It Only Takes Ten To Decide If Healthcare AI Pays Off

 By Katy Feldner, VP, AI Transformation & Product

It only takes 10 minutes to determine whether an AI initiative pays off. Most organizations skip it.

I’ve sat in enough prioritization meetings to know exactly when value gets decided for an AI initiative. It’s in the ten minutes at the start, when someone either does (or doesn’t) ask whether the data behind the idea can support it.

Instead, they pick the use case that sounded good in a steering committee, greenlight it, and only start asking “is this working” after it’s built. By then, the answer is expensive to get wrong.

I ask three questions in that window, and the first one kills more good-sounding initiatives than you would expect.

Is the data behind this specific initiative actionable, not just present? A claims-denial reduction initiative needs claims history that’s consistent across every touchpoint in every system, not just present in one of them. A clinical-documentation tool needs notes that mean roughly the same thing across departments. I’ve had to tell teams, more than once, “the data exists, but it’s split across four systems that don’t talk to each other”, and that’s not a footnote. It’s a meaningful difference between what the initiative can deliver, and by how much.

Then the workflow. Does it have a real decision in it? Not every process does. Some workflows eliminate paperwork with tedious extra steps, and while automating them saves time, it won’t change an outcome anyone cares about. Others have a genuine judgment call in the middle: approve or deny, escalate or wait, flag or clear. Value concentrates around the second kind. If nobody can name the specific decision the initiative is meant to improve, I don’t sign off on it.

Then what does it cost? Weighed against what it’s worth, and whether the organization is ready for it. An initiative that saves a department twenty hours a month isn’t automatically the better bet over one that saves five, especially if those five hours hinge on a decision that costs real money when it goes wrong. Most prioritization conversations ignore this and rank by the size of the win alone, which is how a department’s pet project ends up outranking the initiative that mattered more. 

None of this is an engineering checklist. It’s the same case I’d make to a CFO deciding where the next few million of an AI budget goes. Not “will this work?,” but “why this one, over the other four on the list, and what do we expect back?” Data readiness, decision quality, cost against priority. Most organizations can answer for the last one, but only once it’s too late to change the first two.

This is what counts as value in healthcare AI. A bet placed with open eyes about what the data can support, what decision is being improved, and what it costs to get there. 

Getting this call right upstream — with ROAI™ tracking afterward — tells you whether the bet paid off. Get it wrong, and no amount of measurement buys back what should’ve been decided on day one.