By Angel Mena, MD, Chief Medical Officer, symplr
LinkedIn:Â Angel J. Mena, MDÂ
LinkedIn: symplr
The average U.S. hospital ended 2025 with a 1.3% operating margin, according to Kaufman Hall, leaving no room for the operational failures that are already happening inside most health systems. When credentialing backlogs delay provider onboarding by weeks, scheduling systems operate independently from staffing data, or compliance workflows still run on shared spreadsheets, operations break down. And at 1.3%, any of those inefficiencies is also a financial exposure.
AI is exploding in the healthcare industry right now, but we can’t scale the technology in health systems before addressing the fragmented operational layer underneath it. Layering AI onto broken processes will only further the challenges health systems already face.
Fragmentation brings operational and financial risk
Most health systems aren’t running just one set of IT tools. In fact, sometimes they’re running hundreds of tools at once, accumulated over years of solving immediate problems without a governing strategy. According to symplr’s 2025 Compass Survey, 86% of IT leaders report shadow IT in their organizations, department-level software purchases made outside formal approval processes, up from 74% in 2022. When a majority of healthcare IT leaders can’t fully account for what software is running in their own organizations, the foundation required for AI needs real improvement. Fragmented systems produce fragmented data, and fragmented data produces unreliable outputs regardless of how sophisticated the AI model is.
Shadow IT is no longer simply an IT governance issue – it has become a cybersecurity issue. Every unmanaged application introduces another repository for protected health information, another identity to manage, another integration to secure, and another potential attack vector. As AI becomes embedded across software platforms, organizations that cannot account for their applications will also struggle to account for where their data is being used and by whom.
That fragmentation carries a cost well beyond AI performance and cybersecurity exposure. Clinical cost-cutting is largely exhausted. The next wave of margin pressure, and potential relief, is coming from operations. More than half of clinicians lose over an hour a day to tasks that have nothing to do with patient care and with tight margins, that issue shows up on hospitals’ balance sheet.
The OBBBA reimbursement cuts are phased in starting this year and will be in full effect by 2028. When they do, the Congressional Budget Office estimates 10 million more people will be uninsured. Health systems should expect higher emergency department volumes, fewer lower-acuity prevention visits, and a payer mix that will be even harder to manage. Organizations that haven’t addressed their operational infrastructure before those cuts go into effect will have little cushion to work with.
The foundation that determines whether AI works
The healthcare organizations that get real value from AI won’t be distinguished by how many pilots they ran, but by whether they invested time into ensuring they have a firm foundation first.
That starts with an honest application inventory, including shadow solutions, expanding beyond what’s in the IT budget and understanding what is actually deployed and actively used across the organization. That exercise alone should surface redundancy and create the foundation for consolidating solutions and redesigning the workflows built around them, especially at large networks that have grown through years of mergers and acquisitions. This strategic work is often deferred, though, as overextended healthcare leaders are frequently forced to focus on more urgent operational and financial matters, leaving the same problems to be addressed on another day, year after year and still without resolution.
But with the OBBBA regulations looming in the next two years, health leaders should have a renewed sense of urgency to break that pattern before the cuts are finalized. Data won’t always be perfect, but it does consistently need to come from integrated systems that are understood across organizations. When workforce scheduling data lives across three tools with no shared definitions, an AI model built on top of it will likely produce fast, confident, and wrong recommendations. Healthcare organizations must prioritize their operational foundation first, then automation, to ensure they build AI on data that can be trusted.
The health systems that reinforce their operational foundation with consolidated tools, shared data, and regularly examined workflows will be better positioned to deploy AI, maximize its potential, and be more resilient for the industry regulations and financial pressures ahead.