate what blocks have active versus deferred investments. Spreadsheets and product management tools often focus on what is “in scope” but do a poor job of showing what has been deliberately and intentionally deferred. Keeping all blocks on the page helps avoid the classic "Have we considered XYZ?" confusion while allowing us to re-visit and reconsider prioritizations. Your team maintains its overall blueprint to ensure continuity in thinking and decision-making.

Zoom In and Out: A Macro-Micro Lens for Complex Systems

In the process of building complex systems, it’s easy to lose sight of the broader landscape. The Lego blocks approach allows the team to zoom in and out seamlessly. Traditional roadmap planning approaches often do not allow for top-down or bottom-up traversing of your roadmap and systems in an intuitive way.

Each block can be broken down into smaller units that fit together to form a larger structure. This flexibility enables stakeholders to gain a detailed understanding of specific components while preserving the context of the broader system. Let’s assume your Chief Medical Officer wants to examine the Utilization box from above. You can illustrate the building blocks of this particular box with finer granularity and allow an executive domain expert to traverse the Lego blocks top-down (macro to micro).

The Lego block diagram can be used to establish ownership and accountability areas for individual pods within your tech team. Each pod can be mandated to draw the zoomed-in breakdown for its ownership block to manage resourcing and prioritization discussions. This will foster autonomy, and each team can then traverse the Lego blocks bottom-up (micro to macro) to maintain alignment across the broader tech organization and a cohesive system architecture.

There is an added benefit specific to data teams. As data analysis begets more questions and further analysis, it’s common for teams to lose track of the lineage of data questions and their connection to business goals. The Lego block diagram provides a mechanism for re-orientation, allowing you to start by answering foundational questions and identifying gaps in your knowledge graph.

Modularity and Abstraction: Building for Long-Term Scalability

Components of your systems will need to be changed, upgraded, or replaced as competitors emerge, regulations change, and new technologies are introduced. Having a visual representation of your system as individual building blocks encourages modular design to achieve adaptability and long-term scalability. Note that we have already mentioned the clear ownership and autonomy in the previous section, which ties modular design to the scalability of your organizational design as well.

Postlude

The Lego blocks approach has been a valuable tool for me to build my mental models and help communicate them to my teams and stakeholders. Hopefully, it will help you construct your mental models and think about the foundational pieces. Ultimately, I hope it will help build better products and more robust systems to bring innovation to the healthcare tech space.

This blog1 post has been simmering in the back of my mind for a while. After listening to Giuseppe Paleologo (aka Gappy)'s appearance on Flirting with Models and hearing his nuanced and detailed description of the quantitative research and risk roles in buy-side firms, I decided to put pen to paper. Thanks again, Gappy!

The healthcare landscape has undergone a significant transformation in recent years, moving away from traditional fee-for-service models towards more innovative approaches that prioritize patient outcomes and cost-effectiveness. This evolution has given rise to two interconnected concepts: value-based care and risk-bearing entities. Value-based care emerged as a model focused on quality outcomes and cost reduction, while risk-bearing entities developed as organizations willing to assume financial responsibility for healthcare costs. More specifically, these approaches have reshaped how healthcare is delivered and financed, encouraging a more coordinated, efficient, and quality-driven system by shifting financial responsibility from health insurance companies and government programs (Medicaid and Medicare) to provider groups and care delivery organizations.

In this post, we explore the growing importance of quantitative teams for risk-bearing entities in the value-based care space, where data analysis and insights are required to operate complex care delivery models and appropriate risk pricing and management. Note that the impact of quantitative specialists is not limited to risk-bearing entities or the value-based care space; however, given the importance of risk and value creation through outcomes, healthcare quants should play a paramount role in these organizations.

While job titles may vary across organizations - from data scientists and health economists to actuarial scientists and business intelligence analysts - there are core functions that remain consistent. Ultimately, all of these core functions aim to understand and optimize the value creation of the organization given risk budget and financial budget (staffing) constraints. I focus on risk management and value generation by reducing the cost of care, and I avoid the quality pillar of value-based care.

3 Pillars of Quantitative Analysis in Value-Based Care

1. Growth and Contract Design

In this initial phase, quantitative experts play a crucial role by ensuring that:

• the contract is designed and priced appropriately based on the underlying population and distribution of outcomes,

• the contract design will yield appropriate (and fair) measurement of performance, and

• the contract's downside risk is understood and reflected in the risk budget.

Let's unpack these points!

• Appropriate opportunity sizing: The first stage of the growth and contracting process is quantifying the potential impact of your care model on the underlying contract population's healthcare utilization patterns. First, you need to understand the risk segmentation and clinical characteristics of your population. This will allow you to understand the distribution of cost of care and the prevalence of the high impact sub-cohorts that will vary the financial impact of your interventions on the population. Secondly, for each intervention and sub-cohort, you need to model the impact of your care model on the cost of care. Remember that the cost of care is a function of utilization and unit prices, and your modeling should reflect the potential impact of your care model on utilization patterns and unit prices. Most often, it is a substitution effect where your care model will reduce the utilization of high unit cost services while increasing the utilization of low unit cost services. The substitution may be direct, such as using generics or biosimilars. Or it may be more involved, like a dementia program that deploys home care management services to substitute high cost acute care utilization by shifting where care is delivered. Note that most companies approach this exercise as a point estimation and scenario (tends to be all positive) analysis exercise. However, it is better to take an approach where you can estimate the distribution of your impact at the intervention level and at the aggregate care model level, which will help you with risk pricing and budgeting.

• Appropriate risk pricing: Value-based care contracts may have different designs where the savings (losses) from reduction (increase) in cost of care are shared between two parties. Note that one party keeping all the savings (or losses) is still a shared arrangement. In the design of the contract, quant teams can aid in ensuring that the additional unit of downside risk is compensated appropriately by the additional unit of upside revenue potential. Once the potential distribution of intervention outcomes is modeled (opportunity sizing), then the potential gross savings and financial outcomes in different contract structures can also be modeled and compared using risk-adjusted expected revenue metrics. Note that estimation of your potential gross savings is only  as good as your modeling assumptions, and they will improve as you gather more evidence over time. But even in the early days, using a consistent framework to compare contract structures will be useful.

• Appropriate measurement of performance: In value-based care contracts, performance is measured against a benchmark that is either constructed using historical data on a control population or constructed using a contemporaneous control population. In either case, the role of quantitative teams is to ensure that the over/under performance (reduced/increased cost of care) versus the benchmark will be due to the interventions delivered by your organization as purely as possible by making sure the trending, matching, risk-adjustment, etc. are done appropriately. In addition, contracts may include carve-outs and winsorization to reduce the variance in the measurement. The quantitative teams need to ensure that these approaches do not hinder the measurement of performance by censoring key intervention cohorts. For example, winsorization and carve-outs may work well to reduce the impact of outliers for a primary care model managing a heterogeneous population, but they may require additional considerations for a specialty care model that is targeting the "outlier" population.

• Appropriate risk budgeting: During contracting, there are two important considerations: 1) Making sure the significant negative scenarios are tolerable for the company, and its balance sheet. 2) Making sure the risk exposure is used for more favorable contracts where your organization has higher expected revenue per unit of downside risk.

In the early days of a value-based care start-up, risk budgeting will be a secondary concern. You need to sign contracts and have revenue to grow as a business. However, as the organization matures, making sure the downside is measured and accounted for becomes a more significant and more strictly (even legally) enforced risk capital constraint. Creating a risk-adjusted expected revenue framework in the early days allows organizations to manage their risk budget thoughtfully before they are locked into multi-year suboptimal contracts and hit their risk constraint.

2. Intervention Design, Evaluation, and Optimization

Once contracts are in place, the focus shifts to designing and refining interventions to generate value in a structured, repeatable, and scalable care model. Again, quantitative experts play a crucial role to:

1. Identification and prioritization of intervention cohorts: Can you identify patients who would benefit from receiving an intervention?

2. Monitoring and measurement of operational effectiveness: Can you deliver the intervention to the identified members effectively and consistently?

3. Measurement of efficacy and outcomes: Do you observe the desired outcomes among the members who receive the intervention?

Again let’s double click into each point!

• Identification and prioritization of intervention cohorts: As we mentioned above in the opportunity sizing section, different interventions in the care model may target different high impact sub-cohorts, or the intensity/dose of the intervention may change across different sub-cohorts depending on the needs and the expected change in outcomes across these cohorts. Data analysis and insights may help deliver interventions to the appropriate individuals at the appropriate time. This is necessary to optimize outcomes given the resource and staffing constraints. Approaches like using a generic risk score (like the Medicare Risk Adjustment model) will not work at the intervention level. Especially for specialty care models, a generic risk score that is trained and constructed to work well across a heterogeneous population may not be discerning enough. For example, all metastatic solid tumor cancer patients may be considered "high" risk in a PCP model, but a cancer specific care delivery model may need to distinguish between cancer types and spread of disease.

• Measurement of operational effectiveness and identification of fail points: In an ideal world, the entire identified intervention cohort should receive the intervention. But delivering healthcare interventions is messy and operationally complex. Hence, there will be fail points, and it is important to understand and fix them. Does your team deliver the interventions? Does your team deliver the intervention timely? Can you reach and engage patients to receive your intervention? If patients do not want to receive your interventions, can you understand why? Do you observe variation across demographic groups, geographies, and potentially different teams in your organization? It is important to capture the data to be able to answer these questions and review them periodically with the leaders of your care delivery organization. For example, if you observe large variations across the effectiveness of different teams and individuals, you may incorporate more training. Even better, you may invest in technology solutions to increase overall effectiveness and reduce the variation among team members by codifying your interventions and processes into workflows. Note that these questions and potential technology investments will also improve the scalability of your care model besides efficacy. More on workflows, technology investments, and scalability in future posts!

• Measurement of efficacy and outcomes: Most importantly, the value-based care model is predicated on improving outcomes. Hence, you want to understand the efficacy of your interventions among the people who receive them. This is where you will see the benefit of conducting a formal and deep opportunity analysis by comparing the efficacy of your interventions to the a priori outcome assumptions that were incorporated into your opportunity analysis. If you are not observing the desired and intended outcomes, then you need to ask: 

  • Is the intervention designed correctly? 

  • Do we need to revisit our hypothesis? More importantly, you may need to update your efficacy and savings assumptions in your opportunity analysis to avoid taking financial risk that you cannot successfully manage in prospective contracts.

3. Interim Adjudication and Savings Estimation

In the previous section, we talked about examining the performance of each intervention in real-time. The interim adjudication process allows your organization to examine the financial implications of the outcomes you are driving within the parameters of each contract. Note that this is a retrospective exercise as you are looking at the cost of care in the past during a performance period. However, you can still ask important questions such as:

• Savings decomposition and explainability: Did all your interventions in aggregate lead to positive financial results? Ironically, you can answer this question with a higher level of certainty as you observe more data during a performance period while having less time to improve the results. This is why intervention level and forward looking measurements from the previous section are important to have a faster reaction time. Can you validate the potentially leading outcome measures from your interventions in the actual claims and utilization metrics? In other words, can you explain the financial outcomes during the interim adjudication with the clinical interventions you have implemented? (This is similar to P&L decomposition in finance). Ideally, the financial outcomes correlate to your interventions, which indicates that the results are driven by your care model. Hence, they are systematic and repeatable, as opposed to idiosyncratic savings (losses). Note that understanding the impact of idiosyncratic factors will allow your team to improve the performance measurement methodology.

• Risk and financial budget: Do you need to update your risk budget and financial projections and revenue expectations based on the revenue (losses) you are expected to incur? These are important questions for the leadership organization to manage the balance sheet and cash flows of your organization.

Postlude

I hope this blog post has helped introduce career opportunities to those with a background in quantitative fields and encouraged them to apply their analytical skills to the meaningful challenge of improving patient outcomes and bending the cost curve in healthcare.

For leaders in value-based care organizations, I have also attempted to provide a framework to structure their quantitative teams by outlining the three key functional areas. This intentional role definition can guide hiring processes, ensuring that organizations build well-rounded teams with complementary skills to address the complex challenges in value-based care and ultimately build financially successful businesses to be able to broaden their impact and reach.

While we cover different functions in a fairly linear fashion across the life cycle of a contract, there is a feedback loop across all these functions. As we mentioned above, the opportunity analysis starts with assumptions on the outcomes of your interventions and translates them into a financial model and gross savings assumptions for contracting. The intervention evaluation and interim adjudication help your organization understand if your care model is driving the assumed outcomes and achieving the gross savings targeted in different contracts.

Beyond the monitoring of individual contracts, as you gain (lose) confidence in the efficacy of your interventions and have a sense of your potential revenue (loss) through interim adjudication, you may take on more (less) risk as an organization. As you target different segments and contract structures, you may need to change your growth strategy and contract structures. By embracing this perspective, risk-bearing entities can optimize their contracting and risk budgeting, resource allocation across cohorts in their population and interventions, and drive continuous improvement through data-driven feedback loops.

Hope you enjoyed this article. Please send your comments and questions to practicalvbc@gmail.com.