Consumer tech for medical grade devices

A medical device company has traditionally started with an invention, built bespoke electronics around it, spent years on development and regulatory work, and then gone looking for a code to bill against. That order is backwards in a market where coverage, coding, payment rules, and utilization are largely public.

This pillar tests a different order. Start with what payers already reimburse and how much of it is happening. Ask whether the required capability can be assembled largely from components that consumer electronics already produces at scale and at reliability. Then treat the regulatory, quality, verification, and validation work as the actual product-development problem, and ask whether AI-native methods make that work faster, cheaper, and better evidenced without lowering the standard.

Three strategies, in order: reimbursement-first discovery, commodity components until proven inadequate, regulatory execution as the durable advantage. The first two are available to everybody. Only the third can compound.

Reading rule for everything below. Fact means it is stated in the cited first-party source. Inference means it follows from a cited source but the source does not say it. Estimate means a number we produced and would defend but cannot cite. Hypothesis means we think it is true and intend to test it. Assumption means we are proceeding as if it were true and have not tested it.

Coverage, coding, payment, and utilization for most of the reimbursed device world are matters of public record. National and local coverage determinations state who qualifies and what documentation is required. The DMEPOS and physician fee schedules state what is paid and where. HCPCS updates quarterly. Utilization files show what was actually billed, by whom, and how often.

That means a new entrant can know, before writing a line of firmware, whether anyone will pay for the thing. The question stops being will this be reimbursed and becomes where is reimbursement already proven enough that the work left is building something materially cheaper, simpler, faster, or better.

We do not publish payment amounts here. Medicare rates change quarterly, vary by locality and by competitive-bidding area, and a number copied into an essay is wrong within a year. Amounts belong in the fee schedule lookup at the moment you need them.

The Zero Neo assumption, transferred to hardware: do not design bespoke electronics unless a commodity part is demonstrably inadequate for the medical claim being made.

Accelerometers, gyroscopes, magnetometers, optical heart-rate sensors, temperature and pressure sensors, microphones, cameras, depth sensing, Bluetooth, Wi-Fi, cellular, GPS, displays, batteries, wireless charging, low-power processors, secure elements, edge inference accelerators, and embedded storage all exist at consumer volumes, with mature supply chains and reliability data that no medical-device production run will ever match on volume alone.

This is not an argument that consumer-grade is medical-grade. It is the opposite. It says the component is the cheap half. Medical grade comes from what surrounds the component: calibration against a reference method, accuracy characterization, reliability across the intended life and environment, electrical safety, biocompatibility where there is patient contact, cybersecurity and an update path, manufacturing controls, and clinical evidence proportionate to the claim.

So the question to ask, component by component rather than device by device, is what is genuinely medically specialized here, and what is simply expensive because someone built a custom version before the commodity part existed.

Anyone can buy the same sensor. Component commoditization is not an advantage; it is a starting condition. The advantage, if there is one, is a repeatable capability for regulated product development that a competitor cannot assemble quickly.

The work is well defined and largely document-shaped: requirements and design inputs, design outputs, traceability, the design history file, risk management under ISO 14971, hazard analysis and FMEA, verification and validation planning and protocols, usability and human factors, software lifecycle and cybersecurity documentation, threat models, SBOM, test generation and regression, quality-system documentation, design reviews, change control, CAPA, supplier qualification, complaint handling, post-market surveillance, adverse-event monitoring, regulatory intelligence, predicate research, substantial-equivalence analysis, labeling and instructions for use, clinical evidence planning, manufacturing documentation, and audit preparation.

Most teams reconstruct the relationships between those artifacts shortly before a submission or an audit. The hypothesis under test is that an AI workforce can maintain them continuously instead, so the chain from clinical need to post-market observation is always current and always inspectable.

The standing limit on all of this. AI drafts, researches, traces, tests, and keeps the record current. Qualified regulatory, clinical, engineering, quality, legal, and testing professionals make the judgments and hold the approvals wherever law, standard, or competence requires it. An AI output is never a clearance, a validation, a design review, or a release.

A working, ranked table rather than a catalogue. Ranking is the point: a list of every reimbursed device category is useless, because the interesting question is which few are mispriced relative to what they now cost to build.

Reimbursement mechanism and code are facts, each tied to the cited source. Commodity feasibility, regulatory burden, AI-native leverage, and the Native Alpha reading are our estimates. No payment amounts appear; look those up in the fee schedule for the current quarter and your locality.

The score is a way to argue in the open, not a calculation. Each dimension is rated low, moderate, or high, the reasoning is written down, and the output is a rank order. Anyone who disagrees can point at the dimension they would score differently.

We do not average the dimensions into a single number. Averaging hides the veto conditions, and this framework has veto conditions: if reimbursement certainty is low or the sensing physics has no commodity equivalent, nothing else on the list saves the opportunity.

Signal. A device class stays expensive after its electronics commoditized. Reimbursement is attractive but the product is inconvenient. The incumbent’s cost structure is a legacy manufacturing decision. Regulatory documentation, not hardware, is the barrier to entry. Consumer devices already generate a signal that a reimbursed workflow needs.

Availability. What is genuinely available: public data, components, standards, cleared predicates, published methods, commercial rights. Two standing cautions apply and neither is negotiable. Never infer ownership from inventorship. Never infer freedom to operate from the fact that a component can be bought.

Relevance. Somebody with a name has the problem: a patient, a clinician, a supplier, a payer. If the problem only exists in a market report, stop.

Product wedge. A narrow, useful device that can be built quickly from commodity parts and taken through a known pathway. Not a platform.

Demand proof. Reimbursement, existing utilization, provider purchasing, supplier demand, patient use, design partners, letters of intent, paid pilots, implementation access, recurring use. Interest is not demand.

Rights. The narrowest commercially useful rights. A patent portfolio is not automatically the answer; workflow knowledge, regulatory execution, data, distribution, reimbursement knowledge, and manufacturing economics can defend better and cost less.

Compounding. Each deployment should make the next one cheaper: clinical and longitudinal data, reimbursement intelligence, regulatory knowledge, manufacturing experience, supplier and provider relationships, reusable verification assets, reusable quality infrastructure, and a regulatory AI workforce that gets better with every submission.

Some device categories are expensive for good reasons, and pretending otherwise is how people get hurt. Novel implantable materials. Sensing physics with no commodity equivalent, which is why continuous glucose monitoring sits in this pillar as a control case rather than an opportunity. Extreme measurement precision. Sterile or single-use manufacturing. Unusual energy delivery. Life-sustaining function where failure kills. Evidence requirements that mean a prospective clinical trial. Difficult biological interfaces.

In those categories the commodity component is a small fraction of the cost and the hard part is exactly the part that has not commoditized. The thesis does not apply and we say so rather than stretching it.

The regulatory burden may not be compressible in the way we think. Much of the effort in a submission is judgment and negotiation, and judgment does not speed up because the document was drafted faster.

Quality-system maturity, not documentation speed, may be the real barrier. An auditor is assessing whether a system is lived, not whether records exist.

Reimbursement may be the moat rather than the opening. Supplier enrollment, competitive bidding, prior authorization, and documentation requirements can keep a better and cheaper device out of the market entirely.

Consumer component supply may prove unsuitable in a way that only shows up years in: a part goes end-of-life, a firmware change alters behavior, and the change control obligation is ours.

And the incumbents are not fools. Some prices that look like legacy engineering are actually the cost of service, distribution, and documentation that a newcomer has not priced yet.

The sources this pillar reads, and the ones any claim here should be checkable against. First-party wherever possible: coverage and coding from CMS, classification and clearance from FDA and openFDA, adverse events and recalls from the FDA databases, plus standards, manufacturer documentation, clinical literature, and public company filings where a claim needs them.

The boundary guidance matters as much as the databases. Whether a product is a device at all, and whether a software function is regulated, is decided by the general wellness and clinical decision support policies well before anyone talks about a pathway.

If consumer technology has already commoditized most of the hardware, and AI-native engineering can commoditize much of the development and regulatory workflow, which reimbursed medical devices are still priced as though neither change has happened?