Rehabilitation’s Missing Capacity

Musculoskeletal rehabilitation has always been limited by clinician attention. Ambient sensing and AI may not simply digitize it. They may separate recovery from the appointment.

Musculoskeletal rehabilitation is usually described as a clinical service. A patient sees a physiotherapist, receives an exercise program, returns for reassessment, and progresses until discharge.

That description is accurate. It is also incomplete.

The real output of rehabilitation is not the appointment. It is restored functional capacity: the ability to move, work, care for oneself, and participate in ordinary life. Recovery is produced largely between appointments, through repeated exercise, behavioral adjustment, and adaptation to pain, fatigue, and daily circumstances.

The clinician does not directly manufacture that recovery. The clinician coordinates the conditions under which it becomes possible.

This distinction explains why rehabilitation has remained organized around episodic, clinic-centered care for so long. It was not because the industry resisted innovation. It was because observation, judgment, communication, behavior change, and measurement all depended on the same scarce input: qualified clinician attention.

That constraint is now beginning to loosen.

Ambient sensing, computer vision, remote monitoring, and AI-supported reasoning make it increasingly possible to maintain therapeutic awareness without continuous physical presence. If that shift holds, rehabilitation may not merely become more digital. It may be reorganized around a different production model.

The question is not whether rehabilitation apps will replace physiotherapists. It is whether the functions that once required a person in the room can be distributed across sensors, software, coaches, and clinicians.

The scale of the constraint

Musculoskeletal conditions are not a niche clinical category.

The World Health Organization estimates that about 1.71 billion people live with a musculoskeletal condition, and that 2.41 billion people globally could benefit from rehabilitation of some kind. Musculoskeletal disorders are the largest contributor to that rehabilitation need. Low back pain alone accounts for roughly 570 million cases.

The economic burden extends beyond treatment costs. Musculoskeletal disorders generate roughly 150 million years lived with disability annually, about 17 percent of the global total. In Europe, they are the most common work-related health problem, affecting more than 100 million workers and accounting for roughly half of work absences and 60 percent of permanent work disability.

These figures should be treated as directional rather than precise. Definitions vary, and estimates of productivity loss depend heavily on methodology. But the direction is clear: rehabilitation is not only a healthcare service. It is a system for restoring productive capacity.

That makes the industry’s operating model economically important.

If recovery requires repeated observation, individualized adjustment, and sustained behavioral support, then capacity is limited by the number of clinicians available to provide it. More patients require more clinician time. More personalization requires more attention. More access requires more labor.

The result is a familiar pattern: waiting lists, workforce shortages, and care that is rationed by appointment availability rather than by clinical need alone.

The evidence is visible. NHS community health data showed 323,965 people waiting for musculoskeletal treatment in March 2024, an 11 percent year-on-year increase and 27 percent higher than in January 2023. In the United States, the American Physical Therapy Association identifies a shortage equivalent to roughly 12,070 full-time physical therapists.

These are not merely administrative problems. They are the visible output of a production system whose capacity scales with clinician headcount.

Why the appointment became the unit

It is tempting to see the appointment as an outdated workflow awaiting disruption.

That view underestimates the system it replaced.

Rehabilitation is difficult because recovery is personalized and gradual. Two patients with similar diagnoses may need different interventions depending on severity, comorbidities, occupational demands, confidence, pain response, and adherence. Progression is not mechanical. Increased pain after exercise may reflect normal adaptation, excessive loading, poor technique, or something more serious.

Under earlier technological conditions, the most reliable way to manage that uncertainty was to bring the patient to the clinician.

The appointment served several functions at once. It allowed the clinician to observe movement, interpret symptoms, adjust the program, explain what was happening, reinforce motivation, and measure progress. It was not the product of rehabilitation. It was the coordination mechanism for an ongoing process.

That model was rational.

But it imposed a trade-off. Improving access meant increasing caseloads, which reduced personalization. Reducing cost meant reducing contact, which increased risk. Improving fit required more clinician time, which raised cost and constrained access.

Providers competed within that trade-off. They optimized scheduling, opened clinics, refined protocols, and tried to use scarce clinical time more efficiently. But the underlying production architecture remained the same.

The constraint was not a lack of exercise programs. Exercises can be printed, demonstrated, or shared as video at negligible cost.

The constraint was the continuous execution of the functions that turn a program into recovery.

What technology may change

Ambient sensing and AI do not eliminate the need for clinical judgment. But they may change where that judgment is required.

Computer vision can observe movement through a smartphone camera. Wearable sensors can capture motion, activity, and adherence between visits. Software can provide feedback, prompt behavior, collect patient-reported outcomes, and flag deviations from an expected recovery path. AI can help triage information, identify patterns, and support clinical decision-making.

The important change is not that these tools are digital. It is that they can separate some rehabilitation functions from synchronous clinician presence.

Observation can become continuous rather than episodic. Measurement can occur at home rather than only in clinic. Communication can become asynchronous. Routine progression can be supported by software. Clinician attention can be concentrated where judgment, safety, and human encouragement matter most.

This does not mean rehabilitation becomes automated. It means the production process can be redesigned.

A clinician may no longer need to be present to observe every repetition. A coach may handle routine encouragement and adherence support. Software may detect when a patient is deviating from the expected path. The clinician intervenes when the situation requires clinical judgment.

That is a different division of labor.

The new production question

If these functions can be distributed, the industry’s central question changes.

Historically, providers asked: How many clinicians do we need to serve this patient population?

The emerging question is: How should observation, decision-making, communication, behavior change, and measurement be allocated across patients, sensors, software, coaches, and clinicians?

That is a production-design question, not a software question.

It explains why the most interesting companies in digital musculoskeletal care are not simply building scheduling tools or exercise libraries. They are attempting to reorganize the delivery system.

Sword Health pairs wearable motion sensing with an AI platform and remote clinical oversight. Its model targets observation: the system captures movement data between sessions, while clinicians intervene when the data indicates a deviation.

Hinge Health uses smartphone-based computer vision rather than dedicated hardware, combined with physical therapists and health coaches. That choice lowers per-patient cost and removes a hardware dependency, though it may sacrifice some measurement precision.

Kaia Health uses a smartphone camera to track body movement and provide feedback without a wearable. Its approach emphasizes access and low marginal cost, with the company reporting more than one million lives managed and partnerships with more than 50 employers and health plans.

Vori Health takes a different route. Rather than automating observation, it consolidates decision-making in a virtual team of physicians, physical therapists, and coaches, aiming to reduce unnecessary imaging, opioid prescribing, and premature surgical referral.

RecoveryOne emphasizes asynchronous engagement between patients and a care team. Flok Health is building an AI-operated physiotherapy clinic for the NHS, designed around a single-payer system’s capacity constraint.

These are not variations of the same product. They are experiments in different production configurations.

Capital is a signal, not proof

The market has noticed.

Sword Health raised $40 million in mid-2025 at a $4 billion valuation, and by January 2026 its valuation had risen to roughly $4.15 billion. Hinge Health priced its Nasdaq IPO at the top of its range in May 2025, raised $437 million, and closed its first trading day at a market capitalization near $2.9 billion.

These are meaningful signals. Investors are not paying multi-billion-dollar premiums for scheduling software. They are betting that continuous monitoring, software-supported care, and redesigned clinical workflows can produce rehabilitation at lower marginal cost.

But capital is not proof.

A high valuation reflects expectations about future economics, not demonstrated industry-wide transformation. It does not establish that remote monitoring improves outcomes across all patient populations, that employers will sustain spending, or that these models can operate profitably at scale.

The more persuasive evidence will be operational: retention, adherence, clinical outcomes, cost per recovered patient, clinician productivity, and the ability to expand without proportionally expanding clinical labor.

The strategic question is not whether investors believe the model. It is whether the model can deliver recovery more reliably than the system it replaces.

Where value may settle

If rehabilitation is reorganized, value is unlikely to sit only with the company that builds the best exercise app.

The more durable positions may lie in the layers that coordinate recovery over time.

One is measurement infrastructure. A system that can reliably observe functional status outside the clinic becomes more valuable as care moves beyond the appointment.

Another is clinical operating infrastructure. A system that allocates scarce clinician attention efficiently, escalates risk appropriately, and supports consistent care across a large population may become difficult to replicate.

A third is employer integration. Employers experience musculoskeletal disorders as a labor-cost problem. NIOSH estimates that work-related musculoskeletal disorders cost U.S. employers more than $50 billion annually in workers’ compensation and lost productivity. If a provider can demonstrate faster return to work and lower total cost, the buyer may be the employer rather than the patient.

A fourth is outcome-linked contracting. Fee-for-service reimbursement pays for activity. Bundled payments and outcome-linked arrangements pay for recovery. If payment models continue to move in that direction, the advantage will belong to organizations that can manage the entire recovery pathway, not merely deliver individual sessions.

The common thread is coordination.

The scarce asset may no longer be a clinic. It may be the ability to observe recovery continuously, allocate clinical attention intelligently, and produce measurable functional outcomes at scale.

Conditions and failure modes

This thesis is conditional.

It holds only if the technology can produce clinically reliable observation outside controlled settings. A sensor that works in a laboratory may fail in a cluttered home, with poor lighting, inconsistent patient behavior, or a patient who does not use the device.

It holds only if patients engage. Rehabilitation depends on behavior. A system that produces elegant data but does not change what the patient does between sessions has not solved the production problem.

It holds only if clinicians trust the system. If alerts are noisy, if software cannot distinguish routine variation from genuine risk, or if clinicians must review every data point, the model may add work rather than remove it.

It holds only if payment follows outcomes. If reimbursement continues to reward appointments, providers may have little incentive to reduce the number of encounters.

And it holds only if the new model can serve patients who are not digitally confident, do not own suitable devices, or need more intensive support.

The failure mode is not that rehabilitation becomes digital and fails. It is that digital rehabilitation becomes an additional layer on top of the old system: more apps, more data, more dashboards, more administrative burden, and no change in the underlying economics.

What it means for builders

For health systems, the useful question is not “Should we adopt digital rehabilitation?” It is: Which part of the recovery process should no longer require a scheduled appointment?

For employers, the question is whether they are buying sessions or buying recovered work capacity. If the latter, the relevant measure is not utilization but time to functional recovery, recurrence, absence, and total cost.

For startups, the opportunity may lie less in building another patient app than in solving a specific coordination problem: reliable home measurement, adherence support, risk triage, clinician workflow, or employer integration.

For investors, the question is not “Which rehabilitation company will win?” It is: Which production configuration can deliver recovery at lower marginal cost without compromising safety or fit?

For clinicians, the question is how to use freed capacity well. If technology removes routine observation and communication, the value of human attention may rise in the areas where it is hardest to replace: complex judgment, motivation, trust, and care for patients whose recovery does not follow the expected path.

The shift is not digital rehabilitation

The familiar story is that rehabilitation is being disrupted by apps, sensors, and AI.

That story is too shallow.

The deeper possibility is that rehabilitation is becoming separable from the appointment. The functions that once required a clinician’s continuous presence can increasingly be distributed across time, place, and technology.

If that happens, the industry will not simply become more efficient. It will be organized differently.

The scarce resource will no longer be a clinic slot. It will be the ability to coordinate recovery: to observe what is happening between visits, interpret it safely, support behavior, measure outcomes, and direct scarce clinical attention where it matters most.

The appointment may remain.

But it may no longer be the unit around which rehabilitation is built.

Research note: This memo reflects ATHENSBETA’s interpretation of public information, company disclosures, and the underlying research in the attached analysis. Figures on musculoskeletal prevalence, disability burden, workforce shortages, waiting lists, company valuations, and clinical evidence should be verified against primary sources before publication. The production configurations described are design hypotheses, not forecasts of any company’s success.