At Health Tech Investor Day, Rachel Kern described a potential investment that fell apart over a question considerably less glamorous than anything in a technology presentation: how was somebody actually going to pay for it? The company had made it through an entire management presentation. The payment mechanism still did not make sense. It was a useful reminder that solving a healthcare problem and building a healthcare business are related accomplishments, but neither guarantees the other.
The September 10 event in Washington, Health Tech Law and Policy: What Matters Most to Investors, was hosted by Maverick Health Policy and Goldsand Friedberg, with support from the Consumer Technology Association. I joined Lisa Bielamowicz and Rafid Fadul for the final panel, moderated by Dale Van Demark, on the future of care delivery. By then, the audience had spent the morning discussing federal policy, investment decisions, state regulation, and reimbursement. That sequence gave our discussion of what AI might make possible a useful grounding in what it would take to make those possibilities real.
Julie Barnes opened the day with a straightforward purpose: help investors understand why policy is central to their portfolio companies. What followed made that argument more concrete. These were not simply discussions about avoiding regulatory trouble. They were about decisions that reach into the design of the service itself: who performs the work, who controls it, what evidence is needed, and how the resulting benefit reaches a patient and becomes revenue.
Policy decisions become product decisions
Ross Friedberg and Quinn Shean’s discussion of the corporate practice of medicine made this especially clear. Their subject was the relationship between medical practices and the management companies that support them, including the tension between investors’ need to protect their capital and physicians’ clinical independence. Quinn described concerns about business decisions that materially affect care; Ross explained why access to outside capital is also consequential for organizations trying to deliver it. Neither side of that tension disappears because the company has an excellent technology.
For a founder, this becomes an operating-model question. A company that supplies software to a medical practice is making a different promise from one that organizes and delivers the care. As the product moves from supporting a decision toward directing an action, the boundaries between its technical functionality, clinical responsibilities, and business arrangements need to be made explicit. Discovering late that a critical decision cannot sit where the business plan put it can require changing the service, not merely revising a contract.
AI makes some of these boundaries harder to maintain. Quinn noted that companies and policymakers can use the same phrase, physician involvement, while imagining very different levels of engagement. That difference matters economically as well as clinically. A business designed around occasional escalation has different staffing requirements from one built around review of every recommendation. An investor cannot evaluate the operating model while leaving that distinction unresolved.
There was a related caution in Julie and Brett Meeks’s federal-policy discussion: several of the developments they described were proposals or exploratory discussions, not settled rules. They were identifying where policy might move, not handing the room a set of permissions it could already rely on. For a company whose economics depend on a future change, that uncertainty belongs in the business plan. It should not be hidden inside a confident statement about the size of the market.
The investors described a corresponding gap in product understanding. Michael Brouthers said he increasingly encounters strong technologists who do not understand healthcare implementation. Rachel recounted a health-system CEO’s frustration with vendors presenting capabilities and expecting the customer to figure out the use case. The objection was not necessarily that the technology could not help. It was that the company had left the work of turning capability into a usable service to the buyer.
This is why I think implementation capacity is an investment variable. A deployment that requires extensive custom integration, additional clinical review, and sustained intervention from a few enthusiastic employees has different economics from a repeatable implementation. Those requirements affect the time until the customer sees value, the cost of supporting that customer, and how much work must be repeated at the next site. Calling them “change management” should not make them disappear into a soft category outside the financial analysis.
Payment does more than settle the bill
Robert Jarrin’s explanation of reimbursement showed how deeply the payment system’s assumptions can reach into technology development. He described a valuation structure built around physician work, practice expense, and malpractice liability, and the difficulty of recognizing software-heavy capabilities when software is treated as an indirect expense rather than valued like medical equipment. His point was not that software can never be paid for. It was that the categories used to recognize resources and assign payment may fit poorly with where a new service creates value.
That creates a design tension. If a company discovers that its service becomes financially viable only when delivered through a particular staffing arrangement or recognizable billable activity, the incentive is to build around that arrangement. The result might be a useful innovation. It might also preserve work that the technology could otherwise eliminate or reorganize. Payment is therefore doing more than determining the price after the product is finished; it is influencing which version of the product gets built.
Mario Suarez approached the same problem through distribution and proof. A service can benefit a patient and still struggle to reach that patient because neither the physician nor the payer has a practical route to adopting it. Proving the benefit presents a separate difficulty, particularly when the evidence depends on claims or downstream utilization the company cannot observe. Leslie Krigstein similarly described purchasers demanding demonstrated outcomes and cost reduction rather than another point solution promising the same return on investment as all the others.
Together, those observations sharpen a distinction I keep returning to: value creation is not the same as value capture.
Suppose a monitoring service helps prevent an avoidable hospitalization. The patient benefits. A payer may avoid a substantial claim. But the practice expected to buy the technology might also need to review alerts, contact patients, and handle exceptions without receiving enough additional revenue to cover that work. The clinical benefit has not become less real. The commercial arrangement is simply incomplete. A workable payment model has to connect some of that benefit to the organizations doing, and financing, the work.
Even when the buyer captures the benefit, the financial claim needs care. On our panel, Lisa distinguished reducing work from actually reducing cost. Her administrative-coding example was direct: automating tasks does not remove the associated labor expense unless the organization changes what it spends. She also stressed how difficult those decisions can be for health systems that are major local employers.
Freed capacity may instead become more patient access, less overtime, or a more tolerable working day. Those can be worthwhile reasons to invest. They are not interchangeable with cash savings, and a business case should say which benefit it expects to produce. Otherwise, the vendor and the buyer can agree enthusiastically about “ROI” while describing different outcomes.
Reengineering the service
These questions carried directly into our final panel. Dale began with the deliberately provocative question of whether AI would replace doctors. Rafid immediately asked which physicians and which functions we were discussing, then pointed out that serving someone who currently has no access to a physician is different from replacing one. Lisa likewise argued for decomposing the care pathway into its component parts.
My argument was that AI will unbundle clinical functions rather than replace “the doctor” as one indivisible role. We need to ask which activities are best performed by people, which by conventional software or devices, and which by AI—and then consider what becomes possible when those capabilities are integrated into reengineered workflows. That is a more consequential opportunity than making each step of the existing process slightly faster.
Unbundling is only useful, however, if we put something coherent back together. A patient does not experience diagnosis, monitoring, communication, and follow-up as independent product categories. The transitions matter. A system that detects a problem but leaves nobody responsible for responding has added information without necessarily improving care.
Charlotte Yeh had already given the room a much more human way to think about this. She described a friend whose husband’s diuretic schedule complicated grocery shopping and visiting friends. The treatment existed; fitting it into their lives was the difficulty. Her challenge was to consider the workflow of life, not just the workflows inside hospitals and physician practices.
That example changes what a good product has to accomplish. More data or another reminder may be insufficient when the obstacle is the way a treatment interacts with ordinary life. A useful service has to recognize the problem and connect the patient to appropriate help. Moving care outside the institution does not remove the need for clinical judgment, coordination, or accountability. It changes where those capabilities have to be available and how they connect.
Rafid’s discussion of bringing research to patients reinforced this point from another direction. He described regulatory pathways, reimbursement, evidence, and deployment infrastructure as part of the work needed to get a technology to the bedside. A successful technical result that never reaches patients is not the same achievement as a functioning clinical service.
For me, that was the connection across the day. Policy, payment, and implementation were not separate obstacles surrounding an otherwise finished AI product. They helped determine what service could actually be delivered. A company need not own every component of the clinical pathway, but it does need a credible account of how its contribution fits, and of who will supply the rest.
Rachel’s payment question is therefore worth asking much earlier than the investment presentation. Answering it requires knowing what changes for the patient, what work changes for the organizations involved, and why the party writing the check has a reason to do so. Those are product decisions. Leaving them until later means the product is not yet finished.
My thanks to Julie Barnes and the Maverick Health Policy team, Ross Friedberg and the Goldsand Friedberg team, and René Quashie and the Consumer Technology Association for bringing the day together; to Dale Van Demark for moderating our panel; and to Lisa Bielamowicz, Rafid Fadul, and the many other speakers and participants who made the conversation worth continuing.
-Marc d. Paradis
About the Author: Marc d. Paradis’ professional journey is a fusion of academic rigor with real-world impact. He began his career over 30 years ago as an academic molecular neurobiologist, instilling in him a deep respect for critical thinking and the scientific method.
Transitioning into industry, he held leadership roles that bridged data and healthcare: as Vice President of Data Strategy at Northwell Health, Marc leveraged one of the world’s most diverse clinical data sets to drive patient-centered innovation via a $100M partnership with Aegis Ventures, launching multiple AI-centered startups; and as Vice President & Dean of Data Science University at Optum, he spearheaded the training of thousands of professionals in practical, product-centric AI, data-driven decision making, and ethical data practices. In each role, he fostered cultures of curiosity, critical thinking, and collaboration – precursors to the Constructive Inquiry ethos.
About SIYOM Consulting: Founded by Marc d. Paradis, SIYOM Consulting is a boutique advisory specializing in Data and AI Strategy for Healthcare and Life Sciences. We help health-system executives, pharma innovators and investors identify, evaluate and execute on high-value data and AI opportunities.
Source note for review: Speaker remarks are paraphrased from personal notes, not presented as speaker- and/or audio-verified quotations.