Building the Data and AI Foundation for Health Plans
19 minAudio
In this episode, Mike Sanky, Vice President of Healthcare & Life Sciences GTM, Databricks, discusses how health plans can modernize data and AI capabilities to manage rising costs, improve utilization insights and strengthen governance for scalable AI adoption. This episode is sponsored by Databricks.
Audio and video play straight from the show's own hosting. PodSubs stores and serves no media.
Transcript
From the show. Tap a time to play from there.
- Hello, everyone. This is Erica Spicer Mason with - Becker's Healthcare. - Thank you so much for tuning in to - the Becker's Healthcare podcast series. - Today, we're going to talk about how health - plans can modernize their data and AI capabilities - to better manage - rising costs and utilization. - And joining me for this conversation is Mike - Sankey, vice president of health care and life - sciences go to market at Databricks. - Mike, it's so great to have you on - the podcast today. Thank you so much for - joining us. It's great to be here. Thanks
- for having me. - Oh, we're glad to have you. And before - we talk a little bit more about health - plan strategy here, wanted to know if you - could share just a little bit more about - your work, your background, - whatever you think would be helpful for listeners - to know going into this conversation. - Yeah. So I guess briefly on Databricks, Databricks - is the data and AI platform. We work - across most of the Fortune 500. - In health care and life sciences, we have - more than 1,500 - customers - who are using Databricks on a monthly basis,
- and we do quite a bit of work - with health plans in The US that'll obviously - be the focus of - today's conversation. I've been at Databricks five and - a half years, and my role there - is really to take what we built, which - is a horizontal technology platform, - and help customers adopt it and really extend - it into - all of these industry specific use cases and - applications. - And we do that through a combination of - solutions that we build. So, for example, interoperability - libraries around FHIR, which are
- very relevant to clinical data feeds for health - plans, - as well as through the ecosystem - that we cultivate. So there are - a lot of partners who are building on - Databricks. They may have - solutions that are specific to - industry around prior authorization or around fraud detection - that plug in - as well as a number of data providers, - and we connect to different - MCPs like with CMS, - that - really help our customers
- use this technology - in industry. So that's a bit about me. - Oh, great to learn more about you, Mike, - and the work that you're doing. I'm sure - it's keeping you very busy because as we - know, so many organizations are working to either - implement - AI tools or or expand them and and - make sure that their data is ready to - do so. So, - so I wanna start high level here. We - know health plans are heading into 2027 - facing some very steep cost trends. - You know, there's pressure coming from GLP ones.
- There's rising behavioral health utilization, - shifting acuity. - I'm curious to know what you're hearing from - health plans about where their current tools or - capabilities - are currently falling short of some of those - pressures. - Yeah. It's probably worth just spending a couple - minutes on some of those macroeconomic - pressures that you mentioned because they're very real. - There are health plans operating at a financial - loss, including a number of health systems that - are acting as insurers - and trying to manage their own risk pools. - And a lot of the large,
- you know, diversified health services, health plan organizations, - CVS, Humana, United, - Cigna, Centene, - they're all either exiting or downsizing - in the Affordable Care Act market or Medicare - Advantage. - So, you know, many people maybe outside of - the industry think that health plans are just - minting money year over year, - but the reality is relatively minor changes can - swing them from a profit to a loss, - and that's magnified in these larger plans - that have tens of millions of members.
- So we take a plan that has 10,000,000 - members, and let's say they make 1,000 in - profit from every member over the course of - the year. - That's 10,000,000,000. - But - that $1,000, - that could be a single month retail cost - of a GLP one. - It could be just a portion of the - facility fee for an ER visit, which wouldn't - include the physician's fee and the diagnostic test. - So the point being that, again, these utilization - changes across a large member population, they can - really make an impact. - And then we're combining that with the shifting
- regulatory landscape in areas like Medicaid redetermination - and, more recently, the administration - ending, a nearly $10,000,000,000 - subsidy for Medicare Part d. - So - all this means that there's a lot to - react to and a lot of these decisions - that health plans need to get right to - create benefit designs, to accurately forecast utilization, - and to help to manage these dynamic conditions - and changing member health over the course of - a year, and it's not trivial.
- And many of these organizations that we work - with, I think, self admittedly are data rich - but insights poor. - In other words, they have massive amounts of - claim and member data, but they struggle to - process it quickly enough. - They struggle to apply additional intelligence and logic. - So an example of this could be somewhat - basic things like grouping all claims associated with - the knee replacement as a single unit. - And, fundamentally, they struggle to empower business leaders - that need to make these decisions against this - landscape that I talked about
- with the right data and the right time. - So all of those, you know, contribute to - some of the challenges that they're facing. - Yeah. - I really appreciate how you broke down some - of these costs, Mike, and and what it - means on a larger scale. And you mentioned - some of these challenges around - being data rich and info poor. So I - wanna take that a little bit further. - Most plans already have claims data, pharmacy data, - authorization data. There's no shortage of it. But - I'm curious what you see
- break - between having that data - and then catching a utilization shift within the - month that it happens - early enough for the health plan to act. - What is that looking like? - Yeah. I think a a lot of organizations - are stuck in this back and forth between - the technical teams that are building the pipelines - or building the data products or building the - dashboards and the business leaders that are actually - making decisions. - And people largely are trying to self serve - out of dashboards,
- but they're not really built for deep analytics - and - reasoning and trying to - understand some of the root cause and dynamics - and actions that that an organization can take. - And - oftentimes, you'll see these leaders. They'll get frustrated. - They'll be clicking around in the dashboard. They - still don't get the answer they want without - needing a team to go and a technical - team that requires - rebuilding a - pipeline or data table or a dashboard. - And then what's happened is many of these - large organizations,
- they've - grown to have shadow IT organizations, and that - exacerbates some of these challenges in identifying the - source of truth. - So I think a large part of it - is in the fundamental challenges these organizations have - had in actually democratizing - the the use of those data. - Yeah. That concept of democratizing the use of - data, Mike, I think is a really important - one. - And that leads me to ask, you know, - if nontechnical - teams are are accessing
- important data, - how can they safely query live claims data - without, - you know, five people getting five different answers - at any given moment? - Yeah. That... That's one of the biggest areas - we've been spending time as an organization. It's - building the infrastructure - around that context layer that both humans and - AI agents need to succeed. - And if you think historically - of - where that context or that business logic has - lived, it's been hard coded in data pipelines
- or living in disparate business intelligence dashboards or - it's in the head of Sally or Joe - or someone who's worked in the claims department - for twenty years. - And AI agents, in some ways, are even - worse than humans in that they're likely to - just make something up if they don't have - the appropriate context. - So what we've tried to do and help, - our health plan customers with is build the - capabilities to store that context at the data - layer - so that anyone or anything interacting with the - data, whether it's a human SQL programmer or
- business intelligence dashboard - or an AI agent, they get the same - information - with the same shared definitions. - And this really requires both modeled semantics as - well as learned knowledge. - So let's say that someone wants to understand - the PMPM for commercial members who've had a - knee replacement. - We need a business definition of how to - calculate - per member per month. - We may need to store that acronym as - logic. - We also need to define the time period
- of how we calculate that in a consistent - way, whether that's the most recent fiscal year - to date or a rolling twelve months. - We also would need a definition of commercial - line of business. That's probably a grouping of - many different segments - across a health plan, in terms of their - employer business and administrative services. - And all these definitions need to be stored, - accessible, and governed. - And we have features called business semantics and - metric views that support these. - The other thing that we need to get - right is we need to know the correct
- tables to pull from because you can rest - assured that a health plan has dozens of - fields across many tables that are just labeled - claim, - And we've created capabilities to understand - the most relevant data sources and rank them - and help teams certify them. - So relevance, for example, could be based on - how frequently a given table is used to - populate reports for the CFO. - And all these things together, they create that - governed data intelligence substrate to answer the questions - more consistently and correctly. - So in your question of if five people
- ask the same question, - if they're able to refer to the same - business definitions, - logic, and underlying - data, then they should get the same answer - time and time again. - It makes a lot of sense, Mike, and - I really appreciate you walking our listeners through - that. - And I wanna back up a little bit - too and talk a bit more about adopting - AI in the first place. - Adoption of AI tools is essentially universal. I - think it's safe to say at this point, - especially in the health plan space,
- but readiness - for - various AI tools is another story. So where - are you seeing health plans get stuck when - it comes to adoption? - I think that context layer that I described - is probably the single biggest rate limiting factor - right now. - And we believe organizations need four things to - succeed with AI. - We refer to them as the four c's - if you need a memory device. - So context is one. We've talked quite a - bit about that. - And the other three are control, cost, and
- choice. - So let's spend a bit of time on - the governance challenges facing health plans and health - care in general because those really get to - the concepts around - control, costs, and choice. - There's not a single unifying standard for AI - in health care, and states continue to introduce - their own legislation, which really complicates things for - plans - that operate across multiple states. - So organizations - need a robust framework for evaluations, including around - bias - and observability of these models.
- And just like with their human resources, they - need to control what agents access and what - they do with the data. - Add all of this together, and many health - plans, they can't keep pace with the changes - among the Frontier AI labs. - So they really need to focus on building - an infrastructure that preserves model choice and gives - them the ability to work across models, which - also helps them manage cost because it enables - them to send trivial tasks to lower cost - models. - So beyond getting the context and this data - substrate right, I would say most organizations are
- struggling with governance at scale, - as it pertains to controlling, - what agents are accessing, how they're accessing it, - the cost around it, and then model choice - and choice and cost are - very closely intertwined at this point. - Mhmm. - Yeah. And to your point too, Mike, about - there not being one universal standard, - how so many of these regulations - around AI and governance can vary state to - state - makes it very complicated and complex for plans. - I'm curious to know a little bit more
- about the plans that you've seen get the - governance piece right. - What did they have to put in place - before people could trust the data they were - working from? What does that look like? - So, - for this, maybe I can just share a - couple of customer examples. - We hosted our big data and AI summit - earlier this summer. We had so many health - plan customers attend and share what they're doing. - So one of those was Humana, and they've - been a Databricks customer for a while, including - in support of their value based care initiatives.
- And they recently expanded their use of Databricks - to the office of the CFO. - And the entire focus of their presentation was - about rebuilding finance on a trusted governed data - foundation - and operationalizing - AI across these different functions. - And for them, that meant, including domain product - owners from underwriting, from pricing, from risk assessment, - from revenue management, - from areas like cost and expense management, - and having those domain owners define the business - semantics and the metrics to feed a unified
- semantic model - that could be used across, - the office of the CFO. So it really - was this combination of - the technical foundation, but also the intelligence - from their team and from these different functions - to define - that context layer. - The second example I'll share is from Centene, - and they presented on - the gifts of machine agentic governance intelligence, which - they shorthand refer to as the gifts of - the Magi. - And it was all about delivering secure, repeatable,
- and explainable AI. - And this included a lot of different components - for security. It included things like usage tracking - and monitoring, - a single interface to all external models so - they could track - how the models were hitting their data effectively, - as well as comprehensive guardrails. - For explainability, and this gets into some of - the governance and the regulations, they create an - immutable record behind every AI decision. - So they capture and lock everything, including the - prompts, the responses, the tool calls in these
- multi agent systems, - and they have an evaluation process that's actually - a combination of automated scores - and human review. - And they're able to continuously monitor this life - cycle for things like model drift and policy - infractions. - And finally, in terms of repeatability, they've deployed - the scaffolding and created a library of reusable - templates - to ensure consistency and correctness. - So this includes the context layer, which we've - talked about quite a bit - to support how agents retrieve information.
- And what this has done is it's created - an operating velocity that enables them to rapidly - deploy across different areas of their business. - A recent example of this was a new - solution for care managers to support interactions with - members - and help to close gaps in care. - And, Mike, I I really appreciate the detailed - examples from both of those health plans. I - think that's going to be really helpful, - for folks listening in who are, you know, - just in the early stages of this journey - and to have that concrete view of what - this looks like, - and also appreciate the details you shared around
- the four c's as well. And, again, for - listeners, when we talk about AI adoption, readiness - looks like those four c's, context, control, cost, - choice. - Mike, a lot of valuable insight you've given - us today. So before I let you go, - I wanted to kind of get your look - ahead to 2030. - What capabilities around data, AI, and cost visibility - will become table stakes for health plans, and - what should leaders be building toward right now? - Yeah. So I would definitely encourage leaders to - think about building that proper foundation so they
- can achieve that scale velocity. - And if we think about 2030, we're not - that far away. We're looking a few years - out. What will be table stakes for health - plans? - And I'll throw out a few ideas here - that I don't think are that far fetched. - I think many organizations are already on this - journey. - The first is around fraud detection, and I - think we're gonna move to a world where - fraud is detected before the claim is paid - rather than the catch and chase model that, - you know, many organizations operate in today.
- And that would mean models are scoring every - claim prepayment. They're stopping improper payments. - They're embedding AI to do things like automatically - generate letters if they have to request more - information. - And, ultimately, the goal is to reduce fraud, - waste, and abuse to low single digits. Right - now, for many organizations, it's actually in the - high single digits, and that has a major - impact on their operating margins. - The second is around prior authorization. Obviously, has - has, - gotten a lot of attention. - Recently, there are a lot of regulations
- around standards coming out from CMS, - and I'd like to see a world where - prior authorization is 80% plus automated. - Today, it's probably about half of that. It - depends, of course, on the organization and the - book of business. - And this would mean gold carding, - more top quality providers, - having agents auto adjudicate most requests, - and shortening that median turnaround time to under - twenty four hours, even under twelve hours in - instead of days. - The third area is around the payment models,
- and - I'd like to see a world where all - payments or nearly all payments are value based - and every dollar is tied to outcomes with - those quality metrics aligned - across plans and providers. - And the fourth is around using AI and - supporting a virtual concierge for every member. - A personalized assistant that stays with the member, - understands their history, - and can help resolve payment, benefit, and care - questions end to end across every channel. And, - again, I don't think that's far fetched. We're - seeing this embedded
- in many different functions already, - within health plans, and I think having that - personalized concierge - will really improve the member experience. - Mike, I wanna thank you again for all - of the insight and the practical guidance you've - you've shared with our audience today and and - for the time that you spent with Becker's. - I've learned a lot from our conversation, and - I'm sure our listeners from health plans will - feel the same. So thank you again for - your time today. - Yeah. Appreciate it. Take care. And we'd also - like to thank our podcast sponsor for today,
- Databricks. - Listeners, be sure to tune in to more - podcasts from Becker's by visiting our podcast page - at beckershospitalreview.com.
In this episode, Jon Grandstaff, Senior Vice President of Payment Integrity at Zelis, discusses how payers can reduce payment friction, use AI to support more accurate claims reviews and improve payer-provider...
In this episode, Amy Conlon, MPH, Population Health Sr. Manager at Exact Sciences, and Jeff Huebner, MD, FAAFP, Executive Medical Officer, Division of Medicaid Services, Wisconsin Department of Health Services, discuss...
In this episode, Denis Mrkva, Senior Vice President of Analytics Solutions at Data Axle, discusses how health plans can combine consumer, employer, geographic and healthcare data to identify growth opportunities, reduce...
In this episode, Bob Tavernier, Solutions Executive at Quest Analytics®, discusses how health plans can leverage AI, provider data and strategic partnerships to improve network accuracy, strengthen regulatory compliance...
In this episode, Martina Lee Strickland, Head of Marketing, PR and Brand at Clever Care Health Plan, joins the podcast to discuss the importance of building trust when serving underserved communities. She also shares...