Health system CIOs are under pressure to cut systems, and 76% say application rationalization is critical to their strategy. That makes the EHR’s promise to “do it all” tempting. But the right way to evaluate the EHR against specialized AI vendors isn’t a feature comparison. It’s three questions: Is it safe, explainable, and governed? Is it outcome-driven? Is it worth the cost to switch?
How is Healthcare IT Changing with AI?
In February 2026, the College of Healthcare Information Management Executives (CHIME) released results from a survey of member CIOs and healthcare IT leaders in which 76% of respondents identified application rationalization as critical to their technology strategy. The most frequently cited approach to cutting costs was decommissioning systems. Yet only one in 10 respondents reported having a dedicated application management initiative or portfolio management strategy in place.
Pressure to rationalize technology investments tells only part of the health tech story. Data continues to be locked up by the EHR. Only 16% of CHIME member CIOs reported that their EHR vendor provided vendor-agnostic interoperability. Among the 84% who reported integration barriers, 47% cited cost and 42% cited vendor unwillingness and delays as the biggest obstacles.
The disruption created by AI is adding another layer of complexity. KLAS’s Global HIT Trends 2026 report found that AI is a top IT priority across the globe. Deloitte’s 2026 report on AI adoption found that more than 80% of health systems are prioritizing agentic AI and generative AI as ways to address ongoing staffing, revenue, and operational pressures.
In this environment—where health system CIOs are under pressure to rationalize their technology stacks, do more with their data, and fast-track AI adoption—it is no wonder that large EHR vendors are arguing for greater reliance, investment, and trust.
The EHR is the operational spine of the health system and the vendor relationship with the greatest institutional lock-in. In a recent study published in PLOS Digital Health examining the impact of EHR hegemony on healthcare, the authors state that “this essential infrastructure is now dominated by a single private vendor, raising important questions about competition, interoperability, and public accountability.
We hear this from our own clients. Whispering in their ears is the EHR vendor saying it can do—or is on a path to doing—what many other vendors are already delivering. The argument is that the features and functions supported by more focused solutions are redundant to the “free” capabilities within the EHR. And there is no need to worry about interoperability because the data never needs to leave the confines of the EHR.
But arguing about features and functions is a fool’s errand. Why? Because features and functions are difficult to compare. Unless there is a clear gap, determining whether one workflow is better than another requires significant time and investment to truly understand each solution. Moreover, it is easy, especially in today’s development environment, to hide a lack of meaningful functionality behind a polished interface and user experience.
For IT and procurement departments already strapped for time and resources (never mind the subject matter experts required to determine whether a technology actually solves the problems they face) it is no wonder the EHR becomes the default. Incumbency is the easiest path.
But evaluating the EHR based primarily on features and functionality is the wrong approach. Asking, “Can they do what another vendor does?” is the wrong question.
There are better, more precise, and more forward-looking questions to ask: Is it safe, explainable, and governed? Is it outcome-driven? Is it worth the investment to switch?
These questions get to the underlying problems technology is supposed to solve. They help bridge the gap between what technology delivers and how healthcare actually works. And, importantly, they meet the AI moment.
Let me explain how.
3 Questions for Healthcare Tech Stack Rationalization
Is the AI safe, explainable, and governed?
Executive and clinical leaders are under tremendous pressure to understand how AI impacts care: Who owns the decision? How is it governed? Where does the human fit in the workflow? Yet explainability remains one of the least-assessed criteria in the AI evaluation literature, even as standard software contracts often leave health systems holding the liability.
The Health Sector Coordinating Council Cybersecurity Working Group (HSCC), whose membership includes more than 400 healthcare providers, pharmaceutical companies, payors, and technology organizations, now treats vendor opacity as a risk factor. If a model cannot be explained, it should be considered high risk.
As we’ve moved through our own product development cycles, safety, explainability, and governance have remained central. That shows up in several ways:
- We make AI outputs explainable. Outputs include confidence measures and the reasoning behind them, with access to the underlying source analysis so users can see exactly what prompted the system’s decision.
- We start with the problem, not the technology. AI isn’t always the answer. We first understand the problem, then match it to the right technology approach. And sometimes—yes, I’m saying this as the CEO of a technology company—technology is not the answer.
- We design the human role into the workflow. At every step, we ask: What are the risks, and where does human judgment belong? Sometimes a human is directly in the loop. Other times, a human is on the loop, providing oversight and serving as the escalation point.
Is the solution outcome-driven?
Most AI evaluation stops at technical accuracy and never reaches clinical impact. In a 2025 JMIR systematic review, 59% of reviewed studies focused primarily on accuracy, sensitivity, and specificity, while clinical utility was assessed in only 32% and clinical efficiency in just 18%. The authors urged health systems to evaluate solutions based on clinical effectiveness—whether and how they produce meaningful improvements in care.
Admittedly, outcomes are hard to measure, which is one reason many vendors, EHRs included, don’t focus on them. And in the age of AI, the most important outcome may be something that doesn’t happen: an avoided ED visit, a prevented readmission, or another instance of avoided harm.
We’ve been doing the hard work of tracking and reporting AI-driven outcomes for more than 15 years. We’ve learned two things:
- Always start with a baseline. Every technology implementation should include an assessment of baseline performance for the outcome it is designed to improve. Without it, even when evaluating an incumbent solution, providers are flying blind. Any metric a vendor delivers can look impressive when you don’t know where you started.
- Outcomes measurement is a team sport. Providers should expect vendors to work alongside them to develop the processes and algorithms used to measure performance. This is difficult work. It requires vendors to be comfortable with tough questions, scrutiny, and the pressure testing inherent to healthcare. If your vendor isn’t willing to stand in front of 100 frontline physicians and work through whether they trust the outcomes the technology is designed to support, you have the wrong vendor.
Is it worth the investment to switch?
This question gets to the heart of incumbency. Consolidation almost always feels easier. But is it?
Ripping out a solution that is already in place—and working—comes with a real cost. For one, swapping AI models is not necessarily a one-for-one exchange. These are learning systems that may have been trained and refined within your environment, developing an understanding of your data, workflows, and operating context that takes time to build.
And not all AI is built the same. Research suggests that generally available large language models can be less effective at certain clinical tasks than locally trained or specialized models. Other studies have been more pointed, questioning whether general-purpose models are sufficiently clinically competent for hospital operations. This is a buyer-beware moment. Just because something works in a demonstration doesn’t mean it will perform the same way in the real world.
There is also a people cost. The right technology relationship, as evidenced by how a vendor approaches safety, explainability, governance, and outcomes, is a partnership. The team knows your people, is probably on a first-name basis with many of them, understands your workflows, and works continuously to keep governance and outcomes measurement current. Removing that partner in the name of technology consolidation isn’t simply an adoption challenge. It can quickly become a source of frustration, disruption, and disconnect.
These are certainly not the only questions health systems should be asking. But in the AI era especially, trust should be earned through methodology, not inherited through brand recognition or incumbency.
Starting from a foundation that demands transparency, evidence, and partnership helps distinguish vendors committed to improving patient care and provider performance from those whose business models depend primarily on consolidation, data control, and deeper health system dependence.
