Healthcare's AI Execution Gap

Executive Summary

Health systems are moving quickly to deploy artificial intelligence across clinical care, administration, research, access, revenue cycle, and workforce operations. Yet the governing logic for these investments often remains fragmented. AI initiatives may originate in IT, digital, informatics, research, clinical departments, administrative functions, or vendor roadmaps—each with legitimate objectives, but frequently without a common mechanism for comparing strategic importance, full economic exposure, implementation readiness, or expected enterprise value.

This creates an execution gap. Technical leaders can explain what a capability does and whether it can be integrated. Boards and senior executives must decide whether the underlying problem deserves priority, whether AI is the best intervention, how uncertain the projected benefits are, what other investments will be displaced, and who will convert adoption into measurable clinical, operating, financial, or strategic outcomes. Most directors are not—and need not become—AI specialists. They do, however, understand strategy, growth, capital, risk, and accountability. The institution needs a translation layer between those two domains.

The answer is not to move AI ownership from IT to corporate strategy. AI should be managed as a shared enterprise capability: strategy defines institutional priorities and opportunity cost; finance tests the business case; technology establishes architecture, security, and scalability; clinical and operating leaders redesign workflows and own outcomes; and the board and executive team set ambition, risk posture, and investment thresholds. An independent strategic advisor can help integrate these perspectives, challenge unsupported assumptions, compare AI with competing uses of capital, and create the governance required to turn technical potential into realized enterprise value.

AI activity is expanding faster than enterprise coherence

Hospitals do not lack AI activity. Federal data show that 71 percent of U.S. hospitals reported using predictive AI integrated with their electronic health record in 2024, up from 66 percent in 2023. The same federal analysis found that multiple entities were accountable for evaluating predictive AI at three-quarters of hospitals. That breadth is understandable: AI crosses clinical care, operations, research, finance, technology, and risk. It also illustrates why AI cannot be governed effectively through a single organizational vertical.

In academic medical centers and other large systems, the fragmentation can be more pronounced. Authority is distributed across hospitals, the faculty practice, medical school, research enterprise, departments, institutes, and corporate functions. Grant-funded experimentation, EHR roadmaps, department-sponsored pilots, embedded vendor features, and enterprise platforms may all proceed through different budgets and review processes. Individually, each project may appear rational. Collectively, they may not form a coherent investment portfolio.

The central problem is not that IT leaders are behaving improperly. Their mandate appropriately emphasizes architecture, interoperability, privacy, cybersecurity, vendor viability, reliability, data, and support. Those are necessary conditions for successful AI. They do not, by themselves, answer the enterprise questions: Which institutional problem is important enough to solve? Is AI superior to conventional process redesign? Does the initiative deserve priority over a physician recruit, capacity expansion, acquisition, workforce program, or access initiative? Who owns the business outcome? What evidence will justify scaling—or stopping?

When those questions are not answered consistently, organizations can become technology-led in problem selection: a capability becomes available, then teams search for a use case. Enterprise strategy works in the opposite direction. It begins with material clinical, operating, workforce, research, financial, or growth priorities and asks which intervention—including, but not limited to, AI—offers the strongest risk-adjusted path to the desired result.

The economic exposure is larger than the software invoice

A health system's AI exposure is easy to understate because spending is distributed. Direct costs may include enterprise licenses, EHR modules, cloud consumption, data infrastructure, cybersecurity, integration, implementation partners, and specialized personnel. Less visible costs include clinical validation, legal and compliance review, workflow redesign, training, model monitoring, vendor management, and the time of senior leaders, physicians, nurses, analysts, and operational teams.

Consider an explicitly illustrative portfolio: 20 initiatives with average annual direct and allocated costs of $250,000 each. That represents $5 million a year and at least $15 million over three years before fully capturing shared infrastructure, executive attention, change fatigue, or the cost of failed implementation. This is not an industry benchmark. It demonstrates how a collection of apparently modest decisions can become a material enterprise commitment.

The largest loss may not be the dollars spent on a failed pilot. It may be scarce clinical, technical, and managerial capacity diverted from an initiative with greater strategic value. A positive ROI is therefore insufficient to establish priority. It shows only that projected benefits exceed projected costs under a particular set of assumptions. Investment priority requires comparison with alternatives based on risk-adjusted value, strategic contribution, time to impact, scalability, readiness, and the resources required to execute.

AI ROI is a chain of dependencies

AI business cases often move too quickly from technical capability to financial value. In practice, value is realized through a chain of dependencies. A model must perform adequately; intended users must adopt it; workflows and decision rights must change; operating performance must improve; and the organization must capture that improvement as measurable value. Failure at any link weakens the return.

Ambient documentation provides a useful example. Reducing documentation time may improve clinician experience—a legitimate benefit—but does not automatically produce a financial return. Economic value depends on what happens next: additional patient capacity, reduced overtime, lower reliance on contract labor, improved retention, avoided recruitment, better coding accuracy, or another measurable outcome. If the saved time is valuable but not monetized, the organization should describe the benefit accurately rather than relabel it as cash ROI.

Not every worthwhile initiative needs an immediate financial payoff. Clinical safety, quality, workforce sustainability, research, equity, and access may justify investment. The discipline is to define the intended form of value in advance, establish a baseline, identify the accountable owner, and avoid combining unlike benefits into a single inflated return calculation. Enterprise leaders should understand both who receives the benefit—patient, clinician, payer, researcher, or provider—and whether the organization can reasonably capture it.

Uncertainty is also unusually high. Technologies and vendor offerings evolve rapidly; adoption rates are difficult to forecast; integration and monitoring costs are often omitted; local data quality affects performance; regulatory expectations continue to develop; and causality may be difficult to establish when multiple operational changes occur simultaneously. Business cases should therefore use ranges, scenario analysis, confidence levels, progressive funding, and explicit stop conditions—not a single precise forecast that disguises uncertainty.

The board and executive translation gap

Most board members and senior executives are not experts in model architecture, training data, validation methodology, or technical integration. They should not be expected to become data scientists. Their responsibility is to govern enterprise direction, performance, risk, and capital. Yet AI proposals are often presented in the language of features, algorithms, infrastructure, vendor capabilities, and use cases rather than in the language of enterprise decisions.

That mismatch creates two opposing risks. Leaders may approve investments they do not fully understand because AI appears strategically unavoidable. Or they may underinvest in consequential opportunities because technical teams cannot express their significance in terms of access, workforce, growth, quality, market position, or economic value. The answer is not less technical rigor; it is better translation.

A well-governed board does not need to select algorithms. It should ask whether management has defined the problem, compared alternatives, quantified total exposure, assigned outcome ownership, tested downside scenarios, established monitoring, and created criteria for continued investment. Those are familiar governance questions applied to an unfamiliar capability.

A better governing philosophy: distributed ownership without fragmentation

Moving AI from IT to corporate strategy would reproduce the same structural error in a different department. AI should be managed as an enterprise capability in service of the organization's clinical, operating, workforce, research, financial, and growth strategies—not as an independent technology agenda. Ownership should be distributed, but decision rights, portfolio visibility, and accountability must be integrated.

A practical governing philosophy rests on eight principles:

• Start with material enterprise problems, not available technology.
• Compare AI with other strategic and capital investments—not only with other technology projects.
• Assign every initiative a clinical or operating outcome owner in addition to a technical owner.
• Distinguish technical success, adoption, operating improvement, and realized enterprise value.
• Use progressive funding, evidence-based stage gates, and clear stop conditions.
• Include implementation readiness and change capacity in the investment decision.
• Concentrate resources on a limited number of scalable priorities.
• Retire projects that lack a credible pathway to value, even when the technology remains interesting.

This model does not need to centralize every technical or clinical decision. It does require an enterprise inventory, common evaluation criteria, clear escalation thresholds, and recurring portfolio review. Lower-risk productivity tools may move through an expedited pathway. High-impact clinical models, material capital commitments, or initiatives affecting patient care require more rigorous validation and governance. The level of oversight should be proportionate to risk, cost, strategic significance, and scale.

An enterprise investment screen for AI

A common investment screen allows leadership to compare initiatives across departments and against non-AI alternatives. Scores should inform judgment rather than automate it; the important discipline is the explicit discussion of value, cost, readiness, risk, and alternatives.

Why an independent enterprise perspective can add value

Healthcare organizations generally do not lack AI expertise. Many possess capable CIOs, data leaders, clinical informaticists, researchers, innovation teams, and vendor partners. The more persistent gap is the ability to integrate these perspectives into a coherent enterprise investment agenda. Internal leaders naturally evaluate AI through the mandates, budgets, and histories of their functions. An independent advisor can evaluate it through the interests of the enterprise.

The advisor's role is not to replace technical, clinical, or operating expertise. It is to connect that expertise to enterprise priorities; establish consistent evaluation criteria; test the full business case; surface opportunity costs; translate technical complexity for executives and directors; and clarify who will be accountable for converting a capability into measurable results.

Independence is especially valuable when initiatives cross organizational boundaries, have strong executive sponsors, involve competing departmental interests, or require the organization to stop a promising but underperforming pilot. A neutral perspective can create permission to ask questions that may be difficult within established structures: Is the problem important enough? Is AI the best intervention? Are benefits realistically capturable? What assumptions are weakest? Which projects overlap? What investment is displaced? Who will redesign the workflow? What evidence would require leadership to scale, redesign, pause, or terminate the initiative?

A potential Ridgeline advisory model

Ridgeline Strategy Advisors can help healthcare leadership teams add the enterprise-strategy layer that AI execution often lacks. The work is designed to complement institutional technology, clinical, informatics, research, finance, and risk expertise—not compete with it.

The value of this approach is objectivity plus integration. Ridgeline is not financially aligned with a particular algorithm, platform, department, or legacy initiative. The advisory lens extends beyond the IT capital plan to the broader growth and investment portfolio. It asks where AI can materially improve institutional performance, which opportunities warrant scarce implementation capacity, and what operating changes are necessary to realize the expected value.

This positioning is deliberately restrained. Ridgeline does not claim to validate clinical algorithms, design enterprise architecture, or replace specialized legal, cybersecurity, data-science, or regulatory advice. Its contribution is to help CEOs, CFOs, strategy leaders, and boards make better enterprise decisions about AI—and to provide the connective structure that allows specialist expertise to produce coherent action.

Five actions leadership teams can take now

1. Create an enterprise inventory. Identify active tools, pilots, embedded capabilities, vendor commitments, spending, owners, intended outcomes, and stage of development.
2. Reconnect AI to the strategy. Map each initiative to a defined institutional priority and identify high-value problems that the current portfolio does not address.
3. Rebuild the business cases. Include full cost, benefit ownership, implementation capacity, uncertainty, opportunity cost, scenarios, and evidence required for scale.
4. Clarify decision rights and accountability. Define the roles of the board, executive team, strategy, finance, technology, clinical, operating, research, and risk leaders.
5. Conduct a portfolio review. Concentrate investment on the opportunities with the strongest combination of importance, value, readiness, scalability, and acceptable risk—and stop work that lacks a credible pathway to value.

Conclusion

AI may become one of healthcare's most consequential operating and strategic capabilities. That possibility makes disciplined governance more important, not less. The objective is neither to slow innovation nor to force every clinical and research benefit into a narrow financial return. It is to ensure that the organization knows which problems it is solving, why those problems deserve priority, what the full commitment entails, who owns the outcomes, and what evidence will determine the next investment decision.

Health systems do not need boards that can debate model architecture. They need a translation and governance model that converts technical complexity into enterprise choices about strategy, capital, risk, execution, and value. When that layer is absent, extensive activity can coexist with limited enterprise impact. When it is present, technical, clinical, and operating expertise can be organized around a coherent set of institutional priorities.

Sources and Notes

• ASTP/ONC, “Hospital Trends in the Use, Evaluation, and Governance of Predictive AI, 2023–2024,” Data Brief No. 80, September 2025.
• National Academy of Medicine, “The Opportunity Cost of Market Narratives in Health Care AI,” December 8, 2025.
• American Hospital Association, “How to Build and Implement Your AI Health Care Action Plan,” January 14, 2025.
• U.S. Government Accountability Office, “Artificial Intelligence in Health Care: Benefits and Challenges of Technologies to Augment Patient Care,” November 30, 2020.
• ASTP/ONC, “Accelerating the Adoption of Clinical AI: A OneHHS Approach,” updated July 1, 2026.

The $5 million annual and $15 million three-year portfolio example is explicitly illustrative and is not presented as an industry benchmark. Any organization-specific cost, return, governance, regulatory, or clinical-performance claim should be validated using local data and appropriate specialist review.