Healthcare Adopted AI as Fast as Everyone Else. It Just Can’t Say What It Got In Return.

 By Katy Feldner, VP, AI Transformation and Value Creation

Healthcare adopted AI as fast as anywhere else. Clinician AI usage is already at 67% daily and over 90% weekly, numbers that would make any industry’s adoption curve jealous. Usage was never the problem. So now that everyone is waking up to how AI value gets defined, healthcare isn’t watching from the sidelines. It has the exact same numbers as everyone else, and no clearer sense of the return on them.

Chasing token usage as a proxy for progress is ending, and the AI line items that got approved on faith are about to get questioned by finance. I’d have more patience for the correction if any of it mentioned healthcare. It doesn’t. Healthcare has just as many token-heavy AI line items, and just as few good answers when someone asks what they bought.

That diagnosis — usage isn’t value — is correct, and it’s correct everywhere, healthcare included. It’s the same story, just wearing a white coat instead of a hoodie.

Where it stops translating cleanly is the fix. The replacements on offer are familiar. Tasks completed, developer hours saved, vulnerabilities resolved. Each one assumes a certain kind of workflow, one with a structured input and a binary outcome. A pull request ships or it doesn’t. A ticket resolves or it doesn’t. Software could design its way out of tokenmaxxing because software’s outputs were already countable.

Healthcare’s inputs aren’t structured that way, and its outputs aren’t binary. The input is a claims history spread across four systems that don’t talk to each other, or a clinical note that means something different depending on who’s reading it. The output isn’t “shipped,” it’s a coverage determination a member can appeal, or a triage call a nurse has to be willing to act on. You can’t borrow “tasks completed” for that and call it value. The lesson from software is real. The metric isn’t.

That gap is somebody’s job now, and I’ve sat across the table from that person more than once. If it’s landed on you, you already know it by its symptoms, even without a word for it. You’ve reported a usage number in a room and watched someone ask the only question that mattered, “so what did we get for it,” and knew the honest answer wasn’t ready yet. That’s not a technology problem. It’s a translation problem, and it’s on you before anyone handed you the vocabulary for it.

Optura’s founding bet is that AI in healthcare should be judged on Return on AI Investment (ROAI™), not on how enthusiastically it gets used. I didn’t arrive at that from a trend piece. I arrived at it from watching healthcare organizations sit on AI spend they couldn’t defend, and spending years building the version of “prove it” that holds up against a claims workflow instead of a codebase.

The word for this is new. The question underneath it is old. What counts as value, and what it takes to keep counting after the first win is proven. Those are the two conversations healthcare hasn’t had yet. We’re starting both.