Subscribe
Share
Share
Embed
Is there a single right way to run a home care agency? We sure don’t think so. That’s why we’re interviewing home care leaders across the industry and asking them tough questions about the strategies, operations, and decisions behind their success. Join host Miriam Allred, veteran home care podcaster known for Home Care U and Vision: The Home Care Leaders’ Podcast, as she puts high-growth home care agencies under the microscope to see what works, what doesn’t, and why. Get ready to listen, learn, and build the winning formula for your own success. In the Home Care Strategy Lab, you are the scientist.
Miriam Allred (00:03)
Welcome back to the Home Care Strategy Lab. I'm your host, Miriam Allred. We are live in Palm Springs in California at the Home Care Innovation Forum brought to you by Phoebe. And I'm sitting across from Meghan Willson, the VP of Technology at Team Select. Meghan thanks for being here. How's the conference been so far?
Meghan Willson (00:22)
Absolutely. Wow, so much energy here.
Miriam Allred (00:28)
So much energy. That's what you said, that's what you're gonna take back to the team is excitement.
Meghan Willson (00:33)
Yeah, we you know, we we go to work, we put our heads down, we're you know, we're in Q2 and you know, the roadmap seems endless and then we get to come here and you know, have colleagues and and people with high high energy, high motivation to solve problems. It's really nice.
Miriam Allred (00:49)
And I love what you said. The attendees here, the vibe here is like everyone's just like open to sharing. Open up the playbooks. Let's hash out this morning. Someone asking, How do I let go a C suite employee? It's like, let's just get some like hard, good insights from these providers.
Meghan Willson (01:02)
Yeah.
Let's share. We're in I mean, yeah, we're in a space where sharing is good for everybody.
Miriam Allred (01:08)
Yeah, I couldn't agree more. And that's why we're here on stage having a conversation. Let's just start with an introduction of Team Select for people that don't know a lot about the brand or the company. Just give kind of overview of size, locations, payers, just some of those kind of key details.
Meghan Willson (01:23)
Yeah, we are we are mostly long term care. ⁓ so our our primary payers are Medicaid or commercial plans that that work with Medicaid. you know, we are in 15 states, we have 9,000 employees, we cover five thousand patient lives. so we you know, in the healthcare space we are a niche practice, but we're a niche practice that covers a large body.
Miriam Allred (01:51)
Okay, and this is a loaded question, but just give a like a quick synopsis of your personal story, like where you started and some of the key milestones that have led you to where you are.
Meghan Willson (02:00)
Yeah. She's a I don't know how quick you want it to be. I am a physical therapist and I came into physical therapy practice ⁓ you know, well over twenty years ago. you know, I I moved into home health care about 10 years ago and specifically kind of like had a hunger for tackling ⁓
t for tackling like it moved me into operations because you know we had patients and we had providers but somewhere in the middle you know operations is is a role where if you're successful you get to treat more patients and you get happier providers which as a provider I'm happy to do. somewhere along the way the technology team adopted me and I think it's because operations and technology go so well together. ⁓ and then just recently I've had like this really great
experience where ⁓ the clinical team and our technology team work together to launch clinical clinical work ⁓ it's called care site ai but it's it's really brought for me together a package of getting to be a physical therapist at the beginning and then as you know a leader in technology getting to bring that back home.
Miriam Allred (03:21)
I love it. And what you said to me when we first met was we have to save operational dollars so that we can give more to the workforce. Like that is all of our objective of saving money and becoming more efficient on the operational side so that we can spend more time and more money on the workforce and also on the clients.
Meghan Willson (03:39)
Yeah,
yeah, no, it's true. ⁓ you know, I d intentionally or unintentionally, there's so much that goes into ⁓ partnering patients with providers. ⁓ you wanna spend the ⁓ majority of your time trying to make that magic happen, but there's so much noise in the way. Like, I mean payroll is this like thing, you know. ⁓ when we were a small company, this is a little embarrassing, but when we were a small company, we like had a human that would just like count out visits.
Not efficient at all. And so we found that technology, and I I'm being extreme with my example, but it's true. We found that technology was the the better choice for that and it could run in the background and we could keep that same human, making sure that like the more magical thing happened where patients and ⁓ providers get together.
Miriam Allred (04:30)
Yeah, and you said to before we started recording that you started on paper. You started on paper. And there's there's there's a good thing going with paper communication, like and and technology, we're losing some of the essence of like what that once was.
Meghan Willson (04:43)
Yeah, it was so funny. I just telling this story earlier. ⁓ you know, when I came into my practice as a therapist, we had these binders, they were paper charts, and you just open it up and they were color coded. I mean, what what better system than to just say what's in red is important and then the next person would document and so you would just see this nice clear communication. And then I remember as a therapist moving into documentation on you know on the computer in the software.
And it was just really painful, right? Like we just lost this binder that sits at the bedside. It improved a ton of things. I'm not ⁓ it obviously like we wanted to be in a place where you didn't have to rely on a paper chart to get your medical record. So moving into software helped so many things, but along the way, it I didn't feel like it helped us communicate with each other, and I didn't felt feel like it helped us necessarily provide better patient care.
Miriam Allred (05:40)
Yeah. So as a part of your role, you have overhauled the tech stack. And I want you to share a little bit about what that process looked like because then that set the foundation for what you're doing with AI right now. But I think it was just a couple of years ago you guys had to overhaul a lot of your tech. Explain again in a few words kind of what that meant to overhaul a tech stack of a company your size.
Meghan Willson (06:01)
Yeah. Well, so you know, our tech stack and tech team before the overhaul was ⁓ struggling and it was almost like a very visible situation. Like our our engineers and our team were quite frustrated ⁓ visibly and also just working late. Like, you know, we would have like these fire drills at two in the morning to make sure that you know the payroll system worked ⁓ the night before. ⁓ and then another like really interesting visual that
I you know, was just a takeaway for me is sometimes the guys would be like, you have to crack the door, and I would be like, Well, what's in that door? And it was our it was our server. ⁓ and they were like, It's literally overheating. And so not just to be extreme, it was on fire, right? Like I was like, Wow, our our
Miriam Allred (06:51)
Our whole cupcake's about to go up in flame.
Meghan Willson (06:52)
This is on fire. ⁓ and so we were really lucky to have we were lucky to have an executive team and and a board that really paid attention and and allowed for us to make big moves. And ⁓ we spent 18 months building a brand new data warehouse. We ⁓ actually moved quite a few critical pieces of architecture.
And then we also consolidated into one. And ⁓ we did all of this movement and it while trying to keep the company running and the users from not like trying to kill us, right? Like they were like, why are we changing so often? And we were like, we promised we have a vision with a better tech stack, we could actually we could actually improve our individual niche market. We kind of had a couple of pieces of things.
That didn't fit us very well, so we wanted to rebuild.
Miriam Allred (07:53)
And a lot of this rebuilding was around the time that AI was coming on the scene. So were you anticipating okay, we've got to get a lot of this like foundational stuff in order to anticipate AI or was this even before AI was really coming on the scene? We the force that it's in right now. We
Meghan Willson (08:06)
would
have done it, ⁓ but I I wanna say that was looming over ⁓ the tech team's heads. We had we definitely had an executive group that was ⁓ eager to get AI. ⁓ there was ⁓ you know like I would joke, a little bit just because I like to make light of things and I was like, come on guys, we need to get the AI. But no, in all honesty, we needed to make these moves and we we did have like we had the
Along the way we hoped that it would lead to good AI.
Miriam Allred (08:42)
Yeah. So let's talk about care site AI that you're building because I think it's really fascinating. You guys have a lot of like complex diagnoses that you're seeing every single day. And so you want it to build an LLM that has access to all of this information so that you can do something with that information. ⁓ talk about building the LLM. Is that something that you guys have done solely in-house or you have contracted out companies that have helped you with that?
Meghan Willson (09:09)
We definitely have help. ⁓ w it I mean we have been working with our partners, they're called Questa partners, ⁓ and they have been com instrumental in our journey. In fact, ⁓ you know, when this was this was truly a data first AI second project that CareSite AI was. ⁓ we had this
thought that we were just gonna try to jump in and solve for hospitalizations. And like by the way, we've been trying to solve that our entire career. So it was really, it was really complicated. ⁓ we had different diagnoses, different service level levels, different complexities, and we had the thought of we honestly we were gonna we were leaning into trying to do too much at once.
Our advisor said we were trying to boil the ocean, ⁓ which is, you know, not likely that we would have enough fire and enough time to boil the whole ocean. And so we we we took that take one slice at a time, boil just one cup of water approach. ⁓ and in the data we found really clear ⁓ once we started to look at it in that way, we saw very clear indicators of where to go. ⁓
Miriam Allred (10:28)
And when you say you wanted to solve for hospitalizations, what do mean by that exactly?
Meghan Willson (10:32)
Yeah, so look, we're not we're not gonna solve for for all of hospitalizations and we we have no intentions to, but we felt like there were opportunities to prevent one. And you know w you know our clinical team, when they spoke with us, they were so passionate about saying, look, when our patients, and sometimes they would even say, like our kiddos, when they go to the hospital, they go for weeks and ⁓ sometimes we're we
We're not we're not sure if they're even going to come back. ⁓ and so we thought and they and then they would tell us this other thing that was really important, which was they actually had the notion that their patient was getting sick before critical signs were failing. ⁓ so they would say, look, our you know, we're increasing medication to manage this patient. We're having less tolerance to their tube feedings.
And so the thought that we had and that we had with the data scientist team was is there indicators in in the system that they're putting in there that could tell us that? Mm-hmm. The nurse in the field knows, but but could we use the data in the software to kind of prove that case? If we could, then we would give an earlier flag and maybe treat this sooner. And and you know like if you treat any diagnosis sooner, you just have a better outcome. so that's what we wanted to solve. We were
we wanted to just solve like this this ⁓ scenario and use case where the patient is getting sick but we didn't have the tools to to measure how sick, how bad are the trends, are they worsening or improving.
Miriam Allred (12:12)
Mm-hmm. And I guess my question is have you cracked the nut? Like have you solved for it? Have you do you have enough data and is the is the technology working well enough to where you can like predict this type of thing?
Meghan Willson (12:22)
Yeah, yeah. On on a clean data set, we are a hundred percent predicting our hospitalizations. So on our model, for our hospitalizations, we can give three to five days ahead of schedule time that trends are ⁓ the trends need to be considered. And it's looking at these vitals at at each each person's own ⁓
their own experience. So, you know, s one patient might just be fine in a high fluctuation of respiratory rate, and another patient is not fine at that level. And so we look at each patient at their own level and measure their baseline and then measure each of those vitals against that. Okay. So yeah, we cracked it.
Miriam Allred (13:07)
It but it sounds like it's a combination of data sets, one being the vitals, but then also layered in with the clinical documentation, the notes, the the qualitative data, I'm guessing.
Meghan Willson (13:20)
Hundred percent correlation and and and which is better news for us is you know typically clinicians actually document the things that they're seeing that they actually know is the most critical. So we get a little win there because as you might know, we struggle to get our data and data quality. But ⁓ yeah, yeah, we we get their documentation in and and measure their trends and then that's a feedback loop. So ⁓ that's been really nice for us.
Miriam Allred (13:46)
But I'd imagine the the outcome is only as good as that data. Like you can't it couldn't be solely on the vital data, like that's a key part of it, but all of that like clinical like enrichment of information is as vital. And were you as doing a good job at that or if you had you had to significantly improve the documentation piece to get these types of outcomes? Yeah.
Meghan Willson (14:06)
That is a loaded question. Like, let's just say we definitely feel that we are going to be stronger and stronger the better our ⁓ documentation and data quality is. So the more data that's coming to us, the better this AI model is gonna be. but we didn't lean into encouraging change right away. Change came after they realized in working with this that the more data and better data that went in, the
the better their flags and their their experiences was coming out. ⁓ but th the truth is is that we did a lot of data cleaning and we did a lot of work just to find ⁓ for instance not all patients get certain vital readings. So we did a lot of work to actually just math, you know, math out what was important and what wasn't.
Miriam Allred (14:57)
And analogy coming to mind is like give them a drink of water and then lead them to water. So you showed them if this is all done right, this is the level of information that we can give back to you. Therefore it like motivates them and incentivizes them to do better documentation. And it sounds like that's the approach you took was let's actually show them the outcome and then that will help them document better.
Meghan Willson (15:17)
Yeah, and I'll even harden your case. They actually see it in a dashboard. So like they're looking the AI model is saying, there's a flag here, you should look at something, so human in the loop. But then when they when they go in to look, they'll see the their own gaps. So they can like see where there were gaps along the way. ⁓
Miriam Allred (15:36)
And what are the comments you're getting from your your clinical staff? What do they s what do they say? What do you actually hear from
Meghan Willson (15:43)
Yeah, no, ⁓ so it was I just want to be honest and transparent. It was a it's been a journey ⁓ with a really close part partnership with our clinical team and clinical leaders. ⁓ the very first launch of our pilot, we had a really standard bell shirt bell-shaped curve of people who loved it, people who hated it, and then everyone else was right in the middle. ⁓ we actually spent a ton of time focused on the the percentage of patients
users that ⁓ maybe like didn't like it or didn't believe in it or felt like it was extra work. So we spent a lot of time on that group. ⁓ we actually had to go back and do additional modeling work to to zone in they they had no tolerance for like a false positive. ⁓ and so we had to go back in and do a lot of work. But you know when we got that group, they were the ones that
Miriam Allred (16:44)
Became the promoters, I bet.
Meghan Willson (16:45)
They
became the promoters. That's right. ⁓ it's such ⁓ it like in real time for us, honestly it it it actually there it's a feeling that you get when you actually can get that s user set to to agree that this is a helpful tool. ⁓
Miriam Allred (16:47)
Happens.
Can I ask a weird question? Yeah. Does that bell curve have any correlation to age?
Meghan Willson (17:10)
Probably. Does it probably I didn't no, I didn't look at that. ⁓ I we made some assumptions with age. We also made some assumptions with just your general notion of AI and technology and like that like by the way, like I've I've implemented tools that they have not accepted. ⁓ and then, you know, when the you you do you do a project, you send the dev team out, you get something back three months later and your clinical team is like, Why did you just give me more work? And so like I've
failed them my on my own accord and sometimes. So this this wasn't that project. They were they were together along the way and I forgot where I was going with that by the way. But yeah, I I a s I made the assumption based on what we got and this is just an assumption that there was some high and low tolerance to tech.
Miriam Allred (18:00)
Okay. Last last question here to kind of wrap up. Mm talking to all of these leaders that are trying to bring office team and clinicians around to better documentation, enriching documentation to drive these types of outcomes. What's one thing that they need to I wanna start doing and stop doing? You know, think of all of this pilot that you've gone through. What's something that they need to stop doing? What's something you were maybe like trying to accomplish that just you shouldn't have done, that you learned the hard way? Like think of these providers listening to this. What do they need to stop?
trying to do.
Meghan Willson (18:32)
Yeah,
this is this is tough for me. ⁓ I think that we h had been super hyper focused on workflow and we were focused on that bec because we needed certain things to come out of that. But that left us completely blind to what was helpful and what was intuitive. So yes, you want workflow. ⁓ so f for sure do that, tackle that, but tackle it ⁓
Instead of trying to get to where you're ending, like where are you starting, take it sm one piece at a time.
Miriam Allred (19:06)
Don't boil the ocean. Yeah.
And and then what should they start doing? I guess my my one takeaway is don't give up on the detractors because they can become your best promoters. But anything else that you would advise these operators to start doing when they're thinking about getting this like adoption?
Meghan Willson (19:25)
Lock arms, if you are developing a product that is going to impact a clinical person, bring your clinical team along. I mean, really, really bring them along. And what I mean is we we brought them along so tight every week. We showed them every single piece of technology, even if it made sense or didn't make sense. They actually started to speak like technologists and we started to speak like clinicians, but I think it matters. bring the whole group.
Right, like even don't don't lose that opportunity just because we're all like super busy. bring bring it together.
Miriam Allred (20:02)
And let them be a part of like the messy evolution so that you're not like three steps ahead and then having to like bring them back. It's like they need to be a part of every step, the good, the bad, the ugly, so that there's not this like catch up and this disconnect.
Meghan Willson (20:15)
You are totally right. We had a notion of like we just want them to be proud of what we just made them and we didn't bring them into some of when it didn't go well, and then they weren't able to talk about it. So we did actually change and we've we've done this with a couple of technical products, but we did actually change things along the way. And funny, our technology leaders didn't have to tell clinicians that clinical people told their own clinical people. Which is really great.
Miriam Allred (20:41)
Yeah, and I love what you said. Then you all start speaking the same language and honestly in o in all of these businesses we all just need to be speaking the same language. But the larger we get, the more decentralized we get, the more specialized we get, we kind of lose sight of sight of we're all working to accomplish the same goals and we all need to speak that same language.
Meghan Willson (20:58)
Yeah, and there's nothing more human than ⁓ failures along the way. So I I'm glad you shared that. But yeah, I think that part of part of that messy and failure actually can bring a team together.
Miriam Allred (21:11)
trauma bonding over AI.
Meghan, this has been so awesome. Thank you for joining me live. Thank you for having me. We're at the forum. We'll go ahead and wrap here, but thank you so much. That was awesome.
Meghan Willson (21:21)
Okay,
great. Yay!