Latest / The Clinical Realist / Trust Debt: The Bill Comes Due at Renewal, with Arvita Tripati

Transcript

Trust Debt: The Bill Comes Due at Renewal, with Arvita Tripati · The Clinical Realist

• 0:02
  SPEAKER_01: Somewhere in your organization right now, there's an AI tool in year two of its contract, and it's performing probably worse than the day you signed.
• 0:09
  SPEAKER_01: And it's not because the model changed.
• 0:11
  SPEAKER_01: The model is exactly what it was.
• 0:13
  SPEAKER_01: But your patients changed, the population drifted, case mix shifted, a referral pattern moved.
• 0:18
  SPEAKER_01: And the thing that was validated against last year's patients is quietly making decisions about this year's.
• 0:25
  SPEAKER_01: So here's a part that should keep you up at night.
• 0:27
  SPEAKER_01: Almost nobody is measuring that.
• 0:29
  SPEAKER_01: Clinical ops assumes the vendor's watching, and the vendor is watching, just not against your patients.
• 0:34
  SPEAKER_01: So my guest today has a name for what accumulates in that gap.
• 0:38
  SPEAKER_01: She calls it the trust debt.
• 0:39
  SPEAKER_01: And she has watched a crew from three different chairs.
• 0:42
  SPEAKER_01: Arvita Trapavati is the founder and principal of the HANA Labs, with nearly two decades bringing regulated products to market across digital health, medtech, diagnostics, clinical research, and AI.
• 0:52
  SPEAKER_01: And she's run product and quality, which means she was accountable for what happened after the health system said yes.
• 0:58
  SPEAKER_01: She's run security as a CISO, which means she's the person who killed the deals that had already been promised to the board.
• 1:03
  SPEAKER_01: And her deployment work includes rolling Cardia Mobile out across all 15 NHS academic health science networks, plus it's under implementations at places like Sloan, Cleveland Clinic, and Mayo.
• 1:14
  SPEAKER_01: So she has sat on both sides of the table and has watched buyers do three specific things to themselves.
• 1:20
  SPEAKER_01: Arvita, welcome to Clinical Realist.
• 1:22
  SPEAKER_01: I'm so glad to have you.
• 1:23
  SPEAKER_00: Thank you so much for having me, Sarah.
• 1:25
  SPEAKER_00: It's great to be talking with you today.
• 1:35
  SPEAKER_01: So I want to start with the term itself because you know language is doing real work here.
• 1:40
  SPEAKER_01: So tell me what trust debt for a CMO is.
• 1:43
  SPEAKER_01: Like if they signed two years ago or last month and genuinely believe everything's fine, what are they actually carrying on their books that they haven't written down?
• 1:53
  SPEAKER_00: Yeah, so I think that it's it there's multiple facets, right?
• 1:58
  SPEAKER_00: There is how a vendor has uh they could have built the product uh completely correctly, right?
• 2:09
  SPEAKER_00: And you could have validated it and it worked at that moment, but there's all the little things that happen as the system learns from your behavior that end up reinforcing it in potentially the wrong way.
• 2:23
  SPEAKER_00: I mean, imagine like your Netflix account, right?
• 2:26
  SPEAKER_00: You maybe one month you're really into action comedies, right?
• 2:32
  SPEAKER_01: And uh I have a ton of kids, they mess up my Netflix account every day.
• 2:36
  SPEAKER_01: If I need to watch another anime version of whatever, like, come on, people.
• 2:41
  SPEAKER_00: Exactly.
• 2:42
  SPEAKER_00: And so like you get a bunch of yeses, it learns that that's your behavior, and so it stops showing you all these other sorts of things that uh you might want to see.
• 2:50
  SPEAKER_00: Same way, imagine you're let's say an infusion center and you have scheduling AI implemented, and it's uh learning that oh, okay, well, certain types of folks don't pick up the um uh phone as quickly or don't like uh text back uh as quickly.
• 3:11
  SPEAKER_00: And so it's learning that those scheduling spots are maybe um it's optimizing for quick fill of scheduling, right?
• 3:19
  SPEAKER_00: And so it's going to deprioritize some of those folks for future reach out, right?
• 3:24
  SPEAKER_00: And when you zoom out then over the course of like a quarter, six months a year, you might end up uh finding that you're not serving those patients that are um still part of your population, part of your mission.
• 3:41
  SPEAKER_00: Um those patients that maybe have more strict work schedules, maybe they're say uh retail or hourly employees, and they don't they're not allowed to carry their phones with them and they're only able to respond, you know, at shift breaks.
• 3:57
  SPEAKER_00: And so is that really how you would want to have your organization put forth?
• 4:03
  SPEAKER_00: If not, then maybe it's appropriate to have like governance procedures where say on a monthly, on a quarterly basis, you're looking at how things are drifting and making sure that the outcomes that you're actually achieving are aligned with the way you intended it to uh operate.
• 4:23
  SPEAKER_01: I'm building out a course right now for Cornell.
• 4:27
  SPEAKER_01: It's a whole certificate on healthcare AI.
• 4:29
  SPEAKER_01: And this morning I was working with my um assistant on monitoring and optimization because it's the part I think people think is an add-on.
• 4:39
  SPEAKER_01: And they have all the best things going on for day 30, 60, 90.
• 4:43
  SPEAKER_01: But then when it comes to actually operationalizing it and considering how to monitor, I think first there's a gap technically in how they monitor, but I think there's also a people gap in how they monitor.
• 4:55
  SPEAKER_01: What do you think about the people gap of monitoring?
• 4:58
  SPEAKER_00: Yeah, um I mean people like to pattern match, right?
• 5:04
  SPEAKER_00: And it can be really uh uh people and so it can be really like tempting to fall into old judgment patterns and not like uh not question uh things.
• 5:17
  SPEAKER_00: And so it can be really um really difficult if you're not to to always be at that edge and always be asking that kind of like adversarial uh sort of question to always be you know asking the pointy question.
• 5:33
  SPEAKER_00: Um because you know um a lot of people are people pleasers and and want to get along and it it can uh so you need to be able to create the culture that makes it uh safe to ask those pointy questions, right?
• 5:51
  SPEAKER_00: And I think that this gets into how organizations even bring technology uh in, right?
• 5:57
  SPEAKER_00: If you when you're doing your pilot, when you're doing your initial setup, are you only having the champion and the folks who are excited about technology participating it?
• 6:08
  SPEAKER_00: Or are you bringing in the skeptic?
• 6:11
  SPEAKER_00: Are you bringing in the person or the folks who you know maybe don't have a ton of time to do the heroics to make sure the thing works, right?
• 6:20
  SPEAKER_00: Because then you'll be able to get some of like their realities incorporated into the workflow and see what those you know edge cases are.
• 6:31
  SPEAKER_00: We like to call them edge cases, right?
• 6:32
  SPEAKER_00: But they're not really edge cases, they're just the reality of usage, right?
• 6:37
  SPEAKER_00: And so, yeah, I I think all of this plays into trust, right?
• 6:42
  SPEAKER_00: The the um folks who aren't there for um who who aren't necessarily champions, bringing them in and getting that, you know, um uh other point of view, having that psychological safety to be able to um uh allow pilots to fail, not saying we must bring this in.
• 7:02
  SPEAKER_00: Um, and then you know, uh with that reinforcement, there's also actually the the flip side, right?
• 7:08
  SPEAKER_00: You have the reinforcement of always saying uh yes and you know the system optimizing for it, but then you also have like the the ground truth change.
• 7:19
  SPEAKER_01: The culture of no.
• 7:20
  SPEAKER_01: The culture of no is what I say.
• 7:23
  SPEAKER_00: Well, there's the culture of no, but uh um, but there's like nothing is static or very little few things are static, right?
• 7:31
  SPEAKER_00: Where you have like all the different payers changing their schedules at different times of the year, right?
• 7:37
  SPEAKER_00: And so, you know, how is your governance uh system, like even if your model is is working correctly and not necessarily drifting, the overall system that you have it operating within, like that is part of that becomes part of the product.
• 7:52
  SPEAKER_00: And I don't necessarily know that that is um the way people have to date thought about it.
• 7:58
  SPEAKER_00: And when your system is part of the product, that changes fundamentally what your feedback loops need to be uh versus like if when it's just the technology is the product.
• 8:11
  SPEAKER_01: So yeah, I think it's interesting because when we talk about the population, nobody really owns drift for their population.
• 8:19
  SPEAKER_01: And so it's hard to put it on org chart, it's hard to put it on a racy, it's hard to do anything.
• 8:24
  SPEAKER_01: But who do you think should own it?
• 8:26
  SPEAKER_01: So should it be a CMIO, a CIO, quality?
• 8:30
  SPEAKER_01: Who should own the population?
• 8:33
  SPEAKER_00: Oh well, I think that it needs to be, yeah, the CMIO personally.
• 8:38
  SPEAKER_00: I think that there needs to be like uh a partnership, right?
• 8:44
  SPEAKER_00: There should be a governance committee.
• 8:46
  SPEAKER_00: Hopefully, there's uh a data council and maybe even like a chief data officer.
• 8:52
  SPEAKER_00: But I do think that um, you know, fundamentally it needs to be the uh clinician and uh weighing in on it and being able to talk about what is or isn't appropriate.
• 9:05
  SPEAKER_01: But that's just my I think it's tricky because, like you mentioned, the payers are changing, but our populations are changing.
• 9:11
  SPEAKER_01: Um, I know that for some of the hospital systems I'm working in right now, you know, there's not enough babies being born for certain Medicaid programs.
• 9:20
  SPEAKER_01: And it's hard because your population for some reason just isn't having babies.
• 9:24
  SPEAKER_01: And you're like, well, what's going on here?
• 9:26
  SPEAKER_01: And it means that things are changing.
• 9:28
  SPEAKER_01: Your demographic might change, the popularity of different procedures and things like that by the public may change.
• 9:34
  SPEAKER_01: And so having that finger on that pulse today, I think, is even more important than it was in the past because things are changing so fast that you can't just do it once a year and think you're good.
• 9:46
  SPEAKER_00: Yeah, and I mean it may end up meaning that you divest a service line, right?
• 9:53
  SPEAKER_00: And uh spin something out.
• 9:54
  SPEAKER_00: And so, I mean, there's so many different business implications with this, and I think it really needs to uh be done with like uh that uh clinician partnership and and and oversight.
• 10:08
  SPEAKER_01: So when we think about the clinician piece, I like having CMIO, CNIO in charge of the population itself.
• 10:16
  SPEAKER_01: Ravita, I know that you were a CISO running the reviews that essentially kill products and things coming in.
• 10:24
  SPEAKER_01: What do you think is the moment when you as a CISO know it's over?
• 10:29
  SPEAKER_01: Like what does it feel like in the room?
• 10:31
  SPEAKER_01: What are those things that just scream at you and you're like, all right, people, we're out.
• 10:35
  SPEAKER_01: I don't care what else they say to me.
• 10:39
  SPEAKER_00: Oh gosh.
• 10:41
  SPEAKER_01: So I don't mean to take all your secret sauce away, but you know what I'm talking about.
• 10:47
  SPEAKER_00: Yeah, you know, there's a moment where people are being a little bit hand-wavy and they're like, oh, well, you're they're they're talking about, oh, they're SOC 2.
• 11:01
  SPEAKER_00: And like, oh, okay, great.
• 11:04
  SPEAKER_00: Um what does that actually mean?
• 11:07
  SPEAKER_00: Like, can we see the report?
• 11:08
  SPEAKER_00: And then you find that, oh, they're talking about their data center is SOC2, and then oh, okay, it just like starts eroding that trust.
• 11:18
  SPEAKER_00: And I mean, it's great, their data center is SOC2, but like presenting your company as being that when you know really it's maybe uh a layer or two uh back that that that that just starts um eroding the trust.
• 11:35
  SPEAKER_00: And then it's then you shift kind of mind mindset and gears into like, okay, if they're if they're being shifty about this, what else are they being shifty about?
• 11:45
  SPEAKER_00: Um and so I but I'll but you know you can't be too rigid on that, right?
• 11:52
  SPEAKER_00: There or else you might end up uh, I guess, boxing yourself out of uh new technologies.
• 12:00
  SPEAKER_00: And so I like to think about like what are they ready for now?
• 12:04
  SPEAKER_00: And like what is that decision that we need to uh um enable?
• 12:10
  SPEAKER_00: Um and so like okay, well, maybe they are ready for um a small-scale um rollout with uh one client or one department, right?
• 12:23
  SPEAKER_00: And how do we, you know, how do we use that to minimize risk to us?
• 12:30
  SPEAKER_00: But then how do we also think about that?
• 12:32
  SPEAKER_00: So we're building the evidence that maybe not in, you know, a month, it might not maybe not even be six months, but that way we can build trust so that way, like in a year we can go ahead and have the evidence to be able to roll them out, assuming that they've done their part uh with getting appropriate independent certifications, but assuming that that we can then roll them out across the org.
• 12:58
  SPEAKER_00: And so those are the sorts of things that I uh like to think about.
• 13:02
  SPEAKER_00: Also, when they don't have anybody willing to talk about security architecture, that is another one that is like, oh, okay, well, uh you're just kind of making it up as you go along, aren't you?
• 13:15
  SPEAKER_01: It's interesting because I know on the clinical side, if the clinicians get included in the process too early or too late, it's never the right time.
• 13:24
  SPEAKER_01: Um, everyone says, oh, bring us in early.
• 13:27
  SPEAKER_01: Um, at the same time, that can be tricky too.
• 13:29
  SPEAKER_01: When it comes to security, I think it might be the same exact problem.
• 13:33
  SPEAKER_01: If they bring everyone in too soon, then we might be limiting ourselves.
• 13:36
  SPEAKER_01: But if we bring people in too late, then we're gonna be trying to mitigate in the hardest ways possible.
• 13:43
  SPEAKER_01: Um, there's nothing worse than having a decision that's made without the security piece taken care of.
• 13:48
  SPEAKER_01: Have you experienced that before?
• 13:51
  SPEAKER_00: Oh, yes, uh uh definitely.
• 13:54
  SPEAKER_00: Um so the process that I ended up putting in place to like get around that and um get ahead of it, I guess, is really more of okay, when we're intaking what we want to the the what we want the system to do when we're even first starting our procurement, like okay, here are the gradation of things that we're willing to accept.
• 14:18
  SPEAKER_00: You go off, you source, you figure it out, you you you identify the things that you're excited about, go for it, experiment.
• 14:27
  SPEAKER_00: And then, all right, when we're starting to get closer into implementation, then you know, you have this rubric then of like, okay, well, it it's a super sketchy startup.
• 14:39
  SPEAKER_00: We're gonna be, you know, uh doing X, Y, or Z.
• 14:44
  SPEAKER_00: And um we're we're only going to be able to like do phase one with them, uh, and then we'll uh let that run, and then we'll be able to, you know, if they're successful, then expand to phase two.
• 14:58
  SPEAKER_00: And so that worked well for me at um a couple of different um uh what's it called of those supply chain uh companies that I was at.
• 15:08
  SPEAKER_00: And so um I like to uh use that process.
• 15:12
  SPEAKER_01: I like it.
• 15:13
  SPEAKER_01: Now here's the problem though everything we've talked about so far assumes an organization that has a security team, an informatics team, a legal team, a governance committee, and most of American healthcare doesn't.
• 15:25
  SPEAKER_01: We have ASCs and SNFs and specialty groups, and they're building and buying increasingly sophisticated AI-enabled tools and products, and none of them have that bench.
• 15:35
  SPEAKER_01: So what is the minimum viable diligence for an operator with no CISO and no informatics?
• 15:42
  SPEAKER_01: What do you think they should be asking?
• 15:44
  SPEAKER_00: Oh, that is a really great question.
• 15:47
  SPEAKER_00: Uh so I think that that's where there's like we need to ask uh folks like Chai uh to do better, right?
• 15:55
  SPEAKER_00: Like there's all these different organizations out there that are like putting out standards for um for healthcare, but they're really only talking to big health systems and you know the folks that do have those really um robust teams.
• 16:14
  SPEAKER_00: Um so yeah, unfortunately it does come down to like uh working with consultants to go ahead and write size.
• 16:23
  SPEAKER_00: Um and I don't know that there's a really uh great answer.
• 16:28
  SPEAKER_00: I would say at the minimum, probably ask for the vulnerability scan and make sure that they're uh doing um at least you know on a every uh release basis a uh scan.
• 16:42
  SPEAKER_00: Um and if you get like a deer in the headlights uh look from them, that's probably uh you know a good reason to uh press pause and maybe uh ask them to do their homework before re-engaging.
• 17:01
  SPEAKER_00: Um I remember I was talking to one of the frontier labs.
• 17:06
  SPEAKER_00: Like you would expect that like one of those big uh players in AI would uh have this uh hopefully a little bit more buttoned up.
• 17:16
  SPEAKER_00: But I was talking to uh the head of product at uh one of the frontier labs and uh we're talking about how are they doing evaluations and uh testing their AI uh to make sure that it's uh repeatable.
• 17:32
  SPEAKER_00: And uh they do one test run and it's a broad test uh set.
• 17:41
  SPEAKER_00: I don't know about this.
• 17:42
  SPEAKER_01: I think this is a crooked glasses moment here for that one, Arvita.
• 17:46
  SPEAKER_01: That doesn't sound good.
• 17:47
  SPEAKER_01: That sounds bad.
• 17:49
  SPEAKER_00: It does sound bad.
• 17:50
  SPEAKER_00: It is bad, right?
• 17:51
  SPEAKER_00: I'm like coming from that quality perspective, and I'm like, oh well, even in quality for like what the last 60 years, we've had process capability, right?
• 18:00
  SPEAKER_00: You've had to prove you do three different runs, right?
• 18:04
  SPEAKER_00: Uh that's at a minimum to demonstrate your repeat your repeatability.
• 18:08
  SPEAKER_00: One is a point, two is a line, three is a trend, sort of thing, right?
• 18:12
  SPEAKER_00: And so um, yeah, to hear that they are just doing one test run and this is this is like, oh, okay, well, you're you're just talking consumer, you're talking like fluffy stuff.
• 18:25
  SPEAKER_00: No, this is this is that regulated portion of it that is serving, you know, financials, uh health.
• 18:35
  SPEAKER_00: It's just like all right.
• 18:36
  SPEAKER_01: That's all chest pain over here.
• 18:38
  SPEAKER_01: My goodness.
• 18:39
  SPEAKER_01: Yeah.
• 18:39
  SPEAKER_01: But at the same time, I guess if it was not working when it goes out to the users, there would be huge backlash.
• 18:46
  SPEAKER_01: So I'm hopeful that maybe it's working with one test run instead of three.
• 18:51
  SPEAKER_01: I don't know.
• 18:52
  SPEAKER_00: I guess maybe.
• 18:54
  SPEAKER_00: Um then again, uh maybe that is part of why sometimes when they release a new model, sometimes it seems like it's a little bit um a little bit rough around the edges.
• 19:08
  SPEAKER_00: Uh let's call it.
• 19:09
  SPEAKER_00: So, but coming back to like what can what can folks do?
• 19:13
  SPEAKER_00: I think the just making sure that you have a clear idea of what the outcome is that you're trying to uh achieve.
• 19:22
  SPEAKER_00: Um, the whole user acceptance process that hasn't changed just because it's AI now.
• 19:31
  SPEAKER_00: It just means that you need to be like clear about what those what those outcomes are as opposed to just relying on your vendor-provided test scripts, because your vendor-provided test scripts are going to just um be exercising what they want to exercise, not necessarily what's right for your population and for your specific business.
• 19:57
  SPEAKER_01: Yeah, that's a happy path they're putting together there for you, which hopefully will look great.
• 20:02
  SPEAKER_01: Um, I think the other thing is we think of user acceptance testing and doing UAT at the beginning of things.
• 20:10
  SPEAKER_01: But when I think about how monitoring really needs to be done for you know AI products and tools, UAT might be something we want to do later too.
• 20:22
  SPEAKER_01: Has the model drifted and we see it before we get a bug in a ticket somewhere?
• 20:27
  SPEAKER_01: Like, how are we accepting it a month from now or three months from now or a year from now?
• 20:32
  SPEAKER_01: And I don't know if folks have considered that as a UAT process.
• 20:38
  SPEAKER_00: And I think that gets back to like that governance and monitoring, right?
• 20:41
  SPEAKER_00: Like with your um with that example we were talking about earlier with uh the scheduling, right?
• 20:48
  SPEAKER_00: Uh is that something that like on a quarterly basis that you want to have uh run uh a test uh on and be able to um uh assure yourself that it hasn't drifted?
• 21:03
  SPEAKER_00: Is that something that you should ask your vendor to help you set up and like have be part of a deliverable, like some automation suite that lets you do that um uh against like a gold standard set?
• 21:17
  SPEAKER_00: That way you have continued assurance.
• 21:19
  SPEAKER_00: And I think that as we move through um adoption, like some of these expectations will uh become you know more clearly articulated and hopefully part of the vendor artifact package that gets delivered as part of um any um any go live.
• 21:38
  SPEAKER_01: I think that um one piece that I think is really important, especially for non legal people, is being able to read the contract too, and what is the vendor accountability clause look like?
• 21:49
  SPEAKER_01: So what have you actually signed?
• 21:51
  SPEAKER_01: Who monitors it against what population, what frequency, what is the vendor responsible for?
• 21:58
  SPEAKER_01: What are you responsible for?
• 22:00
  SPEAKER_01: And I think a lot of folks just assume the vendor is going to take care of them, but if it's not in the contract, it doesn't exist.
• 22:06
  SPEAKER_01: And you might not have the robust data science team or statisticians or whatever it is to take your data and disaggregate it or put it in form so you can see what's really happening.
• 22:17
  SPEAKER_01: And I think this especially catches people when an executive decides we're doing a product and we're taking it in no matter what, and we haven't been able to do the due diligence well.
• 22:29
  SPEAKER_01: Or you have a team that just hasn't done this before and they push it through not knowing the I'd say positives and minuses of vendor accountability.
• 22:42
  SPEAKER_00: Yeah, I that is such a great point.
• 22:45
  SPEAKER_00: There's this um sandbox actually that Delaware is uh coming out with for AI.
• 22:51
  SPEAKER_00: I don't know if um if uh that's uh been something you've come across yet.
• 22:56
  SPEAKER_00: Um they're just piloting it this summer where they're talking about basically forming a um a special type of LLC for AI, and with that the liability for so you might have as your company, you might have uh a Delaware corp, but then your AI product specifically would have its own uh corporation, and then the liability would be all wrapped up uh within that AI, and there would be like some amount of monies like escrowed um with insurance policy essentially.
• 23:36
  SPEAKER_00: Yes, and so it's a really interesting concept, and this is where like we drift out of my uh uh core competencies that I'm just like into like legal uh stuff, but this is I find it really interesting because like if you look back at um at state of Delaware decisions around um CAREMARC and like liability, you you'll find that they haven't actually like held too many companies accountable.
• 24:09
  SPEAKER_00: And like what does that mean then for the AI if they're following that same set of principles?
• 24:19
  SPEAKER_00: Um, like will that escrow ever get actually tapped?
• 24:23
  SPEAKER_00: What sorts of companies actually have enough money to be able to set up that escrow account?
• 24:28
  SPEAKER_00: And so are they like basically biasing towards uh larger vendors?
• 24:33
  SPEAKER_00: There's all sorts of different implications there, and like what does that mean for innovation?
• 24:37
  SPEAKER_00: Anyways, it's really interesting seeing that, and then just this week, FDA came out with a discussion paper on um clinical AI, and um there was this uh very interesting uh piece in there, like they're also worried about drift, uh, but they're also starting to uh push back on this concept of um just having labeling in the AI around like, oh, this is just for information, consult your doctor um uh as being a reasonable um liability shield, I guess.
• 25:18
  SPEAKER_01: And so kind of like making it a wellness product instead of a healthcare product.
• 25:23
  SPEAKER_00: Yes, yes.
• 25:24
  SPEAKER_00: And so I find that really interesting and like where that conversation is uh gonna end up going.
• 25:30
  SPEAKER_00: So yeah, it's definitely interesting times.
• 25:34
  SPEAKER_01: That is I'm actually gonna look into that Delaware piece.
• 25:36
  SPEAKER_01: I'm thinking about all the other states and why someone would do their LLC in Delaware versus elsewhere and how that might absolutely choose a certain demographic of organizations.
• 25:47
  SPEAKER_01: So I'm gonna look into that for sure.
• 25:50
  SPEAKER_01: Um, you know, we are almost done today, but you know, we probably have a CMO who signed a five-year AI contract last month and is listening to this on a treadmill right now.
• 26:01
  SPEAKER_01: So what's the first call they need to make Monday morning and who are they calling?
• 26:09
  SPEAKER_00: Uh I think if they have their um, if they have like a chief AI officer or if they have uh if they don't have one of those, then maybe they're a CIO or the legal officer, and really looking into okay, what is our plan for monitoring?
• 26:29
  SPEAKER_00: And do we have our outcomes adequately defined?
• 26:36
  SPEAKER_00: Because I think that people for people know that it's all about outcomes, but sometimes with you know just moving in the human world that we have, we're we're taking a lot of it as inference and we're not being very explicit and uh defining that.
• 26:57
  SPEAKER_00: But for AI, you need to have it carved into stone, right?
• 27:01
  SPEAKER_00: You can recarve the stone, you can change your mind on things, you can, but you need to have it defined in order for it to be able to um do what you want it to do to achieve the outcome that you want.
• 27:13
  SPEAKER_00: And so if you haven't, if you haven't actually defined an outcome, and if you don't know what your monitoring is, you're not gonna be able to say, Oh, we're not getting that outcome.
• 27:24
  SPEAKER_00: We need to iterate.
• 27:26
  SPEAKER_01: Well, it's been wonderful to have you.
• 27:28
  SPEAKER_01: Now for the lightning round, I have two questions I ask everybody.
• 27:31
  SPEAKER_01: So the first one is what do you think everyone in healthcare is getting wrong today?
• 27:38
  SPEAKER_00: Oh.
• 27:39
  SPEAKER_00: Um I think that uh people are overfocusing on uh reimbursement and not enough on value.
• 27:51
  SPEAKER_00: Um take this uh looking uh at it from like really the vendor uh perspective.
• 27:57
  SPEAKER_00: Uh I was talking with a company that uh that was like, oh, we're we're we're going after this one pathway because uh there's um this reimbursement code behind it, but then like in digging into what they were actually doing, it was like, oh, well, reimbursement is like only a three-digit um uh you know money coming in.
• 28:24
  SPEAKER_00: But uh when you look at the value, you're delivering six digits of value.
• 28:29
  SPEAKER_00: So like let's let's go after who who realizes that six digits and and and get a larger slice of the pie.
• 28:37
  SPEAKER_00: And so I feel like I feel like that value versus reimbursement is uh an interesting thing.
• 28:45
  SPEAKER_01: I like that.
• 28:46
  SPEAKER_01: And then last question.
• 28:47
  SPEAKER_01: When's the last time you changed your mind in healthcare?
• 28:52
  SPEAKER_00: I mean that's on a daily basis.
• 28:56
  SPEAKER_00: I keep on learning new things.
• 28:58
  SPEAKER_00: Uh it's it's wonderful.
• 29:00
  SPEAKER_00: Every day I learn something new.
• 29:02
  SPEAKER_00: It's great.
• 29:04
  SPEAKER_01: Well, Arvita Tripati, founder and principal of Ahana Labs.
• 29:09
  SPEAKER_01: Arvita, where should people find you?
• 29:12
  SPEAKER_00: Uh you can please find me on LinkedIn and also at operatinginhealthtech.com.
• 29:18
  SPEAKER_00: Uh, I write a sub stack and publish that.
• 29:21
  SPEAKER_00: Uh, and it's always about that uh confluence of value and uh technology and um and uh regulatory and like how the three can come together uh to help you win.
• 29:39
  SPEAKER_01: Well, I love that.
• 29:40
  SPEAKER_01: So trust debt, model drift, maybe your population is moving.
• 29:45
  SPEAKER_01: If this episode gives you one little tidbit, a question to ask before you sign, send it to the person in your organization who does that signing.
• 29:52
  SPEAKER_01: That's the whole point of the show.
• 29:53
  SPEAKER_01: So I'm Dr.
• 29:54
  SPEAKER_01: Sarah Matt, and this is a clinical realist.
• 29:56
  SPEAKER_01: We'll see you next time.