Real-World Genie Deployments in Healthcare, Pharma, and Supply Chain

Summary

Data leaders from CVS Health, Merck & Co., GlaxoSmithKline, and Premier Inc. share how they deployed Databricks Genie in production to enable stakeholders to query healthcare, pharmaceutical, and supply chain data independently without analyst intermediaries.
The panel covers diverse Genie use cases: explainable AI responses for CVS Health finance teams, external Genie deployments scaled to healthcare systems at Premier Inc., commercial effectiveness planning at Merck, and a global supply chain control tower at GSK.
All four organizations were early Genie adopters and describe the adoption strategies, real challenges, and measurable business impact they achieved, including labor reduction and accelerated time-to-insight.

Real-World Genie Deployments in Healthcare, Pharma, and Supply Chain

Conversational analytics promises to democratize data access across enterprises, but the path from proof-of-concept to production scales differently across industries. This panel brings together data leaders from CVS Health, Merck & Co., GlaxoSmithKline, and Premier Inc. to share how they deployed Databricks Genie to answer critical business questions in healthcare, pharmaceuticals, and supply chain operations.
Discover how these organizations use Genie to enable stakeholders to query data independently, replacing manual analyst work with conversational AI. Learn about implementing explainable AI for finance teams, scaling external deployments to healthcare systems, building agent-based workflows for commercial planning, and creating control towers across global manufacturing sites. The panel also covers adoption strategies, real challenges they faced as early adopters, and the measurable business impact from labor reduction to accelerated time-to-insight.
🤝

Chapters

FAQs

What is Databricks Genie and how is it used in healthcare and life sciences?

Databricks Genie is a conversational analytics capability that lets users query data using natural language rather than writing SQL. In healthcare and life sciences, organizations like CVS Health and Premier Inc. use Genie to enable finance teams and healthcare system partners to ask data questions and receive instant answers without depending on analyst teams, reducing manual reporting cycles.

How does CVS Health use Genie for finance teams?

CVS Health built a research agent powered by Genie that provides explainable AI responses, allowing finance teams to understand not just the answer to a query but the reasoning behind it. This explainability was essential for earning trust in a regulated environment where stakeholders need to audit and verify data conclusions before acting on them.

What challenges did early Genie adopters face in this panel?

The panelists describe challenges including earning stakeholder trust in AI-generated answers, handling edge cases where Genie's responses needed correction, and designing onboarding flows accessible to non-technical users. They also note that early adoption required close iteration on the underlying data models and semantic layer to improve Genie's accuracy for domain-specific questions.

How did GSK use Genie for global supply chain operations?

GSK built a supply chain control tower using Genie to give operations teams visibility across global manufacturing sites, enabling real-time natural language queries on inventory, production, and logistics data. This implementation replaced manual reporting processes and allowed site managers to identify and respond to supply chain disruptions faster than the previous analyst-dependent workflow permitted.

Full transcript

[00:07] All right. Well, hey, thank you very much today for coming. I know this is either during your lunch or you had a quick bite or you're starving and you're going to eat afterwards. So, we'll make sure we keep this quite energetic for the session today. My name is Christina Basmalis. I'm the go-to-market leader for life sciences at Databricks and
[00:23] based in Zurich, Switzerland. American accent, we'll come back to that. But, yeah, we have a I have a great panel. They're they're they're actually super squished together because they love each other very much. I mean, look at they're all this It's a it's a great panel. We have a great panel. We're going to talk about Genie and we're going to talk about really
[00:38] using it in a real-world setting. I just want to know who in the audience is already using Genie at their environment. So, we got about 60 60 70%. Fantastic. Thumbs up. Good? Yeah? Okay, great. So, we're going to we're going to talk about that from a
[00:53] from a you know, what we're doing in health care life sciences. Again, our esteemed panel that we have, well, I'll have them introduce themselves in a second, but we're going to have a couple different questions of how they've how their journey's been in their in their companies, what they've learned, what the value is and what they're thinking about doing next. I
[01:09] would say they're all very much early adopters of Genie and that's that's a great place to be because now they can actually do the journey going forward as we saw it and as you probably saw during this morning the talk, Genie's all over the place. We're very excited around it from a technology perspective and the impact that it's going to bring into all
[01:26] of your organizations. So, just uh I'll go first. I've asked them to do a fun fact as well as tell what their role is. So, Sorry, did I mess that up? Can you still hear me? You guys can still hear me. Okay,
[01:41] they're going to fix it in a second. But, uh so my my fun fact is I I live in Switzerland. I've had the privilege of flying around the world twice in economy class and and one time was during I don't know if you remember about 15 years ago, there was actually a uh
[01:58] a volcano that erupted in Finland and Iceland actually and they had all the volcano ash that was a basically blocked all the travel. So, I was in China. Uh I was there for a 2-day trip. Had a suitcase that had two outfits and we got stuck. And so, that trip when I say around the world, I
[02:13] was stuck in China for over a week plus and um with again two outfits. Didn't go shopping cuz I kept thinking I was leaving. So, you can imagine the hotel bill of doing the the cleaning. And then I finally got to New Jersey. Uh did shopping and then got back to Zurich. So, you know, fun fact of going
[02:28] around the world twice uh which was one that was never planned to happen. Chad, why don't you go next? Yeah. Everyone, uh my name is Chad Novick. I'm a lead director of data science at CVS. And uh my fun facts, I don't know if this is doing too much with the um my fun fact is I actually
[02:45] got here on Friday of last week to help uh pace one of my good friends in a 100-mile race, the Bay Area 100. I don't know if anybody was there this weekend. Uh he got third overall, so I did a decent job. Um Thank you. Um so, yeah. It's a and and at CVS, I
[03:03] lead labor demand modeling. So, uh all the forecasts that determine the drivers that go into our labor model, uh my team is in charge of those and we basically dictate um how much uh or how many hours of uh pharmacist crew
[03:18] uh pharmacy technician go into stores on a weekly basis. Fantastic. Tim, over to you. Hi, I'm Tim Riddle. I work for a company called Premier. We're both a group purchasing organization as well as we have a lot of software that helps keep people home, bring them home to their families. Um my fun fact is I grew up in
[03:37] the sticks of Alaska. The county was the size of New Jersey. We had 50,000 people. 23 hours of daylight, 23 hours of darkness. So, I really made it by being in San Francisco. Love it.
[03:53] BJ. All right, I'm Vijaya Parameswaran. I lead data strategy, products, and innovation at Mark. So, responsible for all things that we purchase from the outside. So, I have two customers here next to me, so I should be careful.
[04:08] And so, to to support all of our business needs, how do we essentially bring that to build data products and to answer questions at scale to make data AI ready. All of that is what I do. Fun fact, uh the So, my kids, when I ask them, "Hey, what
[04:24] do I do?" "You just sit in front of a computer all day, and then in the evening, you just shut it down and come play with us." That's that's all they say. So, that's when you ask them what what do I do, that's essentially how they describe it. That's a fun to to see how they perceive my work. And Francisco?
[04:40] I'm Francisco Cruz. I I work in I'm a senior director of decision science and AI. My my team's actually an implementation team. So, we take all the agents, all the products from Databricks or any other not necessarily competitor, but other
[04:57] technology, we put it together, and then we implement the use cases in AI and ML models for the business to provide business value. I lead right now uh supply chain decision science and AI team along with uh commercial selling
[05:12] team as well. Uh fun fact, I'm very active. I have a third-degree black belt in taekwondo. I teach on the weekends, and I to actually perform some uh halftime games for the Mavericks. So, you you don't want to dance first, do you? I I will if you have music.
[05:28] Okay. Okay. In the back, we got to get some music up and running before the next hour. Okay, fantastic. Thank you very much for the introduction. So, we're going to start with Chad. Um we talked about CVS Health. You said you you focus in forecasting, and you built a research agent that really drove that from a genie perspective. Can you tell us about what's it What was you know,
[05:44] what what's the problem you're trying to solve, and you know, what's what you built actually. Yeah, yeah. So, the goal was really to actually save our data scientists time and enable data scientists to be data scientists at the end of the day. Um we had a lot of data scientists who were basically functioning like analysts half the time, maybe 20 hours a week,
[06:01] where they were spending uh time answering emails from uh stakeholders about the forecast. You know, why is my uh forecast so high? Why am I need to get so much labor for the stores? Um and likewise, we were going into these discussions with finance where we had to explain uh what drove the
[06:17] predictions we were putting out there. They didn't understand shaft plots. They didn't understand all of the kind of visuals that we were putting forward. So, we really need to put together a space to save our team time, and ultimately that's kind of where Genie came in is um how do we create an interface for um
[06:32] all of our stakeholders to discuss with or basically get the answers they need without having to go to our data poor data scientists who were stressing out about answering these emails when they really should have been focused on development. Yep. Fantastic. And I've heard you actually focus a lot on explainable AI um when you were
[06:47] developing that. So, can you talk about how you ensured you had explainable AI especially in the finance department? Yeah, certainly. So, I mean, we modeled it after all the analyses we were running regardless, and I will uh give some credit to Adrian uh my team who we're actually going to give a talk on Thursday, so uh stay tuned for that.
[07:04] But, uh give some credit to her because we basically took all of the analyses that we were doing anyway, these ad hoc analyses, and uh plugged them into the SQL context and instructions pane of uh building out the Genie space. So, for those who are familiar with this, um we spent we basically took a half
[07:20] year to a year of all of our responses and basically added them in added the SQL we were using um and the questions that were being asked of us into that space. Uh so, that that helped tremendously in in basically uh creating an explainable AI. Um and further, we
[07:35] kind of segmented our data into a couple different categories. Um so, first, the key drivers into our model. We had to be very clear about what was what was actually driving the model. To do that and additionally some explanatory features, things like weather for example, when we first built it, it was it was kind of creating these
[07:52] spurious correlations that we didn't want to go down and kind of baking in some additional context. Like weather was really key because we see major disruptions to script volume and sales when there's a ice storm in the Northeast. If anybody's from New York or or Boston, sorry about
[08:07] the weather you had this year. It was pretty terrible. In any case, it was a matter of building the explainable AI was really a matter of kind of translating the analysis we were doing anyway and curating these two kind of groupings of of data.
[08:22] Yeah, fantastic. And then and you got the buy-in from the business on that is they really understood the areas for that. Yeah. They were really happy to get what we provided them with personally. We'll come back We'll come back on that after. So, let's pivot to Tim and talk about Premier. So, you you you've been you were using Genie for quite a bit of time and you decided to scale it
[08:39] internally as well as externally. So, let's talk about the solution that you developed and that pain that you're trying to solve. Sure. So, the first time we used Genie, we were actually trying to build our own text to SQL bot. You know, and then when we got a hold of Genie, we were like, "Wow, we're going to stop that right away." And what we figured out was
[08:56] you know, a lot of our organizations really struggle with with lists of patients, you know, sitting there trying to Nobody who's been in the hospital wants to go back. Right? Readmissions. And And so, really what we did recently, we're actually releasing it this month is embed the Genie API within a workflow
[09:14] application that the organizations can use to create lists of cohorts so they can target better all the organizations we serve how to improve the quality of the care and make sure that somebody that got released from the hospital doesn't bounce back. And so, that's that's one of the things we we've just
[09:30] recently scaled. One of the things that's interesting as a software as a service provider is it's much harder to scale that than it than it internally. No excuses, but it took us longer because we had to build an app, had to call a Genie API, which originally wasn't there, but then it was there, and we had to really make sure all the
[09:46] security and requirements around pushing an app outside of our walls was covered, but we've done that. And we've also started really scaling the AIBI tool as well. We'll save that for tomorrow. Yeah. And Tim, the especially going externally cuz you're you're actually
[10:02] out of the out of the panel here, you're the one that actually has used Genie externally. Is anybody using Genie externally at the moment with their customers? So, you're the only one in the room. What would be a what would be a learning for that? I mean, you said you said it was harder to do, but what did you learn?
[10:18] Um one of the things we recently did was we set up a workspace for a customer. Now, that doesn't necessarily apply to everybody, but it's really worked well for us. Um now, I know they announced a lot of stuff this morning about super fast databases, but still multi-tenancy kind of drives a lot of challenges and
[10:33] complexities. So, we actually just flipped the script and said we're just going to have a single tenant with a single health system for our customer, and that really helped us. It took us a little while to get there. Um we were converting from a legacy vendor that shall remain nameless, but we're up now. And I think that that plus
[10:51] making sure you can deploy I mean, it's one thing to have a site that has Genie. It's another thing to have a site that has 300 sites that has Genie. So, you got to use Dabs, you got to use deployer tools, you got to make sure you're doing it with code and not hand
[11:06] sneakernetting stuff. That's not a good example. is is Scalability. Fantastic. So, VJ, let's pivot to pharma now. So, we had a bit on the healthcare side. We're going to do the the pharma guys now. Um we'll talk first around commercial effectiveness. So, you've Again, you've been using, you know, Databricks and and Genie within your
[11:22] environment. Um and you know, obviously commercial effectiveness helps a company like a pharmaceutical lead to top line revenue. So, let's talk about what you've we've specifically at Merck um, the commercial space in in Genie. Yeah, sounds good. The again, just to
[11:37] paint the picture, right? So, the as a large pharma manufacturer, we sit on petabytes of data, right? So, we have 100 plus data vendors that we purchase data from. And we organize all of this data. This thing
[11:52] is really spotty. Uh, but, uh, so, we organize all of this data and put it together in in various domains and structures. So, as we were looking at saying, "How do we answer questions? What kind of questions are we answering?" So, think of this as Yep. pharma products across oncology, rare
[12:08] disease, vaccines. That's That's the space we operate in. So, as a business leader, what kind of questions are you answering? So, you first question that they're going to ask is, "Hey, how am I doing relative to my forecast?" Now, if you say, "Okay, you're doing well, you're not doing well." Why? What
[12:25] is happening? How is that distributed? Where am I doing well? Where am I not doing well? Hold it up here. Up there? Yeah. Is this better? Okay. Uh, so, the problem was with me, not with the mic. So, how you hold a glass of wine, right?
[12:40] You don't hold it by the bottom of the stem. Yeah, they didn't tell me there is just a glass of wine. So, if they had given me that instruction, I would have Either way, it's the uh So, the questions are very simple, right? So, the questions that they ask every month that you're answering pretty much. It is not the wine.
[12:57] I can tell you. It's definitely the mic. The Either way, it's the uh So, the questions that we answer are very similar, right? So, you're asking from national level performance, you're going down to drivers of performance. You're looking at saying, "I want to slice this by patient, by physician."
[13:13] All kinds of questions. So, as we looked at this, we were spending 2,000 hours a month for one therapeutic area answering these questions. Now, if you do this across therapeutic areas, you're talking significant amounts of time and money.
[13:28] So then we said, "Hey, we can actually enable all of this. The data is already organized into clean domains. We can actually use Genie to enable a conversational environment where the same questions can be asked, answers can be obtained with even better synthesis
[13:43] than what you provide." So that's how we started this journey and so the value proposition was very clearly there. And of course, if you organize the data and brought it to life and and and it's AI ready, this is a no-brainer to do and that's what we did.
[13:59] Fantastic. So VJ, you've you've also moved I said probably from general conversational AI to really agentic with your Genie capabilities. Can you dive into that a bit for us as well? Yeah, so So we what we call our agent verse is is a connected set of agents, right? So as
[14:16] you think about the commercial continuum, you what are you doing? You're talking about developing brand strategy, you're then planning campaigns, you're deploying campaigns, and then you of course you'd have to develop content to deploy to those campaigns and then you're doing measurement. So across this entire
[14:33] ecosystem, how do you orchestrate this seamlessly? So that is through a connected set of agents and that's the agentic workflow that we've enabled. I mean, it's not fully built out and enabled end to end, but most of these agents are there and there's some portions of this that are fairly well connected and enabled.
[14:49] Yeah, fantastic. Francisco, let's see let's see how you're good at holding glasses of wine at this stage. Um so we'll talk now about supply chain. You've also I mean again, GSK is one of our biggest users of of Genie in this particular life science side, actually all of you guys are, but but one of the early adopters
[15:05] working heavily in commercial space as well as supply chain. But we'll talk more on the supply chain side. So you have 37 manufacturing sites, you've integrated them together. How have you implemented Genie specifically in supply chain? Yeah, so let me give you some background with within the manufacturing service
[15:20] cuz I I I recently got on this role as well and had to learn this and so manufacturing supply chain is very interesting in a manufacturing company like like GSK home market. Yeah. It sits right between research and then downstream into commercial. So, if
[15:36] things don't work in supply chain or manufacturing, there's nothing really to sell and then you don't really have anything to convert from uh clinical production to commercial production. And so you have things clinical production to commercial production that
[15:52] you have to make sure you understand. You have quality control, you have manufacturing and then downstream for supply chain and even some forecasting. So, you have multiple personas that exist that are looking at the same data through a different lens. So, it's very
[16:09] important to not waste time like you're saying like getting let's say vendors to come in and build a certain dashboard because you have to look at a uh the slice of a same data throughout all of these sites uh through a different lens through through those personas that I
[16:25] perceived and also maybe some of the manufacturing sites slightly differently. So, when Databricks Genie came out, right? You you onboard everything, your Unity Catalog, you're able to talk to your data, that gave the power to immediately
[16:40] start getting insights for different questions like am I from a manufacturing side, am I wasting anything? Is because there's costs of waste. So, that's impact right now that if you you wake up or a manufacturing lead, you wake up, you look at your data, you can ask that
[16:56] question from that lens. The same data that maybe a quality person seeing that I get it right the first time versus a supply chain, am I actually going to be able to fill the needs of the commercial from every country? So, that's the power of having and
[17:12] conversing with data right away. Uh without that you would have to go the traditional route, be like, "Hey, my BI my BI person, my business analytics person, does your data analyst have time to pull the data, transform, ETL, right? All of that, just to answer a few questions
[17:28] that they need immediate answers." So, if you build your foundation correctly, you put these layers on top, um, you can really transform how you're making decisions. And that's really the difference between getting data and facts and then insights and then turning
[17:45] them into decisions. So, that's how uh, GSK is using it as a whole, but within supply chain manufacturing, these places. So, we're really creating these control tower type situations for the central businesses for them to access
[18:01] their data and ask specific questions. Yeah. So, a great great example of the same data used by different personas to drive different answers and questions or different questions and answers. Um, obviously 37 manufacturing plants is about scalability. So, how did you How did you ensure you're able to scale this
[18:16] across your entire supply chain? Uh, with great difficulty. So, one one thing with manufacturing sites and and uh, one thing with with pharma in general, it's very old school. So, if it's not broken, it's still working, don't really tweak it. But, um, when you see the
[18:34] lift, uh, you really want to make sure that the the proof is there before you make that transition. So, first was proof and then we got all of the manufacturing sites on board with it. Uh, you really need to work with your central tech team or your whatever your infrastructure is
[18:50] for your data team to really get on board and and put that in. Consistency and foundations is critical when you're trying to use the even if you have tools like Databricks. If you don't get on board with that sort of mindset, we you can't scale. So, it's actually more of a
[19:07] ways of working within the company rather than a technological issue. Yeah. Fantastic. So, we're going to pivot a little bit. We're going to stay with Francisco because you got to ask last, now you get to answer first. Um let's talk about adoption, right? So, we all think we all know the the challenge.
[19:23] We can bring these great capabilities, but if your business does not use them, actually they're just great capabilities, right? So, we look at adoption both from how did you get your stakeholders, your business leaders engaged and excited, and then how did you get your business themselves to use the capabilities? So, any any learnings
[19:39] from that? Uh so, there are people who are going to jump in, dive in headfirst, and want to use it, and then there are people that are like, "Uh you're going to take my job or what?" Like that that's that's that's the thing, right? So, um first I think I think
[19:55] and I don't know if this is true everywhere, depending on the mindset of the company, uh you prove it out through one use case, you prove out the value, and you show that, "Hey, you're missing out on this particular." So, with manufacturing, this site is really pulling ahead. You guys are kind of
[20:11] lagging back, and then you get your C-suite to sort of push down. Uh you really need to have that proof of value before you start scaling. And you can show adoption, and then one thing that uh VJ mentioned is measurement. You want to measure what the impact is.
[20:28] Yeah. And align with with your business. So, you really need to prove it out at first before you you jump. There are some companies are they'll just be like, "Hey, this is the mandate from top, and and go for it." And some of our business units that with manufacturing, you really want to
[20:43] make sure the manufacturing leads are are always like, "This is my This is my bubble. I know how it works. I know my supply chain in. I mean I know my my network out." But they they need to be the ones that are pushing it. Yeah. VJ, how about yourself? Same question.
[21:00] Adoption, stakeholder, business leadership, as well as getting the the people to adopt to use the tools. For us, we started with uh one business unit, one therapeutic area, right? So, I think uh the questions were common, very similar across business units and therapeutic
[21:16] areas, and within our own data and analytics teams that were serving all of these business units, right? So, so, the fact that you can actually what would previously take weeks and months, sometimes to answer, now you are able to do that in minutes. That is powerful, right? So,
[21:32] the fact that you're able to do that it's there's no not too many people required convincing to say, "Hey, you should use it." The question is, "How quickly can we scale it? How how much because the ecosystem, the data ecosystem, the questions that we answer is pretty broad. So, now, how can you
[21:48] actually cover more and more? And how do you scale this across the enterprise? How do you scale this across business units within the US market? How do you scale this across all of our 110 markets that we operate within? So, that has been the the bigger push. It's not as much to say, "Hey,
[22:04] convince me that why I should use it." It's more so, "This is great. How can I get more How much How can I get more Yeah. that covers my entire end-to-end ecosystem and for all of the information assets that I purchase and all of the questions that I want answered? And how quickly can I do this across different
[22:21] therapeutic areas and uh markets? Yeah. So, for us, it's it's uh the value proposition was clear again and again. The we had more and more that we were essentially signing up to do. The only way for us to take on more was
[22:36] to essentially say, "There are certain pieces of work that's going to take a lot less effort and and and can run on a self-service model so that we can build and and deploy for other parts of the problem statement that we hadn't yet addressed." Fantastic. So, that was the commitment.
[22:52] Great. So, Tim, again, you're looking at external audience besides internal side of things and adoption there and also meeting them in the workflow. So, same question for you. So, the question. I think one of the things we did at least really quickly internally was we actually have a Genie
[23:07] case study that we used internally to market to our own organization. Yeah. So, it's kind of like guerrilla marketing, right? Well, we we proved that we could do this. We did it with Databricks. And now we're going to Now it pops up all over inside my organization. Externally, we're just in the process of putting this live in
[23:23] production. So, what I would say is the early UAT processes I don't know anybody's got an ICD-10 reference book. This is all health care. How many times have you typed diagnosis code, right? Uh at least in my career, since I started my career as a Six Sigma Black Belt,
[23:39] I've typed dozens of ICD-10. And then I get it wrong, and then the report comes out and it's it's it's junk, and somebody says this is junk, and it's a circle of despair, right? So, one of the first things when we showed it to our users like, "You don't have to do that no more. We're going to query with Genie
[23:55] the IMO ontology, which is our partner, just like Epic and other places, and we're going to give you back a list of diagnosis codes, and you can just approve it to go work on your workflow, right? I know it's a simple example, but anybody's typed 45 ICD-10 diabetes codes
[24:13] into an app, hoping you got it all right, um knows what what that means to a a clinical nurse or a nurse manager or somebody trying to do uh make sure patients come home, right? Yeah. And and so, I think that that one's really where we're going. And I think
[24:28] also with AIBI tool having agent built in, um we're really targeting taking down a legacy reporting tool, right? Because it's a black belt built in to the reporting tool. And so, as we're getting ready to scale that this
[24:43] month, um there's going to be a whole bunch of people that are essentially a black belt without even knowing what a Pareto chart is. In 3 minutes, they'll be able to spit out everything wrong with their length of stay or their blood use, right? And so, I think where we're scaling is just people realize, "Oh, wow The value. I can type a question what's driving my
[25:00] length of stay variance and it'll pop up and it'll be like 14 different things I can go apply to make their organization better and it's really the value 50 60 70% faster than doing old school analytics. And also trusting the data right? And trusting the data because we put
[25:16] Genie through its paces and we we tested it with the Genie space and and all that of course is is critical you can't ask the same question and get a different answer or you lose them. Yeah. Right? But but definitely just the time savings. And every hour a nurse or a quality
[25:31] person gets back is an hour they can spend making their organization better so. Yeah. Chad. So again very similar question on that that perspective I guess I'd like to understand in the roles that you brought this out to which also was an early
[25:47] adopter for you which one's got it and really drove forward from a role persona perspective but you know also business leadership adoption in general. Absolutely. Finance was the first to actually really adopt this that finance financial analysts as well as financial leaders loved it actually out the gate
[26:03] and I think a lot of that was because we had spent so much time human in the loop where we had tested it almost like an analyst or like a data scientist that was just beginning their career not letting it get to the limelight letting them see it kind of just kind of provide a raw output but testing that before we rolled it out to them fully so
[26:21] it had answered a bunch of their emails without them even realizing it. Um Save you some time. Yeah save a lot of time and so that was that was actually so finance was the first that really adopted it I would say. Okay and then as far as general business leadership getting getting your leaders
[26:36] on board went smoothly. Yeah that went that went fairly smoothly I will say we're we're rolling out we've rolled out to more technical users users who are used to logging into I'm going to say the the S word the data bricks conference to snowflake unfortunately
[26:51] Um, but they're used to that they're used to You won't You won't hold it to late. Uh, to to whatever kind of uh, data application. So, they they've been early adopters of this. I think people in general are um, used to having uh, conversational AI in their daily lives and they expect it
[27:08] They know it's a in the business. Yeah. So, in the business setting as well and it was just a matter for us of kind of um, undoing some of the these useless business intelligence tools that we were getting questions on anyway and then providing them with this conversational AI. Yeah, fantastic. Good. All right. So, I
[27:23] actually believe that, you know, you learn a lot when things don't go as expected. So, I'm going to you know, you could use the word failure or you could use the word needs some improvement. Um, so I'm going to make you guys just think about what did what what didn't go the right way during this side of things. So, Tim, we'll start with you.
[27:39] Um, I I think a couple things. First of all, um, because we were early adopter, um, Databricks is awesome. They're a rocket ship, right? But some of the things that might have been niche to our area wouldn't have applied. Like, we can't hand copy files. We have to have an API
[27:54] and then we had to have the workspaces set up. So, I wish I would have thought harder to move faster. Okay. I just I think I we could have done more, right? Um, and we we just didn't anticipate the level of pain to expose this externally. We're we're done now,
[28:11] Yeah. but we didn't anticipate the level of pain and it's transformational for all the people that we're going to give this to. So, I wish we could have moved faster. I wish I would have I would have went harder on the resources. We you know, like everybody everywhere, you got 12 jobs and the 13th job's
[28:28] calling you asking you where your what where where your other 12 jobs are going. Um, and I just wish I would it wouldn't have been a fractional off the side of the desk. Yeah. More strategic from that side to begin with. Great. VJ, from your perspective. Yeah, that Dan said it well, right? I think the fractional piece was of course
[28:45] one when you're trying to do distributed computing. The other one I would say we started this when we started this back out the we didn't quite anticipate the the level of complexity and time and and effort it's going to take. How much context we need to write
[29:00] because what is native to how we do things we just said okay, it should be fairly simple to do this. How can how hard can it can it get to write context? We've done so many of these projects. But it took a good a year plus for us to realize we
[29:16] were not doing it right. By then the LLMs had advanced a fair bit. The the technology stack and how Genie did this had advanced significantly. So that helped. Then it made the work much easier. Right? So I mean we didn't have Genie ontology then. So we had to write our own ontology and and figure all of
[29:33] these out. So I think those are all lessons learned. Right? We started this back in 2004. Took us a good 12 years of of slow walking this to figure out there is better ways to do this and how do we write the what level of context so we need to need to write. How do we bring this data together? How do we have dedicated focus or and then in terms of
[29:51] teams to get this done. And then as the technology evolved it helped matters when we in bringing this to life as well. So that's I think been a pretty rough few months first few months were rough because you think going in you think oh it's it's
[30:08] going to be done in 90 days and 90 days are over you're like okay maybe next 90 days. And then it goes So it took us a while to figure it out that it's not going to be done. I think it's always a challenge of that early adopter because you are using technology that that is relatively new
[30:23] and you know forward thinking and you're going to have some challenges. And and the flip side of that again you're Simon the face of this essentially to to our leadership business leaders across the organization and you're telling them it's coming it's coming. They're like this is what you
[30:39] said last quarter as well. So at some point you kind of have to say, "Okay, you know what? Let's pause this. We're trying to figure it figure it out. We don't want to say give you a half-baked product. We want to make sure that it's good. But, but I think the the good part is we had a lot of support. We we were embracing
[30:56] the culture of progress over perfection, so that worked out well as well. Yeah. Fantastic. Francisco, over to you. Lesson learned. Make sure you hold it like a wine glass. Yeah, there we go. of things. Um There are a few things. Uh first was holding back anticipation from the
[31:12] business. They really wanted this after we proved it out. Um and the reason as we told them, "Look, we proved there was a they were able to run with with Genie code because one, this team had their crap together with data. Their foundation was right. That means
[31:27] everybody else has to get on board. So, that them understanding the amount of foundation work was another one thing. Um but once we got that, um we still had to tell them, "You got to wait. This is still new tech. Uh progressively it'll get better. Uh so, they were paying the
[31:43] system Yep. way faster than I guess not your fault. It's I was new that you guys can handle. So, that some points it would time out and things like that. But, that was because we're we're trying to adopt in those in pushing the system. Um It's These aren't necessarily bad
[32:00] things, but these are just challenges of expectations from not sure who said it, but if you're expecting um you know, to ask a question on chat GPT when it first came out and it gives you an answer regardless if it's right or not. They're like, "Why can't I do this with
[32:15] terabytes of data?" You know, that that's They don't connect the the issues there. Yeah. So, so that that was really sort of the challenges that we had. Um again, it's it's more it's it was more change management, but for them at least they wanted wanted it. And so, we just
[32:30] had to get creative with the engineering until you guys figured it out. But, that's essentially that's how all tech implementations are are sort of done, especially if you're trying to do it fast. Yeah. Yeah. And being again one of the new the new ones. Chad, your perspective? Yeah, certainly. So, I would say the
[32:47] data that we thought was production ready was not production ready. Okay. Which is maybe a little bit different than the challenge that you guys faced, but that was I think the biggest challenge for us up front. So, we spent months trying to get our data curated, getting the semantic layers to the point where we wanted them, pruning away unnecessary fields. We had these codes
[33:05] from initiatives from 10, 15 years ago that our Genie was saying were relevant in driving model that were clearly not. And and that was that was a really key learning for us. And then on top of that also bringing context that it may not have had and recognizing that
[33:21] just because something's not a driver of the model, it may be important in explaining the the history or kind of why there was a variance that occurred. So, it was about getting our data to a stage that Genie could do what it it does as well. to do. Yeah. Yeah. Okay, very good. Um so, we've talked
[33:37] mostly about Genie for business or I guess we're now seeing Genie Genie 1 as the new the new naming that came out today. So, most of you I think all of you are using Genie code as well, right? So, let's let's pivot a little bit into the Genie code side of things and and talk about how you've how you use it. I mean, what what shocked me this morning and I you know, new stat to me was that
[33:54] 60% of all data pipelines are generated Genie code in the last 3 months. So, that's just a wow a wow of how much it's being adopted. So, let's start with VJ. I think you'll go first. Um you know, how did how how have you been using Genie code and any experiences you want to share there?
[34:11] So, I mean, I'm surprised that it was only 60%. So, the you're surprised that it is 60%. I'm surprised that it's only 60% because when we talk about when we talk to our teams on hey, how because I don't think there's not a single user on my in my
[34:26] organization that's writing code Yeah. from from Right? So, they're either of using Genie code or they using GitHub Copilot, whatever it is, right? So, I think they're it's all fairly automated. And uh yes, they you still you still need to
[34:41] know what you're getting, how do you look at how do you look at it to make sure that it actually gets you where you need to go and then your end goals. And uh and then go test it out and deploy, right? And again, across the entire SDLC process, we're we're going
[34:58] through that process to say how do we automate the process and uh now we've automated the process. Now we're saying, "Okay, let's reimagine the process because uh the entire agent tech ecosystem demands that, right? So, because previously we were going through and handing it off
[35:14] and uh we're playing passing the parcel, finish one task, hand it off to the next. Now you can how do we essentially agentify this entire process with some human oversight, human in the loop, on how do you do this at scale, right? So, that's that's a completely different mindset and and uh shift in
[35:30] terms of where we're headed. Yeah, fantastic. Chad, for your perspective on Genie code? Yeah, certainly. So, I'd say we use it as a code reviewer first and foremost. That's been one thing that's totally changed is the way that we do code reviews. Always involves Genie code. Uh we have kind of a template that we use
[35:46] to go through and review. Um and additionally, uh I think one of the other cool use cases, there's really so many, uh but one of the other cool use cases we've had with Genie code is in reviewing forecast output. Um that's something that we kind of develop
[36:01] with Genie code and has saved us um probably 10, 20 hours a week, something of that nature, just uh in in just kind of weekly reviews. So, it's been it's just been accelerating our development as data scientists. Um and I and I love how it also learns the way that you code
[36:17] and then codes in the same way. It just makes it so much easier to read and understand what it's doing. Yeah, fantastic. Really good. Francisco? Uh very similar. I'm not sure how much I can add, but one thing that I do appreciate, um,
[36:33] is everybody converting and accepting Genie code and GitHub Copilot and other AI assistant ways of working, along with reimagining how they're actually developing, right? So, now they're developing agents and skills in order to deploy out. Uh, one thing that Genie code does help, in addition to the
[36:49] pipeline, is if something goes wrong, everybody can look at the same thing that went wrong Yeah. and understand it and learn from it. Uh, it also gives, uh, data engineering capabilities to other personas, like our data scientists don't have to wait
[37:05] Yeah. until one of our teams wakes up in order to actually fix a an issue. Yeah. And so, as long as your, you know, your your your rules and policies around who's allowed to deploy are put in place, um, it's it's sort of changing everybody's role from saying
[37:22] you're only a data scientist, you're data engineer. And then This applies to any uh, AI assistant tool and ways of working. So, it's good that you guys are have put this in place, otherwise, I I think someone else would have done it. Um, but I I think that's really helping
[37:38] transform. All the platforms that are coming out, if they don't think this way, like you guys have put in, they're going to be left behind. Yep. I think Tim, right? Yeah, I'll give two quick examples. The first example is we're in the mother of all migration projects away from legacy databases. Every single database you
[37:54] probably can known demand, no, we had it, right? And And we're almost done as of this month. And what we noticed is, uh, once we got on that, I'm an XDBA, I don't know if you could ever be an XDBA, kind of like it just stays with you. Um, a production DBA, but like a lot of our
[38:10] production teams were writing, you know, in DB sequel, um, and saving 30, 40, 50% of their time just by that one thing. I mean, it's huge. Think about writing a stored procedure by scratch now. It would It wouldn't be done, right? The second example is more of a fun fact. Um
[38:26] I'm moving from a legacy reporting tool who shall remain nameless, and it couldn't do a Pareto chart, right? In AIBI. But I asked Genie code, "What's up with not being able to do a Pareto chart?" And then it said, "We we don't support that." But then it tricked itself and gave me a Pareto chart.
[38:42] And so, think about that if you're moving legacy because now we were early adopter to the clee, the the program that moved Tableau and in Power BI to AIBI. Think about that, but think about what you could add on top of that clee, like headers and footers. I'm talking really blocking and tackling.
[38:59] But that opens up Genie in addition to visuals. People still need visuals. We're not going to completely kill dashboards. Um I just want to make them kinder and not these static, ridiculous things that we have in our organization right now. So, Genie code is really
[39:14] helping with that, and we're on the verge of of hopefully firing a couple of those legacy reporting tools over the next few months. Fantastic. So, streamlining again, simplifying your landscape. Streamlining, simplifying, unifying. Yeah. Right. Exactly. All right, let's talk a little about value. I think all of you guys talked
[39:31] about a little bit of value of your of your solutions that you brought in. But how did you how did you measure that? And how do you how do you ensure that you're meeting the value um and in in that that that area? So, really again, if it's not bringing the outcomes expected, what do you do to pivot if something was was not driving that? So,
[39:48] I'll start with Francisco. We'll go with you first. So, with Pharma, you you partner up with finance. You partner up with finance. Uh essentially with finance uh sorry, with Pharma, you have you have a few. You have your top line and your bottom line.
[40:03] Exactly. And so, your top line, your I'll leave that to you. With your bottom line, you're reducing waste, cost, and everything. And productivity. Well, efficiency. Sometimes you can't be more productive. You don't have nothing else to do. But it's definitely efficiency, and with
[40:18] manufacturing is cost of materials waste. Yeah. So, that's a big thing. Like, if you're if you wake up and you realize something's wrong with one of your machines or any of your processes, and you can attack it, you're preemptively having waste issues, and that's also
[40:33] environmentally sound, right? Like, you need to do that. And so, you partner up, and and then you figure out what you would have spent or you didn't spend. Um and then also, vendor costs, like when you're actually putting that's that's part of your procurement pieces, where you just procurement's like, well,
[40:49] you didn't spend this much, I'm going to take the money back and reinvest somewhere else. So, with with with us at least, we're we're partnering up with our finance team to ensure that that value is there. Yeah. Um also, it's just right, you have productivity, like you're saying, from other parts of the business, because now
[41:07] they're not waiting for you to give them an answer. They can start, and they're really the bottleneck, like, oh, what else should I be doing in order order to um to use these insights that I'm getting. Yes. Jed, over to you. Yeah, certainly. So, we
[41:23] um we measure uh impact on our team, like in our labor demand modeling, in terms of um one of our key outcome metrics, which is weighted mean absolute percent error. And uh we saw a 1% improvement in the 2 months after we deployed our Genie solution. And in terms of business value, that gets
[41:39] translated to um less um overtime in in terms of the pharmacist technician front store crew hours, and also less uh overstaffing of of those roles. And so, if you aggregate that in terms of a 1% accuracy or weighted mean
[41:55] absolute percent error improvement, that's something like 5 to 10 million dollars in in a year. Um and it took us 2 months to get there in with the time that we saved by uh by deploying Genie. It took us a year to get there to the other to the last time we improved by 1% before that. So,
[42:12] um it's not the only thing accelerating us, but it did accelerate us a lot and that kind of goes back to what I was saying earlier. It just enables our data scientists to actually be data scientists. Yeah. So. Fantastic, Tim. Yeah, I think a couple things. First of all, like I talked about with our consultants doing so much faster. We we
[42:27] did deploy to our consulting team and they were doing projects 40% faster. We have hard dollars that are associated with some of these tools that we can take off the thing. Um but I go back to when I used to work with ventilators. Um I told my 11-year-old daughter simplistically that daddy tells stories
[42:43] with health care data keep people safe, right? And what I would say is some of this is intangible because the tools we're giving to our members will describe better what they need to do to fix their care, right? And all of us are in the health care arena. So, I mean, it applies to all all the speakers, but um
[43:00] I'm kind of looking at that more as a missionary feel, like how much better of a story we could tell. So, of course, there's the financial. Yep. Of course, there's we got to take out money. So, I still have a job and so we show that there are results, but also what what could we do better
[43:16] Yeah. um across all the stuff I've heard already. And um I think genius transformational in that. The amount of just thinking about if you've ever worked in a green belt or black belt or learned how to do analysis that you can do right out of the box and
[43:31] completely transform the business in 10 minutes is absolutely amazing. It is hands down the the fastest thing I've ever seen. People train for years to do this and you can do it by pushing a button now. I think the hardest thing is the change
[43:47] management. Yeah. Yeah. Adoption is always key. We should have you do our marketing spiel. PJ, from your your side. It was 2,000 It was 2,000 hours, right? That you talked about. It was 2,000 hours. that I think the the I think the my fellow panelists talked about outcomes, right? So, they they talked about hard
[44:02] dollars, finance. Uh they talked about efficiency and productivity gains. Uh the So, for us, if you look at specific we look at very specific problem statements, right? So, to go from how do you actually go to market, right? So, if you basically when you in pharma,
[44:18] you're talking about identifying patients of interest, physicians that treat those patients, going through your segmentation targeting process. It typically takes anywhere from to go to go through this process takes 3 months, right? So, roughly. Uh for one brand. Now, you have 10 brands, it takes you
[44:33] that much amount of time and effort. So, now if you're able to do this in 1 month, right? So, and then be much more dynamic. Yeah. You're actually reacting to changes in the market. So, that's powerful. You're able to now get to more patients in need. It's not about top line bottom line. So, we're
[44:48] we're able to get to more patients in need and get them the relevant drugs that they should be on to essentially help them and and everybody else. So, that's real impact, right? So, that is great impact for as an organization. That's what we strive to achieve, right? So, in this process essentially if it it helps us grow our share of the market,
[45:05] great, right? So, if somebody else do it, great. But we want to make sure that that is So, that's just one example. And you can you can think about this entire pharma ecosystem and every part of what we do, there are great stories in terms of how
[45:20] this conversational be AI and and conversational AI BI layers and how we're going about this changes how fast we can react to real-time environments and situations and be that much more proactive and and and uh and get get to our stakeholders, get to our
[45:36] customers the right message through the right channels at the right time. Yeah. That's that's all we're trying to do in this process, right? So, we're not necessarily trying to convince them to do something that is uh that they shouldn't be doing. We want to make sure that we get them the right message through the right channel at the right time. Just like you want to be
[45:51] served, right? So, if you're looking for something, that's what you want to be doing. So, that's what we aim to do here as well. Yeah, now fantastic. And I think some of you may not be aware, but we also you talk about, you know, being at the at the commercial side whether it's Viva Vault CRM or Salesforce you can actually launch Genie within the tools right so
[46:08] you can actually bring Genie into the tool launch it ask all the questions you want using all of your data and meet meet that sales rep and that price or MSL rep in their journey in the workflow right so it really can can embed into the the the organizations now so very powerful
[46:24] from that side okay good one last question if we have time I'm going to open up so think about some questions in the audience if we have time left We heard so again all of us have been at the keynote this morning it was you know three hours it was just you know I don't know so much content coming through and you heard all these great things
[46:39] about Genie one Genie anthology I'm going to get all the names wrong I'll stop there there's a few more out there Genie Genie Genie code and and all these great things What are you doing next right so you've heard all these things a lot of stuff maybe you would have done differently had you had anthology you already
[46:54] mentioned that but you've you've heard these great stuff that's coming or is here already what's next right so where are you going Chad where do you want to go Yeah so first off we're scaling that's that's one of the biggest things is we're moving from I think I mentioned we have 10 core stakeholders today and
[47:10] we're scaling to field leaders of maybe 9 to 10,000 people so that's going to be a major a lot of change management in the next couple months that we're really looking forward to and then on top of that we've started building out I actually forgot what exactly it's called I think Genie web
[47:27] apps or Genie apps Yeah We've just started building those out and it's made it really easy to transition work over to our business partners because they can now run workflows in ways that were so much easier than before so those are two areas that I'm really excited about personally Fantastic Tim
[47:42] Yeah we're coming for Tableau and all the legacy BI platforms that don't give good stories and we're we're going to take it out I'm on a mission now because it goes back to the outcomes right if the story isn't told well and
[47:58] and so what are we doing? We're going to scale and we're uh for those who don't know, you can even customize Genie 1. You can put your own logo in there. Yep. And uh I begged the product owners of Genie 1, don't make me reinvent Genie 1, man. And so they're going to give us a version that we can take to our members.
[48:13] Yep. Right? And we're going to scale all of our legacy cool content we had in Tableau and other areas and we're going to give it to them with Genie. And we're going to do it as fast as I can do it. Fantastic. VJ. So for me it's more two things, right?
[48:29] So one is scale, but scale could be across divisions. So from human health, animal health to Merck Medical, Merck Research Labs to manufacturing. The other one is across markets, right? So you have we operate in 110 markets as I said earlier. So how
[48:45] do you go from what we've done in the US very well to all these other markets, standardize and scale, and deliver insights out of the box. Uh the other one that I'm equally excited about with all these advancements and and that we heard this morning is internal self-disruption.
[49:02] This is all about how we serve our external customers within the organization. The other big focus is how do we disrupt ourselves internally in terms of how do we build data products? How do we build our pipelines? How do we all do all this? We I mean we continue to evolve, Yeah.
[49:17] but there's plenty more that that there's opportunities to to go much further along in that journey. So that's So I'm really interested in the technology, changing how you work um as you can see. So are you going to bring on the Genie ontology into the environment? We'll see. We'll see how good it is. And
[49:32] then uh we uh we uh I have a very high bar, so we will we'll learn. Yeah, fantastic. Francisco. Um so BI is gone. So We're trying to move past that. Um I I I think the ontology piece is important to us as
[49:49] well because we're we have so many other vendors. One thing with Databricks that you guys are doing well is you guys are sort of self uh developing in terms of all these. So, all these other vendors that we're using like StarDog, Anzo, and all that stuff,
[50:04] we may not need to adopt, right? Because you guys have everything internally. Uh one thing that I I encourage my team to do though is just be creative with what you're doing. So, I'm I'm aiming for full autonomy autonomous agents for everything,
[50:20] whether they're running on Genie's platforms or my other platforms. Um one thing I appreciate about Databricks is I can integrate all of that. So, running our our team runs out of research, Yeah. uh Karpathy loops, and they we we ping uh Databricks when it's appropriate. We
[50:37] we use it when we can. So, what we're trying to do is go full autonomy in terms of some of the decisions also being made in some of our business units. Human in the loop is great just to check things off, but once you know that it's working, you can open up your head space to tackle other problems as well.
[50:54] Exactly. Very good. Good. One word, just one, and it can't be hyphenated, so just one. Um when you think of Databricks Genie, what what's one word that comes to mind? Tim. Simple. BJ.
[51:09] Promise. Francisco. That's me. Okay, Chad. Intelligent. Now you have no choice. You want to you want to dance? If you dance, I'll get you get you out of the question. Automation. I I I I love automation. So, I I'm just going to say automation.
[51:26] fantastic.

Learn more about the Databricks Data and AI platform.

The information provided herein is for general informational purposes only and may not reflect the most current product capabilities or configurations.