Healthcare organizations adopting artificial intelligence must pair new tools with disciplined data governance, or risk repeating the mistakes of the electronic health record era, industry leaders say. As hospitals and health systems across the United States deploy AI to ease clinician workloads, the lesson of earlier digital transformations is clear: launching a tool is the start of the work, not the end.
The warning reflects a broader shift in how health IT leaders approach technology. Success depends less on what a system can do and more on how well organizations manage the data, workflows and oversight around it.
A Drawing That Still Haunts Health IT
More than a decade ago, a child’s drawing featured in a 2012 JAMA article, “The Cost of Technology,” captured the cost of the EHR transition. It showed a doctor with his back to a patient and her family, turned toward a screen and typing.
The image has stayed with many healthcare IT leaders since. It became shorthand for a common failure: technology that improves record-keeping while pulling clinicians’ attention away from patients.
Digital transformation has delivered real gains. Care access and coordination have improved in many settings, and information now moves between providers more easily than paper charts allowed.
But the journey does not end once clinicians start using a new tool. That principle now applies directly to AI.
Why the EHR Experience Matters Now
The federal HITECH Act of 2009 pushed hospitals and physician practices toward electronic records through financial incentives. Adoption climbed quickly, but usability and workload problems followed.
A widely cited 2016 study in the Annals of Internal Medicine found that physicians spent nearly two hours on EHR and desk work for every hour of direct patient care. Researchers and professional groups have since linked documentation burden to clinician burnout.
AI tools now promise relief. Ambient listening systems, for example, can draft clinical notes from patient conversations so physicians can face their patients rather than a keyboard.
Those promises depend on the quality of the data and the rigor of the implementation. Poorly governed AI can introduce errors, bias and new burdens, including time spent reviewing and correcting machine-generated output.
Governance Moves From Back Office to Front Line
Data governance covers the policies that determine how information is collected, labeled, secured, shared and corrected. In an AI context, it also determines which data trains or informs a model and who is accountable when outputs go wrong.
Weak governance causes concrete problems. Duplicate patient records, inconsistent coding and missing fields can all degrade an algorithm’s performance. A model that works well in one hospital’s data environment may behave differently in another’s.
Regulators have started to address these risks. The Office of the National Coordinator for Health IT finalized its HTI-1 rule in late 2023, requiring certified health IT developers to provide more transparency about predictive decision support tools. The rule asks developers to disclose how those tools were trained, validated and intended to be used.
Industry bodies are also moving. The Joint Commission and the Coalition for Health AI have released guidance on the responsible use of AI in healthcare, urging organizations to build oversight structures, monitor performance and train staff.
The Details That Decide Outcomes
Health IT leaders point to several areas where organizations tend to succeed or fail with AI:
Data quality. Clean, consistent, well-labeled data gives AI tools a reliable foundation. Without it, even sophisticated models produce unreliable results.
Clinician involvement. Physicians and nurses who help design and test workflows are more likely to use tools effectively. The EHR rollout showed what happens when they are brought in too late.
Ongoing monitoring. Models can drift as patient populations, clinical practices and documentation habits change. Organizations need processes to detect declines in accuracy.
Clear accountability. Governance committees should define who approves tools, who audits them and who responds to errors or safety concerns.
Privacy and security. AI systems often process sensitive patient information, which raises HIPAA compliance questions and expands the attack surface for cyber threats.
What It Means for Patients and Providers
For patients, the stakes are straightforward. Well-governed AI could give clinicians more time at the bedside and more accurate information. Poorly governed AI could introduce errors into records that follow patients across care settings.
For providers, the financial and legal exposure is growing. Health systems that deploy AI without clear oversight may face patient safety events, regulatory scrutiny and eroded clinician trust.
Vendors face pressure too. Buyers increasingly ask for evidence of validation, bias testing and transparency, and federal rules now require some of it.
What to Watch Next
Expect more health systems to establish formal AI governance committees and to fold data quality programs into their AI strategies. Watch for additional federal and state rules on algorithm transparency, as well as accreditation standards that tie AI use to measurable oversight.
The central test will be whether AI gives clinicians back what the EHR took away: time and attention for the person in the room. The organizations that get the details right on data and governance will be best placed to deliver it.