Nelson Advisors: The Industrialisation of Clinical and Administrative Labour - Deconstructing the Autopilot Playbook in Healthcare
Julien Bek’s thesis, "Services: The New Software," argues that the next generation of decacorn and trillion dollar technology companies will not sell software tools to knowledge workers; instead, they will sell the completed work product directly, operating as software companies masquerading as services firms. Across the broader macroeconomic landscape, enterprises allocate approximately six dollars to human professional services for every single dollar spent on software. In healthcare delivery, this economic imbalance is magnified to an extreme degree. Hospital operating margins are perpetually constrained by clinical, operational and administrative labour, which collectively account for 55% to 65% of total hospital operating expenses, while enterprise information technology (IT) budgets remain firmly anchored between 3% and 5%.
Traditional enterprise Software as a Service (SaaS) business models have reached operational saturation within health systems. Point solutions and Electronic Health Record (EHR) add-ons aggressively compete for marginal fractions of constrained IT capital, forcing digital health vendors into protracted procurement cycles, enterprise committee evaluations, and vendor consolidation initiatives. By shifting the commercial target from the software budget to operational and clinical labor line items, the autopilot model fundamentally restructures healthcare economics. Autopilots replace per-seat software licenses with outcome-based deliverables, transforming variable human wage costs into scalable, high-margin algorithmic workflows.
However, healthcare is not an ordinary services vertical. Applying the autopilot playbook to medicine reveals distinct structural dynamics: a strategic wedge located in heavily outsourced administrative bottlenecks, an innovator’s dilemma that constrains first generation clinical copilots, an operational progression along a judgment to intelligence continuum and profound regulatory, tort and actuarial barriers that will dictate the terminal market structure of the AI native health system.
Macroeconomic Foundations: The Labour to Software Disparity
The foundational premise of the autopilot thesis is budget arbitrage: capturing enterprise capital earmarked for human labour rather than software tooling. In traditional enterprise software, a vendor licenses a tool to an internal employee for a monthly fee, leaving the client enterprise to bear the fully burdened labor cost of the professional executing the work.
In healthcare, this division creates severe financial distortion. Health systems routinely spend tens of millions of dollars annually on outsourced administrative labor, business process outsourcing (BPO) agencies, clinical documentation improvement specialists, and third-party medical coders, while hospital Chief Information Officers face constrained capital budgets.
Economic Dimension | Enterprise SaaS Model (Copilot / Point Solution) | Autopilot Services Model (AI Native Managed Service) |
Target Budget Category | Discretionary IT Operating Budget (3% to 5% of Net Patient Revenue) | Administrative & Operational Labor Spend (55% to 65% of Operating Expenses) |
Pricing Architecture | Per-seat / Per-provider monthly license ($300 to $1,000 per seat per month) | Outcome-contingent (Percentage of collections, fee-per-cleared-claim, fee-per-enrolled-trial-patient) |
Procurement Stakeholder | Chief Information Officer (CIO) / Chief Medical Information Officer (CMIO) | Chief Financial Officer (CFO) / VP of Revenue Cycle / Chief Operating Officer (COO) |
Value Realisation Mechanism | Theoretical labor productivity gain (contingent on clinician compliance) | Finished work product (clean claim, prior authorization approval, verified clinical cohort) |
Vendor Displacement | Displaces incumbent point software applications or digital health tools | Displaces offshore BPO vendors, domestic staffing agencies, and internal administrative overhead |
Unit Economics at Scale | Margins capped by enterprise SaaS multiples and software seat churn | Software gross margins (75% to 85%) captured on top-line healthcare service fee schedules |
The core flaw of enterprise health software over the past two decades has been its reliance on human execution to unlock economic return. Software platforms digitise records and structure data, but they externalise the cognitive labor back onto clinicians and administrative personnel, driving widespread clinician burnout and administrative overhead.
When an AI company sells an outcome rather than a tool, every iteration of underlying foundation model capability expands the vendor's gross margin and processing velocity. The hospital Chief Financial Officer does not evaluate this purchase as an IT procurement; it is structured as an operational vendor swap that reduces administrative cost to collect and eliminates back-office labor overhead.
The Administrative Wedge: High Velocity Outsourcing Substitution
Per the Bek framework, autonomous systems establish an enterprise foothold where tasks satisfy three prerequisite conditions: the work is already outsourced, the cognitive burden is intelligence-heavy (governed by deterministic rules) rather than judgment heavy, and the final output is objectively verifiable. In healthcare, hospital networks and life sciences sponsors have spent decades establishing outsourced budget lines to external vendors. These friction points serve as the initial wedge for autonomous platforms.
Operational Domain | Domestic Labour TAM | Incumbent Model Displaced | Autopilot Delivery Model | Representative Market Operators |
Medical Coding & Revenue Assurance | $50B to $80B | Offshore BPO firms, in-house certified professional coders (CPCs) | End-to-end chart-to-claim generation; contingency billing (% of net collections) | Anterior, Arintra, Commure |
Utilization Management & Prior Auth | $30B to $45B | In-house utilization review nurses, manual administrative coordinators | Affirmative auto-approval engines; agentic portal submission and status arbitration | Anterior, Cohere Health, Humata Health, Honey Health |
Clinical Trial Screening & Abstraction | $25B to $40B | CRO hourly billable CRA monitoring, manual site coordinator screening | Real-time EHR phenotyping, automated trial matching, structured eCRF generation | Tempus AI / Deep 6 AI, ConcertAI, Paradigm Health |
Payer Claim Adjudication & Integrity | $50B to $80B | Third-Party Administrators (TPAs), legacy payment integrity vendors | Autonomous claims handling, policy arbitration, and reserve modeling | Pace, Strala |
Revenue Cycle Management and Autonomous Medical Coding
Healthcare revenue cycle management (RCM) represents a domestic outsourced addressable market of approximately $50 billion to $80 billion in labour spending. Medical billing and coding are governed by strict, deterministic, and highly standardised taxonomies: approximately 70,000 ICD-10-CM diagnosis codes, 87,000 ICD-10-PCS procedural codes, and thousands of American Medical Association Current Procedural Terminology (CPT) codes.
Traditional medical coding requires thousands of human coders to manually review unstructured physician encounter narratives, operative reports, and discharge summaries to assign billing alphanumeric strings. This task requires high domain intelligence but minimal strategic judgment. The regulatory rules are intricate, multi layered, and carrier specific, but they remain formalised rules.
Autopilot architectures in RCM ingest unstructured clinical notes via multimodal Large Language Models (LLMs), extract documented co-morbidities, verify National Correct Coding Initiative (NCCI) edits, and generate compliant, audit ready claims without human intervention. Rather than licensing this extraction tool to health information management departments for a software fee, autopilot RCM firms contract directly with health networks to take over billing workflows on a percentage of collections basis or fee per claim model. The enterprise sales motion bypasses clinical committees: the vendor replaces offshore BPO contracts or third party legacy billing services, offering lower error rates, accelerated cash collection cycles, and reduced accounts receivable days.
Payer Provider Utilisation Management and Prior Authorisation
Prior authorisation represents a friction-laden administrative barrier between health plans and providers, driving high operational overhead and clinical care delays. The workflow requires cross-referencing a patient’s longitudinal medical history against granular, carrier-specific clinical coverage guidelines, such as Milliman Care Guidelines (MCG) or InterQual criteria. Historically, this has required armies of intake nurses, coordinators, and BPO personnel navigating payor web portals, compiling clinical chart extracts and exchanging clinical justifications via fax.
Autopilot platforms deploy specialised LLM architectures capable of ingesting clinical charts, identifying relevant conservative treatment histories, laboratory values, and diagnostic imaging reports and programmatically generating or approving authorisation packages. On the health plan side, companies like Anterior deploy clinical reasoning agents integrated into core administrative systems like HealthEdge GuidingCare. By automating medical policy review and auto approving care requests that strictly conform to evidence-based guidelines, Anterior’s platform handles millions of prior authorisations across tens of millions of covered lives, achieving over 99% accuracy while driving an 85% reduction in administrative overhead.
Crucially, the platform operates as an affirmative autopilot: it auto approves compliant care within minutes or escalates ambiguous charts directly to medical directors, bypassing manual intake queues. On the provider side, platforms like Honey Health use agentic browser automation to interact directly with payor portals without requiring lengthy systems integration cycles, charging providers a flat fee of $1.50 to $2.00 per completed authorisation. This completely replaces human prior authorisation coordinators and eliminates operational bottlenecks ahead of scheduled procedures.
Clinical Trial Patient Matching and Chart Abstraction
In biopharmaceutical clinical development, patient identification, pre-screening, and electronic case report form (eCRF) chart abstraction represent severe structural bottlenecks. Industry data indicates that approximately 80% of clinical trials fail to meet initial enrolment timelines, with 53% requiring enrolment period extensions and over 40% of trial sites under enrolling. Pharma sponsors routinely outsource protocol feasibility and trial monitoring to Contract Research Organizations (CROs), which bill sponsors on an hourly professional services basis to dispatch clinical research associates for manual chart reviews across trial sites.
Autonomous clinical research platforms disrupt this CRO dominated services model. By integrating directly with provider EHRs and clinical data repositories, systems such as Tempus AI (which acquired Deep 6 AI to deploy real-time NLP querying across more than 750 hospital sites and 30 million patient records) and ConcertAI deploy agentic architectures across protocol feasibility and cohort selection.
Rather than relying on human study coordinators to cross-reference inclusion and exclusion criteria, which feature complex genomic markers, prior therapeutic lines, and strict physiological parameter windows, natural language processing models parse unstructured pathology reports and physician notes in seconds. Strategic partnerships between life sciences entities, CROs like Parexel, and platforms like Paradigm Health demonstrate the viability of shifting clinical research recruitment from manual chart abstraction to automated, real time clinical trial matching. This replaces the CRO hourly billable model with guaranteed, audit-ready patient cohorts delivered directly to sponsors.
The Innovator’s Dilemma: Ambient Copilots vs. Pure Play Autopilots
The first wave of artificial intelligence adoption in clinical healthcare was dominated by copilots. Systems such as Abridge, Microsoft’s Nuance DAX Copilot, Suki and Ambience Healthcare introduced ambient acoustic listening into the examination room, converting physician-patient dialogue into structured clinical documentation. These applications delivered immediate clinician satisfaction by reducing documentation burden and mitigating administrative burnout. However, from an organisational and macroeconomic perspective, clinical copilots operate under a structural ceiling dictated by their delivery model and commercial architecture.
Copilots are commercialised as enterprise software products sold to hospital IT departments, priced on a per-clinician, per-month SaaS licensing model typically ranging from $300 to $1,000 per user. Consequently, they draw capital exclusively from the health system’s 3% to 5% IT budget. Because the enterprise software market in healthcare is highly constrained, copilot platforms are forced into lengthy procurement RFP cycles, competing against incumbent EHR native feature rollouts. While ambient copilots generate meaningful time savings for individual practitioners, they do not dismantle the hospital's fixed labour base. The hospital still maintains its certified professional coding staff, its billing operations, its denials management infrastructure and its clinical documentation review teams.
The structural vulnerability of the copilot model lies in its product paradigm: the system produces an assistive recommendation or draft note, but relies on the human clinician to manually review, edit, approve, and electronically sign the medical record. The copilot intentionally disclaims responsibility for the final output. The human physician serves as the institutional and legal firewall, reviewing the AI's generation during evening administrative hours or between patient encounters.
Under the Bek thesis, if a copilot enterprise attempts to pivot toward an autopilot, promising to deliver a finished, fully executed, audit ready clinical chart or billable encounter directly to the clearinghouse, it triggers an innovator's dilemma. The copilot company’s existing buyers (physicians and IT leaders) purchased the tool precisely because it kept the physician in sovereign control of their note without changing institutional liability structures. Transitioning to an autopilot model requires cutting the human professional out of the routine execution loop, directly challenging the workflow, compensation models and professional prerogatives of the very champions who brought the software into the enterprise.
Pure-play autopilots bypass clinician seat licenses altogether. Rather than positioning software inside the examination room to assist a physician in drafting a note, an autopilot managed service contracts at the enterprise leadership level (such as the CFO or Chief Operating Officer) to take over the operational pipeline, from clinical encounter recording to code assignment, claim scrubbing and final remittance matching. The autopilot enterprise assumes full responsibility for the finished product, monetising directly as an operational service and capturing the massive labor budget line that health systems previously paid to third-party BPO contractors and internal administrative staffing.
The Judgment to Intelligence Migration: A Three Phase Horizon
Bek formulates knowledge work as a continuum between intelligence and judgment. Intelligence work encompasses tasks governed by rules, frameworks, and objective syntax, complex operations that nonetheless resolve to correct or incorrect outputs. Judgment work requires experience, contextual taste, risk tolerance, and qualitative instinct developed over decades of practice. Over time, as artificial intelligence systems capture domain data from continuous human oversight, the frontier shifts: today’s judgment compounds into tomorrow’s intelligence. In healthcare delivery, this operational migration progresses across three phases, transitioning from deterministic administrative processing to full clinical agency.
Transformation Phase | Target Operational Domain | Operating & Corporate Model | AI vs. Human Allocation | Data Compounding & Feedback Mechanism |
Phase 1: Administrative Autopilot | Billing, denials management, prior auth, eligibility verification, scheduling | 100% Autonomous Software / AI BPO Managed Service | 98% AI Execution; 2% Human Exception Handling (escalation only) | Rule-matching against payer clearinghouse acceptance/denial telemetry; carrier policy updates. |
Phase 2: Hybrid Clinical Service | Post-discharge monitoring, routine chronic triage, medication adherence, cancer screening outreach | Tech-Enabled Care Entity (Friendly PC-MSO structure) | 90% AI Agent Task Execution; 10% Licensed Human Oversight & Liability Signature | Proprietary clinical conversation logs; physician-override and nurse-intervention telemetry. |
Phase 3: Autonomous Clinical Judgment | Protocolised first-line care, minor acute diagnosis, algorithmic chronic disease titration, automated care navigation | Licensed Autonomous Digital Clinic / AI-Native Provider Group | 99% Autonomous Model Agency; Clinician operating as system supervisor | Longitudinal clinical outcomes data; multi-modal biomarker responses; verified diagnostic override loops. |
Phase 1 represents pure intelligence execution. Operations such as coding claims, matching medical necessity guidelines, submitting prior authorisations, and auto-appealing technical denials require no subjective bedside clinical judgment. The operational rules are established by the Centers for Medicare & Medicaid Services (CMS) and commercial payer contracts. In this phase, platforms operate at near-total autonomy. Human intervention is limited to exception routing: when an edge case falls below a statistical confidence threshold, it is routed to a specialised human adjudicator, whose resolution is captured to retrain the underlying agent stack.
Phase 2 transitions the autopilot from the administrative back office to routine, protocolised patient-facing interactions. In this domain, tasks include post-operative discharge check-ins, medication reconciliation, hypertension and diabetes remote monitoring, preventative screening outreach, and patient navigation. Because clinical interactions introduce professional licensure and medical malpractice requirements, software cannot act as a direct vendor to the patient without a licensed human clinician of record. The operational mechanism is therefore a tech enabled care entity.
This phase is embodied by platforms such as Hippocratic AI, which has raised over $404 million at a $3.5 billion valuation from investors including General Catalyst, Andreessen Horowitz, and Avenir Growth. Hippocratic AI deploys specialised, safety-focused generative AI agents built on its Polaris constellation architecture, a collection of multi-model LLMs encompassing over 4.1 trillion parameters, where primary conversational agents are monitored in real time by specialised clinical supervisor models. These agents act as virtual nurses and care coordinators across more than 300 protocolised tasks, communicating via ultra-low-latency voice and text across 25 medical specialties.
The system executes approximately 90% of routine patient interaction, data gathering, and protocol guidance, freeing human registered nurses from repetitive manual outreach. When an abnormal symptom, biophysical spike, or out-of-protocol clinical anomaly is detected, the agent escalates the patient directly to an employed human clinician. Similarly, K Health has operationalised this hybrid model at scale through commercial partnerships with health systems like Cedars-Sinai (CS Connect) and payors like Elevance Health. K Health’s chat first platform ingests longitudinal patient symptoms and medical records, generating probabilistic diagnostic and treatment pathways that human physicians evaluate and authorise in a fraction of normal encounter times.
Phase 3 represents the full realisation of the convergence thesis, where today's judgment becomes tomorrow's intelligence. In this terminal clinical phase, autonomous platforms progress beyond routine communication and administrative coordination to assume direct diagnostic and therapeutic agency across standardised clinical pathways. By aggregating hundreds of millions of human verified clinical encounters, override loops and longitudinal patient health outcomes from Phase 2, the underlying models develop domain judgment that equals or exceeds average human clinical benchmarks. The technological barrier to Phase 3 is rapidly falling; however, Phase 3 encounters the industry's most formidable non-technical barriers: legal licensure, medical malpractice liability and regulatory doctrines.
Systemic Bottlenecks: Corporate Law, Malpractice Tort and Outcome Verifiability
The autopilot thesis translates efficiently to fintech, legal document drafting, and customer relationship management, where legal risk is bounded by contract and work product is immediately verifiable. In healthcare delivery, full stack autopilot execution encounters three systemic bottlenecks: the Corporate Practice of Medicine doctrine, the medical malpractice liability regime and the epistemological verifiability problem.
The Corporate Practice of Medicine Doctrine and the Friendly PC-MSO Model
In the United States, corporate law strictly separates business capital from clinical delivery. Under the Corporate Practice of Medicine (CPOM) doctrine, enforced across major jurisdictions including California, New York, and Texas, a non-physician, for profit commercial entity is legally prohibited from practicing medicine, employing licensed physicians to practice medicine, or exerting operational control over medical decision making. A pure-play software company cannot hold a state medical license, bill an insurer for professional medical services under Current Procedural Terminology Evaluation and Management codes, or directly deliver clinical diagnoses.
To navigate this regulatory barrier, clinical autopilot companies must implement the Friendly PC-MSO legal architecture. Under this framework, two distinct legal entities are formed:
The Professional Corporation (PC) is a dedicated legal entity 100% owned and held by a state licensed physician. The PC employs all clinical personnel, including physicians, nurse practitioners, and physician assistants and holds complete, sovereign legal authority over medical protocols, clinical judgment and patient care delivery.
The Management Services Organization (MSO) is the technology-backed commercial corporation owned by founders and venture investors. The MSO owns the intellectual property, foundation models, computing infrastructure, non-clinical office facilities, and administrative workflows.
The PC and the MSO enter into a comprehensive Management Services Agreement (MSA), whereby the MSO provides its AI platform, administrative engines and billing workflows to the PC in exchange for a Fair Market Value (FMV) management fee.
While the PC-MSO model is standard across digital health, state regulators have enacted aggressive legislative and enforcement actions against private equity-backed and tech enabled MSOs that infringe upon clinical independence. California’s Senate Bill 351 codifies strict prohibitions barring MSOs from directly or indirectly influencing clinical judgment, setting physician diagnostic quotas, dictating clinical appointment durations, or managing clinical staff hiring and firing based on business metrics.
Furthermore, high-profile enforcement actions, such as the California Attorney General’s $4.4 million settlement against Carbon Health and its leadership, demonstrate that regulatory bodies actively prosecute arrangements where an MSO exerts de facto operational control over clinical practices, patient billing communications, or provider equity transfer restrictions. For clinical autopilots, every attempt to optimise clinical throughput via autonomous algorithms risks regulatory scrutiny for unauthorised corporate practice of medicine or illegal fee-splitting.
Tort Law, the Learned Intermediary Doctrine and the Malpractice Defence
Software cannot carry standard medical malpractice insurance; only licensed human clinicians and accredited clinical entities can be insured against professional medical negligence. The American civil liability regime historically relies on the Learned Intermediary Doctrine in cases involving medical technology and pharmaceuticals. Under this doctrine, a medical device manufacturer satisfies its legal duty of care by providing adequate warnings, risk profiles and operational parameters to the licensed physician; the physician functions as an informed, independent intermediary who evaluates the technology and bears legal responsibility for its application to the patient.
When applied to clinical AI autopilots, this legal framework produces an operational and structural paradox characterised by two distinct dynamics:
The first dynamic is the negative outcome penalty paradox. If an autonomous diagnostic algorithm suggests an erroneous therapeutic regimen and the supervising clinician rubber-stamps it without independent review, courts reject the argument that the algorithm directed the treatment. The clinician is held liable for medical malpractice on the grounds of abandoning independent professional medical judgment. Conversely, if a validated, highly accurate clinical algorithm recommends a critical diagnostic intervention that the human clinician rejects or ignores, and the patient suffers harm, the clinician faces severe malpractice exposure for failing to heed standard of care algorithmic telemetry.
The second dynamic is the breakdown of products liability protections. If an AI autopilot transitions to true autonomy in Phase 3 operating without human in the loop clinical mediation, the legal system loses its human learned intermediary. Legal scholars and courts increasingly advocate for subjecting substitutive, autonomous clinical AI to strict products liability up and down the deployment chain. Under strict liability, the AI company is held liable for software errors regardless of the standard of care or institutional diligence.
This creates an immense financial liability profile: a pure play software vendor cannot withstand catastrophic class action malpractice or products liability exposure on a standard software gross margin, forcing the enterprise to integrate deeply into captive insurance structures, risk pools, and professional corporate vehicles.
The Epistemological Verifiability Problem
In white collar verticals like legal document generation or computer software engineering, the verifiability of work product is immediate. An enterprise software engineer writes a test suite; the code compiles and passes, or it fails. A commercial agreement contains mandatory corporate carve outs or it does not. In clinical healthcare, true outcome verifiability is non-deterministic, multi-factorial, and longitudinally delayed by months or years. If an AI system designs a disease management protocol for an early stage diabetic or hypertensive patient, determining whether that care management was optimal involves complex biological noise, environmental inputs, patient non compliance, and socioeconomic variables.
Because ultimate clinical endpoints, such as mortality, end stage renal disease, or major cardiovascular events, take years or decades to manifest, commercial autopilot pricing cannot be tied directly to absolute clinical outcomes. Instead, tech enabled care entities are forced to construct outcome-based billing around intermediate, statistical clinical proxies. These proxies include:
Glycated hemoglobin reduction thresholds in type 2 diabetes management
Systolic and diastolic blood pressure control metrics within targeted patient cohorts
CMS-defined 30-day all-cause hospital readmission penalties avoided
First-pass clean-claim submission percentages in billing pipelines
However, reliance on clinical proxies introduces severe second-order vulnerabilities. Whenever an algorithmic system is monetised against a statistical proxy, it becomes susceptible to Goodhart’s Law: when a measure becomes a target, it ceases to be a reliable measure. Autonomous clinical autopilots run the structural risk of optimising proxy markers (such as aggressive pharmacologic suppression in ways that maximise short term reimbursement metrics while failing to improve aggregate longitudinal health outcomes or lifetime medical costs.
The Terminal Market State: AI Native Managed Care and Full Risk Systems
The culmination of Julien Bek’s thesis suggests that the next trillion dollar market outcome will not be an infrastructure software company, an EHR database, or an AI tool suite; it will be an AI-native managed care organisation or integrated health system that replaces the human administrative and routine clinical routing apparatus with autonomous algorithmic agents, capturing the multi-trillion-dollar healthcare delivery line item directly.
Global Risk Capitation as the Economic Engine
In the traditional fee for service framework, hospitals bill for every incremental bed day, diagnostic test, and procedural intervention. In this volume driven environment, an AI autopilot that reduces downstream hospitalisations directly undermines the hospital's primary revenue driver. Consequently, the AI-native autopilot enterprise can only realise its full economic potential under value-based care, specifically through full-risk global capitation.
Under global capitation, the AI native entity contracts with Medicare Advantage, Managed Medicaid, or commercial self-insured employers to manage a defined patient panel for a fixed Per-Member Per-Month (PMPM) or Per Member Per Year (PMPY) payment, ranging from $6,000 to over $14,000 annually for high-need geriatric populations. In this structure, the financial incentives invert: every unnecessary hospitalisation, emergency department readmission, and duplicative specialist consult represents an expense against the capitated premium. An AI-native provider that replaces expensive human nurse call centres, continuous remote monitoring triage, and routine administrative routing with autonomous software agents lowers its operational medical expense ratio, generating substantial operating margins while delivering protocolised clinical oversight.
The Medical Loss Ratio Paradox and Quality Improvement Activities
However, operating as an AI-native health insurer introduces a rigid statutory constraint: the Medical Loss Ratio (MLR). Established under the Affordable Care Act (ACA), the MLR requires health insurance issuers in commercial and Medicare Advantage markets to spend at least 80% to 85% of premium revenues directly on clinical medical care and clinical Quality Improvement Activities (QIA). Only 15% to 20% of premium revenues may be retained for general administrative expenses, overhead, and corporate profit.
If an AI native health plan achieves hyper-efficiency by replacing administrative staff and manual utilisation management with autonomous algorithms, lowering total administrative spend to 5% of revenue, the statutory MLR calculation limits the entity's ability to simply pocket the 15% margin differential. Under federal rules, an insurer whose medical spending falls below the 80% to 85% floor must issue cash rebates to policyholders.
To overcome this regulatory constraint and capture software-like gross margins, the AI-native managed care organisation relies on federal accounting rules governing Quality Improvement Activities. Federal regulations allow specific technology expenditures to be accounted for inside the clinical medical loss numerator, rather than the administrative denominator, if the technology directly supports clinical care delivery, prevents hospital readmissions, improves patient safety, or manages chronic illnesses.
By embedding AI agents as patient-facing clinical care coordinators, real-time biophysical triage engines, and autonomous case managers, the AI-native health plan classifies its computational infrastructure and model operational costs as clinical care expenditures. Furthermore, by organising as a Provider-Sponsored Health Plan or an MSO holding delegated downstream risk from major payers, the enterprise absorbs clinical risk at the provider group level, where traditional carrier-level MLR constraints do not cap operating margins.
Failure Modes of Automation: The Forward Health Cautionary Case
The pursuit of tech native, automated healthcare delivery offers critical cautionary lessons regarding the failure modes of physical automation versus pure-play software autopilots. In late 2024, Forward Health, a venture-backed primary care company that had raised more than $650 million from prominent venture firms, including Founders Fund and Khosla Ventures, and achieved a $1 billion valuation, abruptly ceased operations, closed all physical clinic locations, disabled its consumer software, and terminated its workforce.
Evaluation Parameter | Forward Health Capital Intensive Model | AI Native Software Autopilot Model |
Capital Allocation | Heavy capital expenditure in urban real estate and custom hardware kiosks ($1M/CarePod) | Zero physical clinic capital expenditure; cloud-native API and browser-agent architecture |
Reimbursement Strategy | Out-of-pocket consumer membership fees ($99 to $150 per month) bypassing insurance | Taps existing institutional BPO budgets, carrier fee schedules, and capitated premiums |
Care Delivery Interface | Unattended physical kiosks executing invasive procedures (blood draws, sensor arrays) | Multimodal conversational voice/text agents and automated administrative pipelines |
Failure Vulnerability | Mechanical breakdown, field maintenance costs, hardware depreciation, patient alienation | Model drift, hallucination risk, regulatory CPOM compliance, edge-case exception handling |
Terminal Outcome | Abrupt liquidation, loss of patient records, total loss of invested venture capital | Compounds operational domain data, expanding gross margins and enterprise software valuation |
The autopsy of Forward Health's collapse provides a case study in execution failure for the healthcare autopilot thesis. After operating tech-forward urban clinics in high-rent metropolitan markets, Forward attempted an aggressive pivot in 2023, raising a $100 million Series E round to build and deploy CarePods, unmanned, custom manufactured, AI driven physical healthcare kiosks placed in shopping malls and commercial gyms. Designed as kiosks costing roughly $1 million each to produce, the CarePods were intended to automate blood draws, biometric scans, and routine diagnostics without human medical staff on site.
Physical clinical delivery resisted unattended mechanical automation. Former personnel reported that patients experienced frequent hardware malfunctions, capillary blood-draw devices failed, and technical breakdowns occurred without on-site clinical support to resolve them. By treating healthcare as a consumer hardware kiosk problem rather than an orchestration and cognitive workflow problem, Forward incurred substantial physical depreciation and maintenance costs.
Simultaneously, Forward attempted to fund high fixed real estate costs, expensive custom hardware manufacturing, and an internal engineering cadre through an out-of-pocket consumer subscription fee of $99 to $150 per month, completely decoupled from commercial health insurance or Medicare reimbursement rails. The model failed to achieve the subscriber density required to service its hardware capital structure, burning through runway before proving unit-economic viability.
When capital availability contracted in late 2024, Forward’s sudden shutdown resulted in patients abruptly losing access to their electronic health records, stranding patient care and triggering regulatory and clinical backlash.
The Forward Health collapse underscores a fundamental boundary in the healthcare autopilot thesis: healthcare delivery cannot be solved by replacing human clinical facilities with unstaffed hardware kiosks. Physical clinical care demands human empathy, tactile examination, and an unbroken duty of care. The viable path for an AI autopilot is not the elimination of physical care infrastructure through consumer hardware, but the complete virtualisation, automation and algorithmic augmentation of the administrative, analytical, and protocolised clinical workflows operating behind licensed clinicians.
Strategic Outlook and Nuanced Conclusions
The application of Julien Bek’s "Services: The New Software" thesis to healthcare reveals both an expansive economic opportunity and a formidable set of regulatory, legal, and operational boundaries. Health systems operate under an inverted financial profile: administrative and clinical labor consumes the vast majority of hospital operating revenues, while enterprise software IT budgets remain firmly capped between 3% and 5%. Software vendors that persist in selling per-seat copilot applications to hospital IT departments face an innovator’s dilemma, trapped under SaaS margin ceilings and forced to keep clinicians manually executing documentation to manage institutional liability.
Enterprise value in healthcare AI will accrue to pure play autopilots that capture the labor budget directly. The operational wedge begins where work is already heavily outsourced, cognitive tasks are rule-governed intelligence, and outputs are readily verifiable: autonomous medical coding, prior authorization adjudication, and clinical trial chart abstraction. In these administrative arenas, platforms displace traditional BPOs and CROs via outcome contingent pricing, expanding gross margins with every improvement in underlying algorithmic precision.
As these systems compound operational data, the frontier shifts from back-office intelligence to clinical care coordination. However, scaling from an administrative autopilot to a full-stack clinical entity requires mastering the structural realities of American healthcare:
First, corporate structures must strictly adhere to the Friendly PC-MSO model to comply with Corporate Practice of Medicine mandates, navigating state-level legislative actions such as California SB 351 that penalise MSO interference in clinical workflows.
Second, clinical risk must be managed within the Learned Intermediary Doctrine and malpractice tort regimes, utilising human in the loop clinical supervision until legal structures can accommodate substitutive AI strict liability.
Third, founders and investors must avoid the capital-intensive hardware and out-of-pocket consumer subscription traps that precipitated the collapse of Forward Health, focusing instead on software orchestration and payer-reimbursed clinical services.
Finally, scalable monetisation requires operating within value-based, capitated reimbursement structures where operational labor reduction generates enterprise profitability, while legally classifying AI algorithmic care coordination as Quality Improvement Activities under statutory Medical Loss Ratio rules.
The healthcare ecosystem will not be transformed by consumer hardware gadgets or incremental documentation tools. The ultimate market leader will be an AI native managed care organisation and integrated clinical delivery network, a technology enterprise operating within a compliant clinical services structure, utilising autonomous cognitive agents to eliminate administrative friction and routine clinical routing, running at software gross margins while delivering high-touch, protocolised patient care.
Nelson Advisors > European HealthTech, MedTech, Digital Health Investment Banking
Nelson Advisors specialise in Mergers and Acquisitions, Partnerships and Investments for Digital Health, HealthTech, MedTech, Health IT, Consumer HealthTech, Healthcare Cybersecurity, Healthcare AI companies.www.nelsonadvisors.co.uk
Nelson Advisors regularly publish Thought Leadership articles covering market insights, industry trends, deal commentary, market analysis & predictions @ https://www.healthcare.digital
Nelson Advisors publish Europe's Leading Healthcare Technology Investment Banking Newsletter every week, join 5000+ HealthTech and MedTech subscribers today! https://lnkd.in/e5hTp_xb
Nelson Advisors pride ourselves on our DNA as ‘Founders advising Founders.’ We partner with entrepreneurs, boards, corporates, venture capital and private investors to maximise shareholder value and investment returns.www.nelsonadvisors.co.uk
#NelsonAdvisors #HealthTech#MedTech#DigitalHealth #HealthIT #Cybersecurity #HealthcareAI #FemTech#ConsumerHealth #Mergers #Acquisitions #Partnerships #Growth #Strategy #NHS #UK #Europe #USA#Canada#Commonwealth#CorporateDivestitures #VentureCapital #PrivateEquity #Founders #SeriesA #SeriesB #Founders #SellSide #TechAssets #Fundraising #BuildBuyPartner #GoToMarket #PharmaTech #BioTech #Genomics
Nelson Advisors LLP
Hale House, 76-78 Portland Place, Marylebone, London, W1B 1NT
Meet Nelson Advisors @ 2026 Events
Digital Health Rewired > March 2026 > Birmingham, UK
NHS ConfedExpo > June 2026 > Manchester, UK
HLTH Europe > June 2026, Amsterdam, Netherlands
HIMSS AI in Healthcare > July 2026, New York, USA
Bits & Pretzels > September 2026, Munich, Germany
World Health Summit 2026 > October 2026, Berlin, Germany
HealthInvestor Healthcare Summit > October 2026, London, UK
HLTH USA 2026 > October 2026, USA
Barclays Health Elevate > October 2026, London, UK
Web Summit 2026 > November 2026, Lisbon, Portugal
MEDICA 2026 > November 2026, Düsseldorf, Germany
Venture Capital World Summit > December 2026 Toronto, Canada
Comments