Health systems are pouring investment into AI, but most still can’t vouch for the data underneath it. A recent federal data brief found roughly one in five hospitals don’t know whether their predictive AI models have been evaluated for bias. The reason is rarely the algorithm. It’s unclean data, fragmented records and interoperability gaps that make AI results inconsistent, hard to trust and difficult to scale past a pilot.
Clinical and data leaders from UK HealthCare, University of Rochester Medical Center, Northwell Health and Indiana University Health joined a Becker’s Healthcare discussion on what separates AI that stalls from AI that scales. Their throughline: data is the determining factor, not the model.
This whitepaper is based on their conversation and examines how these systems are building a foundation AI can actually run on.
Learnings include:
• Why AI initiatives stall after an AI partner is chosen, not before
• How leaders decide which use cases to build data around first
• The role governance and clinician trust play in moving pilots into production
• Why real-time interoperability determines whether insights hold across the enterprise