The challenge for healthcare executives adopting AI is the noise when trying to advance an initiative. Between an inbox full of vendor pitches, competitors’ announcements, and strategic deadlines, figuring out where to start is a hurdle in and of itself.

Successful deployment of an AI strategy was never about being first. Many early adopters are now untangling technical debt and ungoverned pilots that a deliberate start would have avoided. The lesson is not that AI does not work; it is that an AI strategy without a foundation is not sustainable. Becoming AI-forward is about building the right foundation of data, governance, and business operating model. That foundation is more within reach than most leaders think.

What does AI-forward actually mean in healthcare?

An AI-forward healthcare organization is one designed to enable AI to be built, trusted, and scaled. Being AI-forward does not mean buying more tools or starting more long-running proof-of-concepts. At a high level, this means building a strong foundation of data and governance while designing a business operating model that unlocks the maximum potential of enterprise data. The goal is not a tool count, but an organization where the right people can build and answer their own questions.

As a practical example, a care manager can ask, “Why did readmission rates spike in one region last quarter?” and receive a trusted answer based on internal data. From there, they can begin taking action within minutes. In this scenario, an end user can analyze clinical, patient, and site-level data ad hoc without having to open a ticket or wait days for an analyst to get the answer. The organizations that get to this point most successfully have built a foundation on a data platform that turns patient-facing users into data analysts rather than “data detectives”.

The three blockers that stall most providers

Most providers do not stall due to a lack of ambition. Health systems have more AI ideas than they can fund and no shortage of AI vendors offering to build them. In a 2025 HFMA survey of 233 health systems, 88% reported already using AI. Only 18% had a mature governance structure and a fully formed AI strategy. What stalls progress is one of three structural blockers.

Blocker #1: Data

The first is fragmented data. Clinical data sits in the electronic health record (EHR), operational data in another system, financials in a third, with scheduling, supply chain, and patient experience scattered even further. The deeper problem isn’t that data is spread out; it’s that nothing reliably unifies it. The same patient can carry a different identifier in each source system, so before anyone can analyze a readmission, there has to be a manual reconciliation workflow. This leads to new use cases that pay an integration tax twice: once to build the connections and again to maintain them when a source system changes. What gets shipped covers a narrow slice of the patient, workflow, or operation.

Blocker #2: Governance

The second is governance that does not fit. In some organizations, it is so loose that nobody trusts what gets shipped. A model produces an answer, and a clinician will reasonably ask, “Based on what data?” In others, governance is so rigid that nothing leaves the sandbox, and every new request becomes a negotiation (a long approval cycle for a model that answers questions in seconds). Both fail in the same way, because what is missing in each case is trust: the right people, with the right data, inside guardrails everyone can see.

Blocker #3: Operating Model

The third is the missing operating model: no shared way to pick priorities, equip teams, or move a pilot into production. Every effort becomes a one-off build, and no win replicates. A model that helps with staffing or Emergency Department (ED) boarding in one unit works, and then is unable to scale because there is no ownership in scaling the pilot, nor is there a clear path to production. The pattern is industry-wide: 83% of healthcare executives are running generative AI pilots, but fewer than 10% are investing in the infrastructure to deploy them enterprise-wide. Across industries, roughly 88% of AI pilots never reach production. In the same HFMA Survey, more than a third of health system leaders reported lacking a process to prioritize and effectively scale AI.

If these blockers sound familiar, it is because each one lands on a different desk: fragmented data with the Chief Data Officer, trust with the Chief Medical Officer, and the operating model with the CIO. None of them is an AI problem. Most providers have a foundation problem, and a foundation can be built without boiling the ocean.

Why now is the right moment for healthcare AI

Coming late to AI adoption only hurts if you repeat the mistakes of those who came before you. Avoid those mistakes, and being late becomes an edge, since you get to build on a foundation rather than clean up years of accumulated chaos. The practical move is to watch the organizations that went first, learn from where they stumbled, and start.

The tools have caught up as well. Modern tooling now enables governance across multiple data sources and centralizes authentication and permission management through integrations with existing identity providers. Moreover, the governance model is not limited to data models; governance also applies to AI and ensures that natural language questioning and agentic interactions are secure and reliable. With this foundation, the same care manager who asked about readmission spikes can type their question in plain English and get an answer they can trust in seconds, instead of filing a ticket and waiting days. At the same time, the CISO and IT security office can be confident that human and agentic users do not have access to data they are not allowed to see.

That is a real shift. Two years ago, getting a governed view across clinical and operational data was a quarter-long project. Today, when the scope is well-defined and the right data is available, a small team can get a governed first-use case into production in days rather than quarters. Premier, for example, configured Databricks Genie for production in three days, giving healthcare teams self-service access to governed analytics to benchmark care and identify preventable readmissions. The moment to start is not when everything is perfect. It is when the tools are mature enough to build something real without a massive upfront bet. That moment is now.

The starting line is closer than you think

Providers do not become AI-forward by accident, and they do not get there by having started earliest. They get there by building the right foundation of unified data, governance people trust, and a repeatable operating model. That is more achievable than most executives think.

The next post in this series is a three-pillar blueprint that moves a health system up the curve, paired with a maturity map and a self-assessment you can bring to your next leadership meeting.

Not sure where your organization sits on the maturity map? Connect with us.