When Mosai sat down with two clinical leaders from BAYADA — one of the largest home health and hospice providers in the country — for a live webinar, the conversation didn't stay theoretical for long. Within the first few minutes, the panel moved past "what is AI in healthcare" and into the harder, more useful territory: how do you actually build AI that clinicians trust, and what should it be trusted to do?
The panel included:
Elliott Wood, CEO of Mosai
Justin Searle, home health leader at BAYADA
Amy Hirsch MSN, RN, CHPN, Divisional Director of Clinical Practice and Support, Hospice, at BAYADA
This is the first post in a four-part series unpacking the conversation. Here in Part 1, we're starting where the panel did: with the core philosophy that shapes everything Mosai builds — AI's job is to aid clinical judgment, never to replace it.
Ask any provider evaluating AI vendors what they're most worried about, and a version of the same fear usually surfaces: that the technology will start making calls that belong to a clinician. Justin Searle addressed that fear directly, and early, by explaining the guiding principles behind BAYADA's home health practice:
"I wanna root at least part of this conversation in a couple of the guiding principles that we have in our home health practice. The first is around the quality of care that our clients deserve... and the second is around the idea that our frontline clinicians are our lifeblood." — Justin Searle, BAYADA
From there, he was unambiguous about where AI fits — and where it doesn't:
"We're thinking about AI, certainly not in terms of a replacement perspective, but to help enable them to be working at top of title, top of license... AI aids judgment. It can't replace it. Ultimately, independent clinical judgment has to remain superior and supreme, full stop." — Justin Searle, BAYADA
That's not a soundbite BAYADA reaches for occasionally — it's the operating principle behind every AI decision the organization makes. Elliott Wood confirmed it's the same principle Mosai has built toward since its earliest days, long before "AI" became a catch-all marketing term in healthcare:
"We've always referred to it as clinical decision support. That means it is not clinician replacement or clinical decision replacement." — Elliott Wood, Mosai
Why this matters when you're evaluating vendors: any AI partner should be able to explain, specifically, how their tool protects clinical decision-making rather than sits above it. If the pitch centers on the model deciding rather than informing, that's worth pausing on.
Amy Hirsch brought the conversation down to the floor level — where a beautifully designed model still has to survive contact with a skeptical, experienced nurse. Her framing was simple: the technology only works if the people using it understand what it's for.
"Everyone's at different levels, and so everyone has a different understanding of technology as well and AI. And so when we go to our clinicians and tell them we're going to be implementing a new tool, it creates some change management. But in the end, it really is the clinicians are our product... without our clinicians, we wouldn't have patients." — Amy Hirsch, BAYADA
That's a meaningfully different starting point than "here's a new tool, please use it." It reframes the entire rollout conversation around what the clinician gets back — not just what the organization gains.
One of the more clarifying moments of the panel came when Elliott Wood drew a distinction that's easy to lose in a market where "AI" gets applied to almost anything tech-enabled. He separated two categories that often get lumped together in vendor pitches:
Workflow automation — automating menial, low-value tasks. As Wood put it, drafting documentation or entering referral data into the EMR by hand is "a complete waste of someone's time today," and automating it is a legitimate win — but it's not clinical insight.
Clinical decision support — taking a large volume of information from the EMR (including narrative notes) and distilling it into something contextually important for a clinician making a real decision in the moment: a risk score with context behind it, a summary of whether a patient meets hospice eligibility criteria, a signal that a care plan needs to change.
"Whether that's a risk score that has context around it, whether it's a summary that's giving insight into should this patient continue in home health, does this patient meet hospice eligibility criteria — there are a variety of different ways these analytical solutions can be leveraged that aren't taking away that clinician's ultimate judgment... but it is giving that clinician all of the evidence that they need in order to make a decision." — Elliott Wood, Mosai
Both categories are legitimate AI use cases. But they solve different problems, and they require very different levels of sophistication to build well. Automating a data-entry task is a relatively contained engineering problem. Building a model that clinicians actually trust with a life-or-death judgment call — that's a much longer road, built on years of iteration and real clinical feedback, not a single product cycle.
That distinction is exactly where Part 2 of this series picks up: how Mosai's models have evolved over more than a decade to earn that trust, and what that evolution looks like in practice on the ground at BAYADA.
This is Part 1 of a 4-part series recapping Mosai's live webinar with BAYADA. Next up: "From PDF to Platform," a look at how Mosai's clinical models have evolved over a decade — and what that means for providers navigating a market full of newer entrants making similar claims.
→ Watch the full webinar on-demand
→ Continue reading with Part 2