By: Eva Keller
There is a pattern playing out across health systems right now that most executives would rather not talk about. Leadership approves a new AI tool, the procurement team signs off, the IT department runs the integration, and then six months later, the clinicians still aren’t using it. Not because they’re resistant to technology. Not because they don’t understand the value. But because nobody asked the right question before the build even started.
Dr. Ashok Gupta, founder of TheraNow and a Doctor of Physical Therapy who has spent over a decade deploying technology inside live clinical environments, has watched this pattern repeat itself more times than he can count. His diagnosis is straightforward. “Healthcare doesn’t lack innovation,” he says. “It lacks implementation.”
That distinction matters more than most people in health tech want to admit. The industry has spent years celebrating new tools, platforms, and AI capabilities while consistently underestimating how difficult it is to make any of them stick in real clinical workflows. The result is a graveyard of well-funded products that failed not because the technology was wrong, but because the integration was an afterthought.
Gupta uses a simple analogy to explain where most founders go wrong. “A lot of founders like us start building a product first,” he says. “You’re building a house, but we don’t have power, water, or sewer.” In other words, the infrastructure that makes a product usable within a health system must come before the product itself. Distribution strategy, workflow integration, and data pipeline compatibility are the utilities. Without them, even the most sophisticated AI tool is just an isolated piece of software sitting outside the systems clinicians actually use every day.
Epic controls a dominant share of the EMR and EHR market in the United States. For any enterprise health tech product, integration with Epic is not optional; it is the difference between existing inside a clinician’s workflow and requiring them to leave it entirely. “If your clinicians have to actually leave the software and get onto a standalone website,” Gupta explains, “that’s where you’re asking to break the workflow.” And in healthcare, broken workflows don’t get repaired. The tool gets abandoned.
This is the core reason clinicians reject AI tools even when leadership mandates them. The mandate addresses adoption from the top down. The workflow problem operates from the bottom up. When a new tool adds steps, creates friction, or requires a clinician to context-switch mid-patient interaction, the path of least resistance is always to work around it. Over time, workarounds become habits, and the tool quietly disappears from daily use, regardless of the policy.
Gupta sees this failure mode as entirely preventable, but only if builders are willing to prioritize workflow fit over feature richness from day one. “If your tool requires a massive behavior change, adoption fails,” he says. “If it reduces the workload inside the existing workflows, ROI follows.”
The practical implication for health system leaders evaluating new AI tools is to stop asking what the tool can do and start asking where it lives inside the existing workflow. Does it require clinicians to open a separate platform? Does it ask for data input at a point in the process where clinicians are already focused on patient interaction? Does it create a new task or eliminate an existing one?
TheraNow has built its entire AI strategy around eliminating tasks, not adding them. The focus is on using AI to remove the administrative load that was never supposed to be part of clinical work in the first place, such as documentation, coding accuracy, and compliance checks. “We’re focused on building infrastructure inside health systems,” Gupta says, “not just software.”
That framing is the difference between a tool that gets mandated and a tool that gets used. In healthcare, those are rarely the same thing.
