Why AI governance is a revenue strategy, not a compliance cost.

1Adoption is widespread. Enterprise value is not.

RAND reports that, by some estimates, more than 80% of AI projects fail—roughly twice the rate cited for non-AI corporate IT projects. Its authors do not reduce that result to a single cause. Their interviews surface failures of problem definition, data readiness, technical infrastructure, talent, and stakeholder communication.^[1] That distinction matters: an 80% estimate is not evidence that every failed project lacked a governance committee, and it should not be converted into an invented dollar-loss total.

McKinsey found a similar scale gap: almost all respondents reported AI use, yet nearly two-thirds had not begun enterprise-wide scaling. Its highest-performing cohort represented about 6% of respondents and was nearly three times as likely as others to report fundamental workflow redesign.^[2] This is an association, not proof that workflow redesign alone causes return. It does show that value capture is an organizational system problem, not merely a model-selection problem.

The useful governance question is not “Do we have a policy?” It is “Can a reviewer trace the claim, decision, evidence, owner, and response?”

2Governance becomes valuable when it produces evidence.

“Governance” is too often used as a synonym for policy, committee, or restriction. Those mechanisms can matter, but none is commercially useful by itself. Evidence architecture is the operational layer: intended-use requirements, acceptance criteria, source provenance, validation records, human-oversight design, change history, monitoring thresholds, and named accountability.

The revenue mechanism is therefore indirect but concrete: better evidence can reduce avoidable rework, expose missing assumptions earlier, support multiple reviews from a shared record, and make post-deployment value measurable. It cannot rescue a product that lacks clinical utility, product-market fit, a payment pathway, or responsible execution.

3The medical-device validation-to-value gap is real.

FDA states the boundary plainly: its principal role is evaluating device safety and effectiveness; after authorization, payors and providers make separate coverage, payment, use, and recommendation decisions. FDA also notes that evidence submitted for authorization may not always overlap with evidence needed for coverage, which can contribute to delayed or denied coverage.^[3]

That does not mean FDA evidence and payor evidence are “orthogonal,” nor that authorization has no commercial value. It means the decision criteria are not identical. FDA's Early Payor Feedback and Parallel Review pathways exist precisely because earlier coordination can help sponsors design evidence generation for both regulatory and coverage questions.^[3]

CMS payment is also specific to the service, code, setting, practitioner, and applicable coverage policy. The 2026 Physician Fee Schedule is a payment rule, not a universal reimbursement pathway for “AI.”^[6] A sound commercial evidence plan therefore starts with the intended clinical service and payment mechanism, not with the assumption that clearance automatically creates a billable event.

4Lifecycle evidence is directionally aligned across high-consequence sectors.

Medical-device AI

FDA's January 2025 AI-enabled device software guidance proposes a total-product-life-cycle approach to risk management and submission documentation. It remains a draft, is not for implementation, and contains nonbinding recommendations.^[4] It should be read as regulatory direction—not misquoted as a new universal postmarket mandate.

FDA's December 2025 final guidance on real-world evidence separately explains how the agency evaluates whether real-world data are fit to support device regulatory decisions.^[5] Together, these sources strengthen the case for planning data provenance, operational measurement, and lifecycle documentation early while preserving the legal status and scope of each document.

Automated driving

Automotive regulation reaches the same structural idea through a different route. In June 2026, UNECE's WP.29 adopted a Global Technical Regulation and a draft UN Regulation on Automated Driving Systems, following GRVA's January recommendation and adoption of the drafts.^[7] The comparison is not that medical devices and vehicles share one compliance regime. It is that high-consequence AI increasingly requires a traceable safety case, defined operating conditions, validation, and post-deployment evidence.

Documentation can support a showing of due care. It is not an automatic legal defense, and it does not replace safe design or responsible operation.

5RIGOR: a reusable evidence lifecycle.

Health AI's RIGOR framework organizes evidence work into five connected domains. RIGOR is the method; Constat is one product implementation of that method for FDA-device evidence. Keeping those roles distinct prevents a consulting framework, a software product, and a research report from competing for the same identity.

The framework's practical advantage is prospective reuse. A monitoring plan designed only after launch cannot recover every missing baseline. A coverage strategy designed only after authorization may discover that the necessary outcome was never measured. Evidence planning does not make all requirements identical; it makes the differences visible soon enough to act on them.

6What “revenue strategy” should—and should not—mean.

Governance is a revenue strategy when it changes how an organization defines, validates, buys, deploys, monitors, and retires AI. The economic contribution may appear as avoided rework, faster review, stronger procurement readiness, better adoption, clearer coverage planning, reduced incident cost, or evidence that supports a differentiated claim.

It is not defensible to infer a universal “governance premium” from an opt-in industry survey, to attribute every failed AI dollar to absent governance, or to promise that governed tools become reimbursable. Governance is one part of a causal system that also includes clinical utility, workflow fit, economics, leadership, data quality, technical performance, and execution.

The strongest commercial claim is modest: evidence architecture makes value easier to test, review, reuse, and defend.

7Constat: the record as a product surface.

Constat applies this evidence discipline to FDA AI/ML-enabled medical devices. It connects source-quoted premarket evidence, postmarket signals, reimbursement pathways, and category-level compliance patterns. The browser console and Constat MCP expose the same evidence lifecycle to people and AI agents, with denominators and source context kept near the claim.