Building health care AI pilots is easy; getting them to succeed at scale is a challenge.

According to a recent report from Bessemer Venture Partners, AWS, and Bain & Company, only 30 percent of GenAI health care pilots reach production, where an even smaller number will go on to be successful.

This is partly because many health care organizations are in “throw it at the wall and see what sticks” mode with GenAI. They’re operating dozens of pilots with little central coordination, planning, or metrics and hoping for ROI somewhere with something. Inevitably, some pilots will show promise and might be implemented in a single department or process. Others will fail because of the many obstacles on the journey from pilot to scalability.

The result for health care is a disorganized and unproductive approach to a transformative technology that could bring enormous benefits to an industry in dire need of them. It’s time health care took a more comprehensive approach to developing GenAI.

Not just an AI problem

Anyone familiar with the history of innovation in health care should not be surprised at the struggle with AI.

Health care has long struggled to move innovations – whether new drugs, diagnostics, or technologies – beyond the proof-of-concept phase. Pilots often fail to address the broader operational, financial, and cultural factors required for long-term success.

One major issue is misaligned goals. Pilots frequently serve narrow special interests rather than aligning with organizational strategic priorities. Additionally, scalability planning is often neglected, with many pilots focusing on technical feasibility while ignoring operational and financial scalability, workflow integration, data interoperability, and regulatory compliance.

Stakeholder engagement is another critical barrier. Without buy-in from clinicians, administrators, and patients, adoption stalls. Unclear ROI further complicates matters; if the value proposition isn’t well-defined, organizations hesitate to invest further. Evidence gaps – such as limited real-world data on outcomes – make it difficult to justify expansion. Finally, integration challenges arise when pilots operate in isolation, making it hard to embed them into existing workflows or systems.

These issues lead to “pilot fatigue,” where organizations repeatedly test small-scale projects without a clear path to system-wide implementation.

Platform vs. point

The goal for health care organizations should be effective, coordinated AI integration across hundreds, even thousands, of uses, rather than dozens of different point solutions.

To break free from pilot fatigue, health systems need to adopt a platform-based approach to digital health. Unlike point solutions, which address specific problems in isolation, platforms provide a unified infrastructure that supports multiple tools, integrates with existing systems, and scales across departments or populations.

Platforms prioritize interoperability, scalability, and adaptability. They are designed to support dozens or even hundreds of applications, enabling widespread adoption across workflows. In contrast, point solutions often solve immediate needs without considering broader impact, making them harder to scale.

For AI, the platform approach is particularly critical. AI is a general-purpose productivity enhancement tool that can be applied across thousands of workflows in a health system. The real value of AI emerges when it scales to dozens or hundreds of applications, unlocking efficiencies and insights that transform care delivery and operations.

Signs that an organization is relying too heavily on GenAI pilots rather than building to scale include: multiple small-scale projects with no clear path to system-wide implementation, a proliferation of point solutions that don’t integrate with limited cross-departmental collaboration, a lack of metrics, leadership indecision, and little discussion of infrastructure, funding, or long-term adoption.

How to build a GenAI platform

It’s not too late for health care organizations juggling dozens of point solutions to convert to a platform approach. Here’s what’s required:

1. Leadership alignment: Scaling efforts must have clear support from executives and align with organizational goals. Leaders play a critical role in securing funding, staffing, and infrastructure while fostering trust and collaboration across teams.
2. Real-world evidence: An evidence base against defined measures is crucial. Real-world data demonstrating clinical and financial outcomes is necessary to justify scaling. Success should be measured against clearly defined metrics, such as improvements in health outcomes, patient satisfaction, operational efficiency, or revenue.
3. Workflow integration: Digital tools must fit seamlessly into clinical, operational, and administrative workflows. Integration with existing technology infrastructure—such as electronic health records (EHRs), enterprise resource planning (ERP) systems, and patient communication tools—is equally important.
4. Data governance: Robust policies for privacy, security, and ethical use of AI are non-negotiable. Health systems must ensure compliance with regulatory standards while maintaining patient trust.
5. Change management: Comprehensive onboarding and training for clinicians and staff help address resistance and emphasize the benefits of new tools.

One last element that deserves special attention: Don’t underestimate economic and cultural costs. Scaling a pilot often requires significant investment in infrastructure and training, costs absent from pilot programs. AI proponents often mistakenly assume that everyone in the organization shares their enthusiasm for the tools when, in fact, the end users are dubious about the whole idea of learning new ways to do things. Remember, they’ve been promised before that new technology will solve their problems only to be disappointed with the results. It’s necessary to bring end users on board by involving them from the beginning and keeping their needs and concerns in mind at all times.

Evaluating ROI

While AI is a relatively new technology, it can be evaluated at scale by using an old standard: the Triple Aim. Does it improve the experience of care through better outcomes for patients and greater satisfaction for providers? Does it improve population health overall and address equity concerns? Does it reduce per capita costs through improved efficiency, increased revenue, etc.?

With a GenAI platform that extends throughout operations, an organization should be able to evaluate results in each area.

Lastly, ask the users: Do they like it? Is it improving their performance? Is it making life easier? They are the ultimate arbiters of success.

The path forward

Every health care organization will adopt GenAI to one degree or another. It’s no longer a question of if, but how well. The ones that see meaningful returns, real impact, and sustained value won’t be the ones endlessly piloting small, siloed tools. They’ll be the ones that approach GenAI as a strategic capability – designed to scale, embedded across systems, and aligned with both clinical and business goals. The organizations that embrace this mindset won’t just adopt GenAI, they’ll lead with it, setting the pace for innovation, efficiency, and better care across the industry.

Kedar Mate is a physician executive.