Achieving AI Success with a Solid Data Foundation

Summary

• Enterprise data foundation strategy: This video showcases GE Vernova's transformation from fragmented legacy systems to a unified Databricks-based architecture integrating pre-spin-off and enterprise data ecosystems.
• Generative AI integration: Learn how incorporating GenAI into overarching data strategy enables high-performance analytics and data-driven decision making across a global energy enterprise.
• Data quality and resilience: Discover how building a resilient data framework adaptable to continuous evolution supports 125+ data sources consolidated into common analytics models.

Achieving AI Success with a Solid Data Foundation

Talk by Kevin Tollison, Partner, AI Consulting, EY ; Santosh Kudva, Chief Data Officer, GE Vernova
Join for an insightful presentation on creating a robust data architecture to drive business outcomes in the age of Generative AI. Santosh Kudva, GE Vernova Chief Data Officer and Kevin Tollison, EY AI Consulting Partner, will share their expertise on transforming data strategies to unleash the full potential of AI. Learn how GE Vernova, a dynamic enterprise born from the 2024 spin-off of GE, revamped its diverse landscape. They will provide a look into how they integrated the pre-spin-off Finance Data Platform into the GE Vernova Enterprise Data & Analytics ecosystem utilizing Databricks to enable high-performance AI-led analytics. Key insights include: Incorporating Generative AI into your overarching strategy Leveraging comprehensive analytics to enhance data quality Building a resilient data framework adaptable to continuous evolution Don't miss this opportunity to hear from industry leaders and gain valuable insights to elevate your data strategy and AI success.
Databricks Named a Leader in the 2025 Gartner® Magic Quadrant™ for Data Science and Machine Learning Platforms: https://www.databricks.com/blog/databricks-named-leader-2025-gartner-magic-quadrant-data-science-and-machine-learning
Build and deploy quality AI agent systems: https://www.databricks.com/product/artificial-intelligence
See all the product announcements from Data + AI Summit: https://www.databricks.com/events/dataaisummit-2025-announcements

FAQs

What challenges prompted GE Vernova to redesign its data architecture?

GE Vernova inherited two large, fragmented data ecosystems from its parent company with over 20,000 dashboards and 500-600 data sources. Consolidating these inefficient, duplicate systems into one unified platform became critical for business efficiency.

How did GE Vernova consolidate its diverse data landscape?

This video shows how they partnered with EY to integrate their pre-spin-off Finance Data Platform with the GE Vernova Enterprise Data & Analytics ecosystem using Databricks, unifying data from ~125 different ERPs into a common model.

Why is Generative AI important to this data transformation?

This video emphasizes that incorporating GenAI into the overarching data strategy unlocks high-performance analytics and enables faster decision-making, helping organizations like GE Vernova compete in the modern energy market.

What business outcomes resulted from this data foundation improvement?

By consolidating fragmented ecosystems and building a resilient, evolving data framework, GE Vernova improved data consistency, reduced operational costs, and enabled analytics that serve all business functions from engineering to finance.

Full transcript

Hello folks. Um, can you guys hear me? Thank you. Thank you for coming to our session. Uh, I'm the executive for data and analytics for G VOA. Uh, after getting the side, we will go into more details about my company. Yeah, thanks. Uh, thank you all for coming here. here. We know you had a lot of options and appreciate you spending your time with us. I'm a partner at EY and I have the privilege over the last two years I'd say uh working with Santosh and GE Vernova uh as a strategic partner for their AI and data journey and we look forward to sharing some of the journey with you today.
Thank you. Um so if you if you see this this is this is a tale of two companies one companies that was and one company that is going to be. So um so if you look at it all of you know G right G existed for 100 plus years um it was made of many different companies healthcare aviation uh transportation oil and gas etc. And as part of that organization, we built a very large data and analytics ecosystem that got data from about 125 different ERPs into a common model to drive financial analytics, financial decisioning at an enterprise level, at a large G enterprise level, right?
And you know over the last two to three years, um G has been split up into three different companies today. Um G Vernova being one of them. Um so we are in the business of energy generation. So we generate electricity u reliable sustainable electricity from gas, nuclear, renewables like wind, solar. We supply the energy uh through the grid and have the software to run those operations better. Right? So this is where as we have a company we had uh historically we built a very large ecosystem for the energy enterprise.
Okay. Uh for the large part of it um just from a metric standpoint um we had about 20,000 dashboards. Um we got data from uh 500 600 different sources. Um and that was a scale. It covered all the functions from all the internal functions from engineering, commercial, sourcing, finance etc. So we built these these two large ecosystems and as we kind of separated from G set up our own company. We've we found that having two large ecosystems to support us is going to be very expensive and very inefficient for our business users to use two different systems, get different answers from those systems.
So we kind of two years back we partnered with EY to move on a journey where we were bringing those two systems together. So whatever we had within the G Vernova business, how do we take what was at a G level separating from GE and merging it into our ecosystem. Okay. So that's where we said, hey, we need to do this primarily because we need to have one ecosystem for our business users, right? And then as this whole AI genai um uh things came into the picture, how do we build an ecosystem that allows us to drive analytics at scale to do self-service to do AI genai kind of outcomes for the organization and then the biggest other driver is obviously the cost right having two large ecosystems very very expensive uh at scale but so cost cost optimization was was the one critical factor as we looked at it.
Okay. Yeah. Yeah. Yeah, maybe just if I if I add you know one or two things there. I think you know from an EY perspective you know we've been fortunate for the ecosystem and we'll talk a little bit more about the finance side of it you know to bring that into EY as part of this journey and be able to work with them in order to evolve and simplify and I think as we look at the right these things about that partnership the first one on the trusted data you know building that future for AI I think the word trust is key and we'll talk about that as we go through this it's not just trust in the data it's trust in the AI systems and it's trust in the people consuming that they're going to make the right decisions when using that data.
And so we'll talk about that and layers of trust as we think through an ecosystem to uh be viable. So when we said we had an existing ecosystem in place, right? So what you see on the leftmost side is kind of a single ecosystem that covered all the functions for our internal outcomes. Okay. And historically if you see we have been very comfortable in the structured data space right. So getting the structured data doing the catalogs doing governance um security access control everything we have been very comfort comfortable with that.
But with Genai coming into the picture how do you create that bridge between structured data and unstructured data to deliver out of these outcomes. Right? So this was kind of our landing zone that we kind of were partnering towards and that's where team came in and did take it over. Yeah. Yeah. Thank you. And so the dark blue items in there are domains that as GE went through their transaction, EY was fortunate to bring that in-house.
And what I mean by that is we acquired a wonderful team of GE professionals about 120 of them that run and operate the finance data ecosystem for GE when it was GE corporate as mentioned as part of their simplification strategy and looking at the ecosystems. We've been partnering to bring that together to be that holistic circle that we see there really working towards integrated insights uh optimal cost some of the things that he spoke about before you know you can call it a lot of things at EY we like to call it AI ready data it doesn't mean the data itself is ready it means all the things on the right side around the operating model are ready for AI uh the data is a critical component to it and so I'll just touch a little bit around you know what we're producing and bringing through here and I'm not going to hit on all of these capabilities, but a couple of them are really key.
And I think the strategy is imperative. Santosh outlined that broad strategy of what they're working towards in order to maintain and evolve the insights as they work towards Vernova. Uh very critical in order to get to a more simplified uh a more service-based orientation for data to be consumed by AI, by people. We were talking about services versus a product orientation last night a little bit and I think that's really something to consider as you all are on your journeys as you go through either consuming and or producing uh platforms and ecosystems for your customers, your clients, your company.
Uh the other thing that I think is imperative here is master data management. not not solely for integrating the data together, but one of the key things for that we'll talk about in a little more depth is around how it's being utilized in order to maintain access to the data. That also ties in very well with the data risk and compliance. So, when we think about this broad ecosystem with uh lots of data, uh lots of systems, how do we know who's accessing it?
How do we know that they're accessing it for the right purpose? and in a highly regulated environment of which GE Vernova is in. How do we make sure that we're adhering to those regulations or our own policies? And so, you know, there's a lot of things on here that we could get into depths on. I just think those three critical things around the strategy, the utility of the master data and connecting that into the risk and compliance is a very imperative thing as you think about, you know, what could go wrong as you start to scale uh an ecosystem as as broad as what we're talking about here.
And uh one of the things I started with mentioning that we have um 21,000 dashboards. Everybody cringes at that number, right? But if you look at it, most of the dashboards have been built by our business users. Okay? Um they can connect to our data lake ecosystem, create a dashboard, publish it in production, share it with 10,000 users without any involvement of it. Okay? So that's where I mean you have 20,000. For example, we built just 7,000 last year.
Okay. But we at the same time cleaning up 3 to 4,000 of them every year because of non-use and nonrelevance going forward. And those operating rhythms are in place. But while we are doing this, governance is critical, right? So historically, if you see one governance meant, people would also cringe at governance, right? Hey, governance means it's a roadblock. We will do our own stuff uh in this space. But we kind of changed the whole paradigm where we said government self-service is the key.
Okay. Because our business users were also changing. I remember like 20 years back I when I joined this uh company I had a user said I just need a button right give me a button and all my dashboards should be in. Now you see users who are like hey just give me the tool give me the data I'll figure out what I need to do. So how do we prepare an ecosystem that enables delivering outcomes at scale because as an IT organization first of all we will never have the scale to meet all the requirements right second we also will not have the business knowledge on to how to extract the best value from the data so as we have been thinking about all of these in terms of organizational readiness training talent etc we had to keep all those dimensions in mind as we as we went along.
Yeah. Yeah. It's a it's a great point and you know governance is something really hard to instantiate as a culture and an organization but you don't get to the scale outlined there on the left without it. Um as we look at this operating model I also think it's really really critical a couple of the points in here in order to get right. The first thing is I have never worked with an organization that has regretted doing something around the talent for data and AI literacy.
How are we thinking through how people will be consuming it from a services perspective? Understanding what that means, where to go, how to get the data. I think the integration that you all have done with your functional counterparts in finance, insourcing, etc. , it's really a two in a box model between your team and those functional counterparts parts in order to move forward in the world. Um, I think that instills a level of trust from a people and relationship perspective that, you know, not all digital organizations have out of the gate.
I think as we look at other elements on here, um, I think that there's a lot of things you you all have been doing as you move out of GE and are GE Vernova around the org readiness, the technology, and the architecture. um great individuals on your team like Oscar who's here. You know, really putting together that foundation from a platform perspective and thinking about it as a platform and not a set of individual products is critical to get scale.
Um the last thing that I'll just say about that operating model is at the top around the business outcomes. Um and we talk about AI ready data and we talk about the data on the left you know getting your data right understanding the capabilities that you need to win with and understanding this operating model are all in service to those business outcomes without those there's no ROI there's no value to the business nobody wants to talk about a really nice data database data lake call it what you want unless it's driving something and I think the imperative thing here is it's driving business critical processes that we'll talk about here in a moment it's helping close the booked uh procure to pay order to cache all the sourcing activities it's not just an insights it's actually driving transactions into the ledger and so there's a lot of wealth around this that's much more than just an analytical layer and I think the business outcomes there it's driven so much process efficiencies over the years that's one of the things that we're really excited about uh at EY I'm going to transition a little bit into um some of the elements and uh promise to talk a little bit about you know this second piece here about enabled by data bricks because it it is an imperative thing that we think about how are we going to do this at scale you talked about cost how are we going to do this in a cost effective manner at the bottom there's a number of elements uh from data bricks don't take this as an exhaustive list of the services and features from data bricks but some of the key things that I'll touch on on the left side we look at the systems uh I got some some metrics from the team this morning so I'll I'll look to make sure I don't misquote those and you were talking about all the ERPs, you know, there's 92 major sources that are coming in.
Uh there's 500 terabytes of data that we have in this uh asset across the GE entity that's being maintained and managed in order to support their financial processes. There's this concept of reports that means a lot, 21,000. There's golden outcomes that we call them around 360 of them that traverse many processes. I I think a couple of the things that are most important in the center before I go to the bottom row though is when we think about our excitement around this, one of the things that's critical for us is this is a socks compliant environment.
The level of rigor that is in there from a controls perspective gives us trust in the data in order to report financial statements. And when we think about data being ready and do we trust the data, if you are trusting the data in order to close your books in order to perform tasks that are getting you through your close process, it adds a lot of credibility when you start thinking about the trust in it to have AI be able to help support insights at scale.
I think that's a critical thing. There's not a lot of organizations that I've worked with that have uh really mastered that in a large scale and I think it's something that'll continue to move forward. I think that control environment, say what you want about socks controls, I think that control environment and discipline is something that's just imperative when you think about uh the level of scrutiny on the data that will continue to be had.
Uh the other thing that I think is important here that everybody wants to talk about is the machine learning engine. You know, great the data is being processed. What else are we doing around the data? There there are uh 30ish machine learning features in here that are supporting critical processes. We think about cash flow forecasting. We think about commodity classification. And we think about, you know, call it anomaly detection for multiple reasons.
One, data quality evaluation. Two, risk management from a functional perspective. How are we managing the risk uh from a controls compliance perspective as we look at the data? There's plenty more that we could talk about, but that integrated nature of these as that data is streaming in from the ERPs. And mind you, these ERPs, some of these are like the largest global implementations of ERPs that have happened over the last decade.
I mean, GE is a large company. So, this is a lot of a lot of data that's coming in and it's being supported and underpinned by these elements from a data bricks perspective. And I I'll just hit on a couple of them very briefly here, but when we think about the Unity catalog, you know, fine grain access is really critical. This type of data that we're talking about isn't public knowledge before the books are closed. So, how do we make sure that we know who has access to what and handle those individuals and resources in a persona group?
How do we think about using Delta Lake for time travel? As you're closing your books, you really need to understand how the data is flowing through these ERPs are going to other systems like your closed console in order to make sure that you understand, you know, day-to-day as you're going through it, reconciling that data. Very time consuming process when you're downloading things from Excel to execute that. We also think about delta share uh forward looking here.
I think some of the things with SAP we're really excited about as we you know take this uh capability broader to market as well as with Vernova about really simplifying that integration when you think about that SAP data bricks uh uh uh kind of uh uh future view of how that'll come together um liquid liquid clustering um CI/CD on the workflows piece and then I think the audit trail with change data capture is also a critical component and you know one of the things that I'll just say is you know recently and actually as as we've uh helped bring these ecosystems together you know data bicks is now able to be that performance layer for the query engine as well whereas with other I call them platforms andor technologies that we work on with other clients you might have a performance layer on top of your database and so getting to a more simplified platform is also uh very advantageous that we've seen So um this is a slide we put together some time back.
I've not updated it. Uh but this is when the whole genai thing come came into the picture to see how we should think about about uh this whole space. Uh what you see on the left is what we as it are very comfortable with structured data metadata management catalogs etc etc right. Um then this whole world of unstructured data comes into the picture. Yes. Genai gives us the ability to navigate that, extract value from that.
But the biggest value is when you bridge those two together, right? Um everything from simple non-Ji use cases like our auditors, they say that hey you have this transaction, can you tell me what is a document associated to it? Historically we show them the transaction in an ERP and then we swivel chair and try to figure out from a document management systems. Right? So u so that is one example. And then then you get down to use cases where you want to make um costing suggestions or procurement suggestions based upon uh the specification of the part to the what were the purchase prices of the part.
So you are looking at data from the structured data standpoint to the specifications from the documents and being able to link all of all of that together right and the way we have thought about I mean we had a big big asset and you may say it's a big stone on the on the neck is that we have 21,000 dashboards right but these are dashboards that we spent a lot of time building right and what you see now is that hey geni coming in but it is very expensive right so how do we build a framework that takes advantage of everything that we have built.
So we we have been able to do some projects where based upon user's question it can go to the analytics catalog and pull up the right report instead of trying to compute all of that on real time when the user is asking that question. Right? So on the left side you see hey I have a report I can just pull up the report very quickly or a dashboard very quickly and show it to the user for what they are looking for. Right?
Hey, if I do not have the report, then yes, go ahead and look at the data and work through that. Right? And then you also have what is what we are thinking about is kind of the source of truth for the unstructured data. Right? Right? I mean we have a bunch of contracts u procurement contracts, sales contracts, drawings creating a layer that gives us the metadata of the documents and that answers a lot of the questions that are basic questions for our for our users and only in the extreme case where none of these questions are answered by any of these structured uh constructs of the data then yes you go in and navigate the whole uh document to look at the document and say I want the specs of this part I'm buying and pull that out from the document.
Right? So we kind of get into a layered approach keeping in mind that things we have already built the assets we already have and what we can do to take advantage of those versus where we have to do very resource intensive compute intensive work to navigate the uh the documents or the unstructured data to deliver that right and the last on the last one that we have been able to do is also drive action. So uh we have been able to go through this process where people would say whom do I want to procure this part from and this the system will suggest which contract to use and which vendor to buy the part from but also submit a purchase requisition in our ERP to initiate the purchase of that of of that part right so basically to drive actions in that space so we have been thinking about that it's it's a evolving space um with um um and EI and the EVI team coming and putting all our financial document financial data into it.
It really kind of takes us to the next level in terms of the availability of the data sets for us to do uh our our analytics outcomes, right? I mean as mentioned, it is socks. Uh we use our application to close about a $40 billion business, right? Um I never like the analytics ecosystem to be used for socks but we are using it for socks. Uh we also have PII and our crown jewels in the data. So all of those things we have to take into account.
So even when we look at the layer the metadata layer the access control becomes a critical part. Hey I am a user I have only access to this data. So you'll get answers to that data. Right. Yeah. Yeah. And it's uh you know I think one of the things you saw that I really like is is you know it's integrated and it's driving to action. And it's not just stopping at an insight that a human reads. It's really driving an integrated uh outcome.
And when you think about all these processes and how large the organization is and many ERPs, you know, we sometimes call it a virtual ERP from the finance side. Really bringing together and stitching it together in a different way. And so as we as we look at this on the left side, you know, we've been calling it a finance data appliance at EY. There's a lot of things that it's doing for us in terms of the services we provide and how we're uh accelerating our managed services that we provide to organizations.
It's also giving us a foundation in order to demonstrate and realize the promise uh uh you know the proof over promise if you will of AI on the right side. And so we have an asset. I'm going to play a video and I just want you to uh you know maybe listen as I talk through h how we're taking this very robust large broad data platform that has high highenriched highquality data and really making it into you know an AI asset as we think about what's moving forward and so this is uh this is an application we work with various uh clients on and kind of thinking through well what comes next as you have everything that has established from a geni my perspective and how do we think about this?
So, you know, how do we get the management reporting with the insights across operational and financial? We need to know the financial impact of production efficiency, etc. How do we think about having a an ecosystem where it's not a bunch of people downloading data or creating a dashboard, but where we're able to go in and ask the agents questions about our financial performance and be confident that we're getting a coherent and reliable response.
And so what you'll see happening here in a moment, this is going through a 6 plus 6 forecast for any financial professionals out there or anyone familiar with the budgeting process. And it's going to be showing a couple things. First, it'll show you that it's curating insights. So to Santosh's point, when you look at your unstructured, your structured, how do you really get to that point where it's going to curate insights for you based on what it knows about your business and the access it has to your data.
So it's going to go through and show us that. Then we're going to ask it a question. I'm just preemptively showing you what'll come up here. Um, it's going to ask a question about the uh variance that it's been experienced from the uh AOP and it's going to go in and there's a agent framework that's been developed in here. Uh, and it's taken us a while from a reliability modeling perspective in order to perfect this uh at least to the point that we're at today to make sure we have reliability in that response.
So, we've uh essentially we'll see some questions come up. It's like a 600 and some odd thousand variants. It's looking at external and internal factors in order to attribute and look at the drivers of business performance. It's going to explain what's happening there. And the way that this happens is that there's an orchestration agent, a reasoning model that's a manager. Think of it like the manager in your FPNA department or the manager in your business department.
It's got three analysts that it's going to ask to perform this task. That analyst is going to create a checklist of what they're going to do. They're going to analyze data. They're going to write code and they're going to generate a response. the manager agent is going to look at those three analysts and say okay did you come up with the same answer that's going to drive uh better visibility to the users. It's going to drive better confidence and reliability in it and it's going to give a robust uh explanation as well as a synthesized executive summary.
And so when we think about this we think about okay well how did the data allow us to do that? There's a couple of things. This data is not pre-agregated for the reports or anything else. It is there in its transactional form. So when we ask questions, there's nothing lost in translation. We're able to give it that guide kind of blend, as I like to call it, blend that implicit knowledge of a language model that it has with explicit knowledge about what our data means.
We need to tell it what we already know. And so we're bringing in that capability here in order to really do this at scale. I think it goes through another one where it's asking questions on documents about warranty periods. to Santosh's point earlier, if all you have is your structured data, you really aren't going to get to that action ever because you're going to have to take a step outside. And so this will go through um it'll do a couple other things and then it'll get into the forecasting.
So how do we start to use machine learning? It's been done for decades in order to forecast our business. That's wonderful. How are we going to use more generative AI techniques in order to get explanability in order to run scenarios? And I think as this comes to a close here and we open up for questions and final comments, I think a couple things that you know are imperative. One is we need to have more agility in our business.
Having all the access to data, having all the access to insights needs to drive agility. It needs to drive action as Santosh uh mentioned and it needs to be done in a trusted way. And so when we think about all of that coming together, it really allows us to have an ecosystem where you know we can get access to our data, we can trust our data. So how do we not only unleash our people on it, dashboards, reports, transactional processing, how do we unleash AI on it in a trusted way and evolve and we think you know at least in the finance domain uh that we are are going deeper and deeper in we think this is a very uh enriched area when you think about an AI system and not a set of point solutions.
Um so appreciate you all spending some time with us. Santos, anything you'd like to say in closing here before we open for questions? couple of things right as you as we start navigating this AI and the genai space right I mean we have been historically very comfortable in the deterministic kind of model where we write ETLs where the answers are very definite right as you think about the space right there are outcomes which need to be deterministic right if I need to ship a part to service a gas turbine at a customer I cannot be 95% accurate okay um so or If I'm I have socks and I have to make a journal entry or report to the street, I cannot be 95% 80 98% accurate.
Right? So where where can you use those um you you need to figure out what is the right balance on where those use cases are and what the right use cases are as you are targeting with your business community. Okay. Um the second thing that we found out is that when we have a large ecosystem of 100,000 tables and you just let loose genai on top of it, it's not going to work. Okay. Um, data modeling, meta, metadata management becomes super super critical.
Okay. Getting the data in a userfriendly language. If you put cryptic titles of the columns etc etc. How do you link data between like hey I have my ledger versus my subleddgers. Is there a link that can be created? All those things become super critical. Right? Hey earlier it was doing the data modeling. We would build a dashboard. No big deal. Put whatever column names you wanted. U didn't matter. But now when you put Gen AI or AI on top won't work okay gets into hallucinations gives completely wrong answers.
So so the data model that we have been focusing on to get the data model right becomes a big factor for us to get the enabling. Okay. To enable this. Okay. So with that uh I think we will open up for questions.

Learn more about the Databricks Data and AI platform.