Post by Vi

A patient can be the perfect candidate for a treatment and still never get it. Right condition. Right need. Wrong health plan, and suddenly the drug costs 10X more. The clinical fit was perfect, but the economics weren't. Most AI targeting in healthcare & life sciences never sees that, because it's only looking at one slice of the picture. And that blind spot points to a bigger problem: not the model, but the architecture underneath it. We sat down with Yiftach Meitar Head of Product at Vi, about what gets less attention than models and features: the system AI actually runs on. Most health organizations already run complex data stacks, and AI gets added on top as one more layer. What you end up with is point solutions. One for activation, another for engagement, something else for operations. Each works on its own. None scale together. And each only sees its own slice, which is exactly how the patient who can't afford the drug slips through. Yiftach's team took a different path: one platform that sits above the existing stack and plugs into it directly. No replacing systems. No replacing teams. And the learning never stays local. Every deployment makes the underlying platform stronger. Which raises the obvious question: how do you know it's actually working, and not just riding seasonality or a client's own improvements? Yiftach's answer comes down to one discipline most AI vendors skip: a randomized holdout group. See how Vi proves the impact is real. 🎥 Full video in the comments 👇