Payer Digital Transformation: The Impact of Data + AI

Summary

• Digital ecosystem: Payer organizations leverage multimodal data from apps, prescriptions, labs, and clinical notes to improve member experiences and drive efficiency.
• Analytics at scale: Advanced analytics and robust data governance enable personalized engagement, streamlined claims, and optimized care management.
• Interoperability solutions: Thought leaders share strategies for overcoming privacy and interoperability challenges while using AI on the Databricks Data and AI platform.

Payer Digital Transformation: The Impact of Data + AI

Talk by Aaron Zavora, Sr. Directory Industry GTM, Databricks ; Jagadish Venkataraman, VP, AI/ML Engineering, UnitedHealth Group ; Neeraj Sharma, Vice President Healthcare, Fractal
Payer organizations are rapidly embracing digital transformation, leveraging data and AI to drive operational efficiency, improve member experiences and enhance decision-making. this video explores how advanced analytics, robust data governance and AI-powered insights are enabling payers to streamline claims processing, personalize member engagement, manage pharmacy operations, and optimize care management. Thought leaders from Evernorth will share real-world examples of data-driven innovation, discuss strategies for overcoming interoperability and privacy challenges, and highlight the future potential of AI in reshaping the payer landscape.
Here’s more to explore:
Economist Impact: Unlocking enterprise AI: https://www.databricks.com/resources/webinar/unlocking-enterprise-ai
The State of Data + AI: https://www.databricks.com/resources/ebook/state-of-data-ai
See all the product announcements from Data + AI Summit: https://www.databricks.com/events/dataaisummit-2025-announcements

FAQs

How do payers use data to personalize member engagement?

Payers analyze multimodal data including app interactions, prescription history, and clinical notes to understand member needs. This video shows how Databricks Data and AI platform enables personalized experiences.

What role does AI play in claims management?

AI streamlines claims processing through predictive models, fraud detection, and automated decisioning. This video features real-world implementations from UnitedHealth and Fractal.

How can payers overcome data interoperability challenges?

Databricks provides governance and federation capabilities to unify disparate healthcare data sources. This video discusses strategies from industry leaders.

What is the impact of longitudinal data analysis?

Longitudinal data enables payers to identify patterns across years, improve care outcomes, and optimize interventions. This video demonstrates applications for millions of members.

Full transcript

All right, thanks. Um, hello. I'm I'm Jagadesh Meat Raman with United Health. Um, I lead ML platform engineering for the enterprise. Uh, you will hear about the platform a little bit in our talk today and I also have a couple of applied science teams working on healthcare AI use cases. Good afternoon everyone. I lead uh healthcare client relationship at Fractal and advise uh different healthcare leaders on AI innovation and scaling.
Glad to be here. Thanks. Yeah. So, we're going to focus on um you know AI upskilling and uh essentially it's a three-way collaboration between um United Fractal and Data Bricks as you will see in the talk. So, we can let's go dive in. All right. Um so AI like you all know is very much here uh especially in the last two three years right uh literally touching all facets of life. So for me again again I truly and strongly believe that the real potential of AI is going to be realized when health outcomes become better due to AI and the healthare system overall starts working better for everyone when we start tapping into the potential and really that aligns well with the tagline for United that you see here helping people live healthier lives and helping make the health system work better for everyone right so that's really what we are striving to do um as a company and so just to sort of um set the stage so United uh as both a payer and provider uh touches over 148 million lives uh within the US and in this day and age that by definition uh also means we have um you know multi-year longitudinal multimodal data um across all of these um patients from various different engagements and you see some of those buckets kind of being called out here.
So you have engagement from the patients. They could be interacting with the myuhc app. They could be searching. They could be calling customer service advocates with benefits questions. So there's voice data. There's search data. Then you have data in the form of um you know prescriptions and labs which is very much you know structured longitudinal data every time they make a visit uh to these clinics. And then there's tons of unstructured notes data in the form of like encounter notes.
It could be discharge summaries. It could be some you know clinical summaries that the clinicians are creating. and also lots of like you know policy documents, standard operating procedures and so on right so there's tons of unstructured data all of that and but then the real power is not just in the uh you know the vastness of the data but truly in the potential to be able to leverage it and make it come to life uh to make things better again I'm calling out some of the uh use cases here if you broadly kind of bucket it into care and consumer if you will so on the care side uh you know you could be looking at how do you prevent hospital readmission sort of early intervention for patients and how do you bring them in before sort of a severe attack comes in or if you have care gap in the form of like you know somebody needs to get a flu shot or a vaccine shot and you need to or there are other programs where there could be a spike in glucose level coming up where there that could be an early intervention from a care provider.
So how do you sort of leverage data to make those early interventions uh for the patient? On the consumer side, again like I mentioned, there are still millions of phone calls being made uh every year by members calling in to understand which provider is in network, what is covered like, you know, um what is their dependent coverage looking like, where should they go for a certain specialty? Again, tons of these uh information that are lodged in long PDF files that people are sort of scrolling up and down to figure out the answers.
Um and then uh again on the search side again on the myc app people are going and uh they want to be able to go and ask um you know very simple questions like hey I have this health issue and I am in this geography like why should I be looking and the answer should be completely contextualized to their plan and their sort of deductible and things like that's uh in their doc in their plan document right and then last but certainly not the least uh just operational automation right like you can imagine healthcare is just riddled with administrative burden from the moment like a patient walks into a clinic to the time like care is delivered and like a claim is settled.
So everything from all the different faxes that are being sent to like notes and summaries that are being generated there's like huge opportunities where to be sort of organically able to reimagine uh the healthare system right and all of this as you see in the footnote here I'm going to talk about that in a little bit more detail in a later slide we want to make sure that everything that is being developed from an AI standpoint is uh sort of done in compliance with our internal group called the responsible uh use of AI group which is a diverse set of clinicians legal procure u uh compliance and security team members who are sort of reviewing these um use cases for like what population is impacted how are you sort of defining accuracy numbers how are you testing for bias fairness and so on we'll get into a little bit of that um in in a later slide um with that I will transition to um niraj to talk about how fractal is sort of uh stepping into this whole job thanks so before I start All this using this data and AI Jagish what you think the biggest challenges that you have had in last few years when you started uh this whole AI journey right so what you think to experiment to pilot to scale and on that spectrum what kind of challenges that you faced yeah so I mean I would say like I mean already uh you know companies like health tech companies specifically suffer from inorganic tech stacks I would say like you know there are a lot of mergers and acquisitions that happen and teams kind of have their own way of um developing uh you know models and um solutions.
So even more so exaggerated in this AI age where literally every other week there is a new model coming out a new framework coming out a new way of doing things and agentic systems coming in. So really for us to just provide that clarity to teams uh as to like how exactly we should go about standardizing the use of some of these tools what to use where what makes sense how do you kind of sort of um be um you know skeptical about like what's coming out at the same time um sort of move fast and be able to sort of uh go from uh experimentation to production right so that is lacking and the whole point of this upskilling exercise is to sort of be able to give them that clarity so they understand like what tools are out Got it.
Thank you. Uh so that's something which u I would like to highlight uh that Bane did a survey with a vast majority of uh executive leaders and the number one hindrance that 44% of those executives said is lack of in-house expertise that is really stopping them or slowing them down that what they want to achieve what they see the potential of AI that can be done on their own businesses but they are not able to do because of the in-house expertise that they lacked.
So that's almost half of that executive saying that. So in that context and that's where we'll talk about what we are able to do uh with UGG. Uh but just looking at upskilling what business values it can provide you is uh multiffolds. So the very first thing is uh you can try to find the talent outside or that also is challenging but uh when you get the new talent they don't know what your organization does and bringing them on boarding and all that takes time to get all the knowledge that your particular business needs.
So if you're able to upskill the current uh team, you're able to retain the knowledge that you have within the organization. They are working on different tool sets or different technology, but now you're able to use their knowledge which is very specific to your organization and then you give them the tools and skills that's required for AI scaling. So that's one. uh you also unlock the product productivity gains by getting now the tools are there even if you are doing engineering work there is a AI component that will make you productive it can write your own code you can further improve on it but it will make you productive but you need to know how to use those right so that's the second part of it it certainly is lot more cost effective we'll see what the initial gains that UAG has that we have for this uh particular program but it's lot more cost effective to either hire or take the talent get those trained in market versus trying to do it yourself right so that is a lot more cost cost effective and last pieces which as a partner um I would like to have lot more partnership with all of you but from your perspective UNG's perspective all of you clients it reduces the dependence on the partners because you control your own destiny.
Partner ecosystem is important. You will definitely need that for scaling. But if you are able to do that upskilling yourself, you reduce the dependence and you can control your own destiny. So these are kind of key business drivers that really um allow why that upskilling can help you in different ways. So uh with that intent in mind uh fractal and u joined forces and uh I'll tell you a little bit about fractal before getting into what we did together.
So Fractal is AI first company. We are in 25 years now. We started before AI was AI and uh we have been doing lot of uh research. We're very research focused. We invest almost 7 to 8% of our revenue going back on research. So there are things those what we see now on LLMs we were able to do before even chat GP moment. We were including those in our own products. We are seeing ahead of the curve. There are uh pathbreaking uh research that we do and present in a lot of different uh conferences worldrenowned and present that research and bring that to uh our clients and we bring that to um clients as our solutions as well as some things that we make available just as a opensource.
So what you see va is one of uh our uh medical um model that we made available uh primarily for Indian market where the access to care is not readily available or there are not too many clinicians available so you can upload all your clinical data or ask questions like it's like medical GPT which is free for all that we have made available or we have these this research going into the commercialization. So some products like Coentic that you see here, it's our agentic platform that allows you to do all these multi- aents that you bring for your business processes, automate your processes for um either automation or just assistance kind of co-pilot which is very purpose-built for your own uh specific business.
So that combined research which is made available either for commercialization for all the clients or even for general public is uh what we do. U the way we look at it um to get the results particularly in this age of AI is just one part of it. The way we look at it is if you look at the first equation on the results, it's a function of AI but what needs to happen on engineering to make that AI work all the plumbing that needs to happen getting the data what all components that you need from different systems.
So that's why you see the engineering is squared because that's how the weightage of engineering is. The third component that we do which generally is overseen or not given as importance at as it should is the design part of it. How you look at the adoption. You build a system and clinician say are you telling me how to do my job? No, you bring them on up front. You design a system that has the adoption because you have worked through them.
So how you build a system which is human- centric design and all those aspects of design if you don't do that mindfully the adoption and the efficacy of these systems are not going to be as much as you would like to see. So that's kind of the first s uh equation that you see here how we um go about getting the results for our clients. The second equation that you see here is how to make it effective within an organization and talent is one part.
The upskilling that we are talking about today is one component of it. The whole governance and Jagdish will cover about how UAG has a robust governance process be data governance, AI governance, the reviews board and all that. So we do that. So out of these equation there are five pieces three on top three here. We cover five pieces as fractal. One thing that we lean heavily and advise clients and organizations to do is the culture because that is very specific to you.
Your individual organization has the culture that is going to be very essential to make it effective and that's what uh we say it is be change management be the mindset and how you encourage people energizing the business and all that is part of the culture you bring that in that's how it makes it effective so to that effect jug how UNG looks at AI and broadly the innovation what kind of culture you have would like to just pick your brain on that.
Sure. Yeah. So I think a couple of cultural aspects that we can um sort of highlight right. So one is really what we are talking about today in terms of upskilling and having people sort of think of AI as a companion if you will like that's kind of where it's going whether you're in software development or whether you're in product management where not. that so just trying to sort of um have that AI first mindset both to be relevant and to be sort of more productive right and then the second one really I think is about reuse and that's something we uh are emphasizing a lot uh within the organization again for an enterprise our size you can imagine there are multiple tech pockets and teams that are sort of aligned with various line lines of businesses building their own thing so how do you essentially create a central searchable reusable repository where people are coming in and sort of being being able to effectively leverage what has already been built and then just build the last mile rather than solve everything from scratch again and again right and for that once again like a standardized framework becomes crucial so that everybody's kind of building on the same tech stack otherwise like things just don't plug in right so that's kind of where we are makes sense thank you okay u so when we started on this uh scale and having the conversation uh at u scale thousands of people uh how you balance the personalization that uh you want to make it as effective as possible while doing at a scale.
We can do it this kind of room even this 100 plus people 200 it's hard to get that effective in a classroom setting and you make it virtual it even becomes less effective. So how you make it uh more effective and doing it at a scale at what UAG needs. So there were different kind of challenges. One uh one size does not fit all. We have different level of uh skills and different personas. Some software engineers to some junior data scientist or some ML engineers. what we are trying to achieve with different kind of uh skills.
So there are different personas we need to really look at uh who we are targeting and how we balance it out. The second piece was uh even on their own journeys each individual's skill set is different or skill level is different. Uh we are not starting at here or here. Uh it's a spectrum. So how we cater to that kind of cohort which is very varied in their skill levels as well and then also the work does not stop our business commitments uh they are what they are. uh so we have we need to do this in this quarter by this uh year end all these business projects those are already committed and business is counting on the technical team to deliver those where is the time so you say okay let's get all these thousands of people and put it in classroom and say okay you are going to be now J&I experts great but what happens to the business projects those we already committed right so meeting those commitments and is still able to do that it's just like just changing the wheel while car is running.
Right? So those are key things that we had considerations on when we are designing a program like this which we are being very practical about making it effective as well as considering what is happening today. We are not jeopardizing that. So fractal analytics academy our uh approach is that uh it's best to learn with experience. So there are lot of different uh courses out there. There are lot of content. There is no uh end to what you are able to find.
Uh but what we really try to do is um our approach is three C's. Jagish talked about two C's. So my competitive split came in and I said okay I need to top it up with three C's. So the three C's we have is we make it very contextualized. So anything that you are learning you want to make sure that uh it is for your specific industry for your specific business. If you're talking about within payers you have Medicare government programs or commercial whatnot there are nuances or if there are functions where uh somebody is helping the clinical team versus enrollment team versus provider of claims.
So there is a context to what kind of programs you are going to help business with and that's where you're going to build those models be traditional AI or ji right so their context is very important so that's one thing that we uh really emphasize on the second C that we look at is how you look at the connections of different systems from data perspective as well as how different points come in when you build a system if you're looking at it from the point of view of I need to build a system or an ML model that does um let's say um identify what my proactive uh outreach should be or who I should be look at like hospital readation that you're talking about Jes right so there were different components that you are going to make connection on rather than a linear You need to find out the connects or dots and say okay this is how I learn to solve a problem.
It is not just a linear thing there I can solve it. And the last piece is this not only the technology is new but how to solve business problems you need to be creative. It is not that something is solved before. There are no problems with these technologies how you make the best use of it. You need to be creative problem solver. you need to learn that this is how I'm going to apply these technologies to solve the problem.
So bringing that creativity. So this is what we focus on that how build a program build content to keep these things in mind which makes it lot more effective for a learner to be project ready. The end goal is that whenever and JDish will talk about some of the things that we are doing on once somebody goes through this program the idea is not to just get a certificate that oh I'm JI certified how that person is able to effectively work on a project immediately after finishing that curriculum that's the whole intent to make them project ready and by doing all this contextualization and problem solving we are able able to make them product ready which is going to talk about.
All right. Um thanks N. So yeah so everything just building on whatever NR just said we essentially have our internal program uh we are calling AI dojo. So in collaboration with our opt university our internal learning management system that we have and fractal that n talked about. So really the goal is to uh create course content for uh you know beginner level people uh all the way up to the advanced people um essentially giving them um sort of basics of data transformations all the way to like how do you go about creating models and how do you start leveraging the genai tools that are out there right so we have over 120 hours of course content that we jointly uh created with fractal um highly contextualized uh I'll get to that in a minute and we have over 20,000 uh you know engineering talent uh across the enterprise globally like working on all kinds of like digital tools to AI modeling and so on and we already have uh 15,000 plus of them already enrolled in the course and sort of waiting um uh to sort of go through it and we have uh at least a couple of different um basic pathways that we begin with uh you have the genai uh pathway and then you have the classic machine learning pathway so you can guess which one people are choosing and then the there's the advanced AI track uh where essentially people who have already sort of dabbled in a bit of AI and want to sort of focus more on like advanced retrieval patterns or like you know graph techniques or fine-tuning a small language model things of that nature.
So essentially you're you have like curated course content for like the advanced um path as well right so that you essentially graduate with um you know specific capstone projects on uh on each of these. So yeah so just to sort of um um reinforce some of that. So it's not about just coming in and uh you know being presented with a um dummy data set of like you know iris flower petals or something like that and walking away thinking that hey I know how to do AI.
So it's really about uh how do you um sort of uh get data sets that are very relevant whether it's like mimic you know discharge summaries or diabetes data sets and things of that nature where you you develop an intuition for the data uh and at the same time you know what techniques to throw at it and how are you able to sort of complete the exercises and walk away right so it's highly contextualized content maybe I don't know n if you want to say something about you know what kind of data sets maybe a couple of examples that you picked sure so again on contextualization what we've built different hands-on exercises.
So as they are going through so even those three pathways that Dougl showed on ji there are four five different courses as they graduate from one to another there are hands-on exercises and we look at let's say look at the clinical data set or this is a claim data set. So we have those contextual data sets for those hands-on exercises. On top of it, once you are done with those pathways that is required uh for you to get to the certification and be project ready, there's a capstone project at the end which depending on the function you are uh coming from.
So if you again you are going to support the clinician versus claims versus member experience those data sets which is very specific to U. G G and very specific to that function is what is being used in the capstone project. So when you are coming you are getting the real life experience that when you're working on a business project for that particular function you will encounter the same set of uh data elements and different kind of tables that you are going to join that I'm looking at the claims versus member versus provider or these files those are coming from provider and third party if it's geni and unstructured data here are the clinical labs so all those things we make it as real as possible and contextualized to that particular function which is not only just UAG specific but within UAG there are RX and there are all sorts of businesses right so from RX to um all the USC the insurance side of the business within insurance you have all these different subsegments so that's a kind of contextualization to the level which allows these learners to a relate because they are already in that function and also use that data sets to solve the problems what we want them to learn.
Makes sense. Thanks. Yeah. And then so once you have the data piece sorted out, it's really about like you know what support you can offer uh to the learners as they go through the process in the form of like you know office hours or like you know uh community groups or teams channels or whatnot, right? Um and then most importantly we make sure that all this course is sort of complemented from like video content that people are looking at on the uh sort of SLM to hands-on exercises that they actually are able to work on in this United AI studio platform that I'm going to uh talk about so that they know once they graduate exactly how to go about sort of creating their new project from scratch.
So on the right again all of this is completely intertwined. Uh so in terms as you can imagine we deal with sensitive data all the time. So really understanding at an enterprise level how to deal with PHI what are the policies in place like how do you think about uh data use rights and governance is kind of ingrained uh in the platform itself. Uh and then that's where the responsible and ethical use um comes in as well.
So when the model is ready to uh eventually go to production uh you kind of go in front of this group. I'll talk about that in a minute. Um and then again all of this is hosted on this enterprise platform which also again I'll dive in in a little bit. So I'll just add on there u just to be clear that uh all of this and where the office hour and community uh comes into play. U what I mentioned that we want it to be a self-arning experience.
So all the content is available. It's on self-pace you learn but if depending on your skills what you're finding challenging and whatnot that's where the office hour personalization comes in. So that's the kind of marriage of scale and personalization that it's a self-arning on your own pace which we believe is the right tal uh content for different skills. But you're not alone. Wherever you are stuck, there is a support system that is personalized and one-on-one you can get that uh support.
That's where that uh scale and personalization uh marriage happens. Yeah. All right. So, so again just want to do a little bit of a deep dive into like what is this platform that I've been alluding to, right? Again, internally we are calling it United AI Studio. Um so really um you know started out a a couple of years ago more as like end to end ML life cycle enablement for teams the coming looking to come in um you know connect to internal proprietary data sets and um go end to end to deployment right but now it's obviously evolving into also like an agentic um framework where you have like MCP clients and servers sitting in and you know data sources talking to one another and so on.
So that's uh the natural evolution for this. So really if you look all the way at the left right I think it's again given the number of different lines of businesses that are there across the enterprise it is very important for uh for teams to have clarity in building uh right up like okay what project am I working on what am I funded for like do I have like the building transparency in like in what resources I am using and storage I'm using so right from the outset they have the ability to bring in their own Azure subscription or GCP subscription and create the project right there.
So we intend this for this to be a thin layer on top of the cloud service providers. We are not trying to reinvent the wheel and sort of create the services that already exist. But really the goal is to be multiloud so that there is not a vendor lock in and especially given the pace at which LLMs are evolving. We want teams to be able to experiment with the Geminis or the GPT4. 1s or whatnot, right? So we want that capability and um so so they're able to come in and choose where they want to bring their subscription go in and then as you can imagine for data can be on like you know data bricks s3 buckets like you know GCP buckets whatnot so to be able to sort of offer them um sort of a slew of data connectors right there with data governance in place so that data station process is there so that they can then um go do the appropriate training and get the data to flow.
So at that point is where they start sort of experimenting and preparing the data going through ETL pipelines you know transforming the data and then start experimentation. So this is again spinning up compute and that's where like you know data bricks comes in um especially at scale but being able to sort of spin up compute and u sort of create a data bricks workspace um run your experiments and at when whichever point they think that this model is ready for prime time.
So that is where they are sort of the platform is natively integrated into this AI review board that I mentioned and that AI review board uh essentially is uh again a team of uh you know clinicians legal uh compliance people that and we are uh united is a member of CHI which is a coalition for health AI. So that is a conglomeration of um you know academic and industry partners who are all trying to collectively figure out what are the right metrics to be looking at.
So if you're looking at like a clinical notes summarization model or like a medical question answering model what I mean there are obviously benchmarks being set by like the big players out there but how much of what they are showing you on the public data sets actually translates to proprietary data sets and which of those metrics actually is relevant for your specific use case. So those model scorecards are coming out of the chai workg groupoup discussions and figuring out you know what is the definition of bias, fairness, hallucination, toxicity, all of those uh metrics for the relevant population that is being impacted.
It's all of that plays a crucial role. So when you think you're the when when the teams think rather that the model is ready for prime time, they go hit the submit for review um button and that triggers like a live review with this panel that uh that I just described. Right? So that's when they go present the results and once this group blesses it and basically says okay uh you can go live that is when the platform allows them to even deploy the model as an endpoint right and then it becomes exposed as an endpoint you can integrate into a clinician workflow into a nurse workflow into like a customer advocate workflow whatever it is right so that is the endto-end uh life cycle and as you can imagine now this is um sort of evolving into orchestrated systems of not just point solutions but multiple solutions strung together in in agentic fashion.
Right? So that's kind of where this is headed next. And again, just to bring that all um together again as a platform customer, there are multiple things um the users could be doing. And so we um you know, naturally sort of enable um you know, multi multi cloud uh playgrounds where teams are again able to come in and sort of look at the latest and uh greatest models out there. It could be notebook LM, it could be like, you know, Gemini Flash or whatever it is.
And they're able to come in and sort of experiment with it. And then this reusable catalog that I mentioned when talking about culture. So really creating a framework where teams that have taken models to production feel comfortable that this is a component that others across the enterprise can actually reuse. So we give them a meth method to sort of a framework to go and publish this to the catalog so that and then attach metadata to it so that like a planning agent can understand the tool or some other team that wants to use the component can understand it and so on.
So just creating that whole um catalog and then obviously the UIS projects that I spoke about right so the other the other thing that is uh that we're working on again once again with uh in collaboration with fractal is this um notion of an AI gateway so everything that uh you know especially from an LLM standpoint or or a third party service standpoint everything that is um leaving our system and sort of going to a cloud provider we want to make sure that there is complete transparency and monitoring and observability in terms of like what activity is going on and also from a cost and capacity optimization standpoint right so you can imagine like Azure for example um offers what are called like provisioned throughput units so this is essentially like dedicated capacity to like the GPT models right but then it's a use it or lose it kind of scenario where if you're not hitting that API all the time you're still getting charged for it this is like a dedicated capacity but then if you go the pay as you go route uh essentially you run into SLA issues and you run into like latency issues it's a best effort right so now teams are trying to trade off between uh you know best effort and dedicated capacity so they are kind of caught in the middle where they don't necessarily have enough traffic to like buy an entire PTU but at the same time they don't want to be stuck with like unknown latencies and so on right so this is where an AI gateway comes in where essentially we as a platform take ownership of capacity optimization and essentially say we will pull together 2 three five how many of our applications make sense and kind of make them integral multiples of the PTUs that we can purchase is so it's completely transparent to the teams u as to how the PTU is being used and they have full building transparency in terms of like how um this is actually sort of being charged uh to their accounts right so again all of this um is sort of um again once again in collaboration with fractal building out the LLM gateway and uh the way it is hosted in the platform so so where does data bricks again truly come in um so apart from teams experimenting with data bricks workspaces once they are within United AI studio.
Um there is also um you know especially for this um AI dojo program in particular um we want users to come in and just focus on the course content. So we don't want them to be coming in and worrying about what service to provision, spinning up compute, you know, clusters and uh you know being um delayed by you know VMs being spun up and so on right so essentially the serverless offering from data bricks uh was a magic bullet for this whole thing where you could potentially have hundreds if not thousands of users coming in parallel and being able to sort of um you know if especially if they're training uh fine-tuning an SLM or like training an XG Boost model and things of that nature then really being able to sort scale up capacity as needed um you know for uh for their needs across thousands of users, right?
So really there is also a role-based access control just ingrained in there so we know exactly who has access to what and at the same time there is very minimal sort of administrative effort in trying to maintain uh you know or rather than submitting tickets or like you know spinning up uh clusters and so on. Uh honestly the scale at which we are doing this it also becomes a great template for a lot of the hackathons that kind of hack happen across um the enterprise in smaller scale but then this is like a very sort of um easy one to replicate for like you know 50 users or 100 users where they have like a very targeted um approach to go do it.
So lastly, I just want to maybe um touch on uh you know, so dojo really is at the intersection of creating, you know, business value and like um uh the outcomes of the program itself. Um so the goal really is for us to have make sure that the internal talent um is retained number one and two they it feels like they I mean they feel like they're getting the right upskilling that they are need that they need to sort of um evolve um with the evolving landscape and uh also like how can we essentially um scale to the AI use cases again you can imagine hundreds of use cases across the enterprise across various lines of businesses and so for team I don't think it's practical ical for us to just go out there and hire an entire new workforce to come in and sort of uh start doing this.
So it's really the natural way to sort of get there is to make sure that the people that understand the system who understand the data have a natural progression towards using some of these AI tools that are out there and uh sort of make the best out of it right so the so really the way to scale to deployment for these use cases is through internal upscaling um and it's also like a great uh talent retention uh mechanism and again like mentioned uh it's um in compared to like an external certification at thousands of employees this is um much much smaller costs uh incurred to the enterprise to set something up internally and like um upskill them and there as well we already have like very high learner engagement like I mentioned we have over like 13 15,000 users and then we are seeing much higher NPS scores I'm told it's much higher than 70% right now based on the last couple of weeks um but but really that uh that is it so I think that just the three-way collaboration between us fractal and data bricks primarily targeting the internal audience to sort of get them to that next level we are starting with the developer persona we are starting with the engineers who are comfortable coding uh and getting there and then once we hit that I think the next target would be like the non-developer persona the clinicians and the product managers and um others who who will also benefit from this but we kind of starting with the people who are like hands-on um coding I think that is pretty much it you want to end yeah I would just say as a next step as said so these thousands of technical engineers and all those personas is the first target which is deep uh upskilling but uh United has 400,000 people uh nontechnical people what we do with them and they need to be AI aware they don't need to have all this uh skill that is required to do hands-on work but how they make uh best use of AI for productivity or what tools are being given to them how to best utilize that so that's one part of it.
We also started and we are seeing that some uh leaders in industry and United being one pioneer which we started with their human capital team conversation on how to reimagine uh the roles of different functions not just the upskilling but completely reimagining how the roles will look like in AI world. So it's very early stages. We started to have brainstorm on how to reimagine the roles. Not a lot of organizations are thinking about it.
But that's where United is going and some of other clients we are starting to have that conversation. uh the way we look at it is how many different parts of any individual role uh those on a daily basis or yearly basis what kind of components of that role is and what we believe is going to be impacted by AI versus not. So the impacted by AI which is completely eliminated or transformed or is still being required like empathy and some of those creativity teamwork.
So which is going to be impacted which is not is where you will redesign the roles very early stages but that's a kind of uh the next stage of this upskilling where the word is headed. Yep. Um I agree. Yeah. So I think really I mean we're already seeing like you know with product management for example instead of having to write like business requirements documents from scratch. People are essentially being able to just do a brain dump into a document.
Multiple people that are engaged in the project are just able to sort of come in and write out what they want to see and then you're essentially llmizing the whole thing and saying create this with timelines and plans and risks and blockers and whatnot. And if something changes tomorrow, all you have to do is like go change a line of text in the raw text that you gave it and everything else on your road map and everything just adjusts to it.
Right? So the whole I think a lot of things are being reimagined. I think right from ticket submission to like peer programming like Satya talked about at build and so on. So I think a lot of things are evolving and uh this is just the first step in the journey to sort of help the developer community catch up to what's going on and then everything else the end users who are most important. I think especially from a trust standpoint, we want the nurses and the clinicians to be able to trust what they are using and understand what they're using.
So definitely that is like the next step we will sort of go towards. So with that we'll wrap up. Thank you. Thanks everyone.

Learn more about the Databricks Data and AI platform.