Sponsored by

The 8-week course that teaches you how to actually evaluate deals.

Learn the same investing frameworks used at BlackRock, KKR, Ares, and other top firms.

Over 8 weeks:

• Underwrite live deal scenarios end to end

• Build and analyze institutional-level real estate models

• Make go/no-go decisions based on risk-adjusted returns

Join the June 8 cohort of the Wharton Online Real Estate Investing program. Use code SAVE300 to get $300 off tuition.

Explore the Program

Policy Pulse: The Long Tail of AI in Healthcare Is Where the Transformation Lives

"We don't have to pay AI a living wage."

I heard someone say that on a panel last week. On another panel the same week, the topic was the second stage of AI. This is what policymakers are setting up for next. Two very different framings of the same technology, and both stuck with me.

Last week's Policy Pulse was about what AI is taking apart in healthcare, including the eroding moats, the care management margin being absorbed by tech-enabled platforms, the moment where "humans alone can't compete." This week, I want to flip the camera around and look at what AI is building. Because most of what we're seeing today is the head of healthcare's AI trend, while most of what I'm excited about is the tail. And it’s long one!

What I Mean by Head and Tail

Quick framing before I dig in.

The head-and-tail concept is borrowed from Chris Anderson's 2004 Wired piece and in his book “The Long Tail: Why the Future of Business is Selling Less of More.” What you should know as you read on: when the cost of reaching consumers drops, individualized and niche solutions can reach the masses. The supply and demand mismatch disappears. Think about Netflix and Amazon as the classic examples of market disruption via the long tail of possibilities opened up by technology .

In healthcare, the head is the concentrated, obvious stuff the high-visibility, high-dollar use cases dominating the market today. Tools built to maximize revenue inside fee-for-service systems: automated coding, ambient documentation, billing optimization. Easy to see because they're loud, well-funded, and already at scale.

The tail is the vast ecosystem of micro-tasks that happen between doctor visits. Tracking medication adherence. Troubleshooting home health devices. Coordinating transportation. Managing weirdly specific workflows like outpatient therapy scheduling or DME onboarding. Thousands of small, scattered workflows that actually hold the key to improving long-term, population-level outcomes.

Here's why this matters:

In traditional economics, healthcare organizations couldn't afford to put enough human care managers on this work because human labor is a constrained and expensive resource. Commonly, care managers would be deployed to help manage the care for the highest 1-5% risk (in a VBC program).

Software developers largely skipped it also, because each workflow was too small to justify custom-built software, although in the last few years progress has been made here.

AI changes the unit economics. 

Generalized models adapt to thousands of small tasks at very low marginal cost.

Suddenly it's viable to solve the small stuff at scale.

The cumulative impact of fixing thousands of minor friction points ends up larger than the impact of optimizing a handful of big administrative processes. Technology starts being about supporting a patient continuously, throughout their actual daily life.

That's the foundation for everything below.

The Head: Where AI Is Showing Up First

The first wave is doing what you'd expect a profit-maximizing tool to do inside a fee-for-service system (squeeze more revenue out of every encounter).

Ambient AI and AI coding assistants can now detect every possible billable code, extracting every available dollar from the payer. That payer could be CMS, a commercial insurer, a self-funded employer, an individual paying privately, or someone chipping away at a high-deductible plan.

Tech can integrate inputs automatically and maximize billing across any clinician type or site of care. This solution was imagined first.

• Billing and coding optimization (RCM)

• Ambient documentation

• Prior authorization support

• Radiology read assistance

• Chatbots and triage front doors

Is any of this contributing to better outcomes? Some, I’m sure, especially if it gives clinicians time back. What it will do and already is doing is push medical cost trend higher for transactional care. That's the head. It's loud, it's quantifiable, and it's already here. 🙃

The Tail: Where the Real Transformation Lives

The tail takes more imagination. It's the part where AI does what humans in healthcare have struggled to do. It’s because the economics never worked, the technology wasn’t ready yet, and the workforce couldn't stretch far enough to reach most patients.

Also:

AI makes small problems economically actionable

As noted already, this is the unit economics shift in action. Healthcare has effectively long- ignored countless small issues because human attention is too expensive and time-constrained. Now it's viable to do, at scale, what was previously valuable individually but operationally impossible. These are things like:

• Medication adherence follow-up

• Discharge instruction reinforcement

• Transportation coordination

• Diet and symptom monitoring between visits

• Caregiver education

• Social barrier identification and tucking patients into support

• Home device troubleshooting

• Specialist referral navigation

• Behavioral “nudges” to manage chronic disease

AI extends care beyond the clinic walls

Healthcare today is still mostly organized around encounters, including office visits, medical testing, admissions, procedures, and more. AI shifts the center of gravity to continuous, longitudinal support: home-based care coordination, chronic disease nudging, home monitoring, caregiver “copilots,” behavioral reinforcement, low-acuity triage, recovery optimization, and medication titration support.

These add up to innumerable of low-intensity interactions happening continuously, outside any formal care setting. This is the long tail in action.

AI reaches populations we've historically failed

Some populations are operationally hard to serve well, such as multimorbid patients, rural patients, elderly and frail individuals living alone, disabled populations, people with behavioral health comorbidities, those with rare diseases, non-English speakers. They all often need coordination, patience, repetition, monitoring, and navigation.

Not to beat a dead horse, but these are all things humans can't scale profitably or with limited human resources but that activities with a high ROI.

Here’s the VBC angle again: AI can redirect attention toward patients and their individual needs that fee-for-service economics couldn’t.

AI cracks healthcare's workflow fragmentation

Healthcare is made up of thousands of workflows, each mostly too small for traditional software vendors to build for.

Infusion scheduling.

Home oxygen paperwork.

SNF transfer coordination.

Pharmacy synchronization.

DME onboarding.

Wound photo monitoring.

Transportation benefit routing.

Building purpose-built tools for each niche workflow required custom development, dedicated training, and more. In addition, there was no logical billing mechanism for these tools to work. The CMMI ACCESS Model is changing this.

The deeper shift is that AI doesn't have to live inside a single application. Much workflow pain is the result of systems living in silos. These are the handoffs between the EHR, the payer portal, the pharmacy system, the SNF, the patient's phone, clinician to clinician, and more.

AI can sit across all of those. Long-tail workflows then finally become addressable.

What This Already Looks Like

Some of this is shipping today. Here are two examples. With ACCESS launching this summer, this will likely become more visible.

Lark runs a 24/7 automated AI health coach for employers and health plans. It’s chat-based, pulling in connected data (weight, activity, sleep, glucose), delivering personalized nudges without clinician review. Published evaluations in diabetes prevention and diabetes care show improvements in engagement and intermediate outcomes like weight loss and A1c. The strongest evidence is on intermediate metrics, not hard endpoints like hospitalizations. But this is the direction.

Withings sells a connected scale, blood pressure cuff, and watch that auto-sync to an app. The app analyzes trends and can trigger nudges or share reports with clinicians or family.

Always-on AI coaching and connected monitoring change behavior between visits, catch deterioration early, and close care gaps before they become expensive. This can lead to fewer preventable ED visits, hospital admissions and readmissions, missed post-discharge follow-ups, and complications from uncontrolled chronic disease. These create medical cost trend improvements and savings that can recur and compound year over year in value-based care structures.

Why AI Succeeds Where Humans Struggle

Some clinicians are worried about how this might reduce the need for human-led healthcare. I see it improving upon the work clinicians can do and allowing their top-of-scope work to carry greater impact on each patient they touch. Might who they touch change? Sure.

Despite challenges scaling this, we've made real progress with things like nurse-led care management and proactive patient outreach. But it depends on a patient's willingness to take phone calls and be transparent with them. It's easy to tell a nurse, "yes, I weighed myself," or "no, I didn't eat the high sodium canned soup."

I can attest to the incredible difficulty in activating most patients in the crucial work of between-visit physical therapist-prescribed care. It’s notoriously hard to impact what patients do when you aren’t in front of them or in their ear.

AI can now become part of the patient's life flow. It can nudge, encourage, course-correct, and be relentlessly consistent in a way no clinical workforce can. It can deliver something that looks a lot like empathy, at a marginal cost that makes it actually deployable.

And AI never gets tired or has a bad day. It doesn’t have a limit to the number of people it can interact with per day. Its reach is hard to wrap your arms around, right?

We're Too Impatient With VBC Transformation

Here's where I push back on the prevailing narrative that says “maybe VBC doesn’t work.”

In the book Same As Ever, Morgan Housel notes that improvements in cardiovascular disease management have prevented something like 25 million deaths, and how he’s struck by this not getting get more attention. He then explains exactly why it doesn't. Slow improvements, especially in things that don't happen, never make news. Look no further than vaccines to see another example of this in action. Countless lives saved with their safety and utility in question by some. Baffling, but explainable in this context.

VBC could be put in the same bucket. I regularly hear from skeptics who expect VBC to have already transformed the system. The truth is, there are massive improvements in pockets, among early adopters, among the practices and risk-bearing organizations willing to learn what success in risk actually requires. Look at advanced primary care practices reducing hospital admissions for Medicare beneficiaries by 40–50%, or reductions in spend and surgical complications inside episode-based acute care.

But we've been at this for less than 15 years. Medical schools aren't teaching VBC patient management in any meaningful way, and neither are most other professional programs.

Healthcare is HARD. It doesn't work like other parts of the economy, either. And its historical doesn't financially reward longitudinal relationships. FFS is baked into the fabric of how we deliver care.

So yes, change takes more than a minute. More than a couple of decades, even. Compounding effects on population health take time inherently. Improving health is not unlike the compounding of improving personal finances: a series of small action steps that build on each other and turn into new habits and behaviors. Another favorite book of Morgan Housel’s, The Psychology of Money, talks about this.

AI Is Arriving Right on Time

And now we have AI to help. I think it's here just in time.

It's going to complement VBC by doing what VBC has needed all along, which is bringing patients themselves into the care team. Newer Medicare models, including the ACCESS Model, lean further into patient engagement and create the financial logic where consumer-facing AI tools actually get rewarded for keeping people healthy.

This is when the head of the trend stops mattering as much, and the tail starts pulling the system in a different direction.

This is also part of why I keep saying this time really is different. The head is loud. The tail is where the long-term win lives.

From Direct Care to “AI-Augmented Practice Operator”

If you are a clinician or manager working on the delivery side of healthcare and want to lead the adoption of tools for your organization, here’s career direction that may emerge.

Instead of fearing automation, forward-thinking leaders looking for career leverage are stepping into roles where they manage the AI tools that handle patient outreach, allowing their staff to operate at the top of their licenses. They are leaning into integrating continuous, longitudinal tech into healthcare clinical settings.

These practice managers and clinical directors are shifting into operations roles dedicated to digital health deployment. They are doing things like designing hybrid models where AI handles the initial patient touchpoints, freeing up the human staff to step in only when complex clinical decisions are required.

The long tail of AI demands a new kind of clinical leader. If you are looking to future-proof your career, moving from direct care to digital practice orchestration might just be the smartest bet you can make right now.

Ultimately, a fragmented healthcare system requires a new kind of operator to run it. The tech is finally arriving, and the roles are right behind it.

*Disclaimer: All opinions and ideas expressed in this article are solely mine and none represent a recommendation or should be viewed as advisement of any kind to anyone to do anything.*