Operational Intelligence for Energy: Databricks at Scale Across Chevron, Shell, Constellation, and Alinta

Summary

Leaders from Chevron, Shell, Constellation Energy, and Alinta Energy demonstrate how they use the Databricks Data and AI platform to operationalize AI across energy workflows—from standardizing KPI definitions across five refineries to enabling traders to analyze 200+ natural gas pricing hubs in minutes.
Shell onboarded traders to Genie in just 8 minutes by prioritizing trust and simplicity over complexity, while Chevron unified refinery operations with real-time margin insights and Constellation deployed NOAH AI for nuclear engineers and AI-powered outage scheduling.
Alinta Energy, running on Databricks for 7 years, executes 5-minute energy settlements at enterprise scale and built an executive AI strategic engine in 4 weeks, demonstrating that operational intelligence requires both governed data and a platform resilient enough to survive organizational disruption.

Operational Intelligence for Energy: Databricks at Scale Across Chevron, Shell, Constellation, and Alinta

The energy transition is the defining industrial challenge of our generation. Grid modernization, renewable integration, emissions monitoring, and energy trading require operational intelligence at scale: turning petabytes of operational data into real-time insights for the people who operate the grid, refineries, trading desks, and nuclear plants. Databricks' unified data platform enables energy companies to eliminate data silos, build trusted single sources of truth, and deploy AI directly into frontline workflows.
This forum brings together leaders from Chevron, Shell, Constellation Energy, and Alinta Energy to share how they are operationalizing AI: standardizing KPI definitions across five refineries for real-time decision-making, enabling traders to analyze 200+ natural gas pricing hubs in minutes, deploying nuclear safety systems on governed Databricks infrastructure, and running 5-minute energy settlements at enterprise scale. Each company demonstrates that winning the energy transition means giving data and AI directly to the people who understand the problems and know how to solve them.

Chapters

FAQs

How does Shell use Databricks Genie for natural gas trading?

Shell enables traders to analyze over 200 natural gas pricing hubs using Genie's natural language interface, with traders onboarded in just 8 minutes. The key lesson from Shell's deployment is that traders need to trust the underlying data before they will use the tool, so the team focused on data quality and simplicity rather than feature complexity.

What AI use cases has Chevron deployed on Databricks for refinery operations?

Chevron built three primary solutions: a work transparency tool for real-time field operations visibility, unified KPI dashboards that standardize metrics across five refineries, and a Nova Margin Insights application for real-time refining optimization. This video notes that data quality is ultimately a people problem, and Chevron's success depended on organizational alignment as much as the technology itself.

How does Constellation Energy use Databricks for nuclear plant operations?

Constellation Energy deployed a NOAH AI system that provides AI-assisted guidance for nuclear engineers on complex operational decisions, alongside AI-powered outage planning and scheduling tools. Following the acquisition of Calpine, the team built a shared semantic layer on Databricks to align definitions and enable consistent analytics across the merged organization.

What is Alinta Energy's 5-minute settlement use case on Databricks?

Alinta Energy, a vertically integrated energy retailer that has run on Databricks for 7 years, uses the platform to execute energy settlements on a 5-minute cadence, requiring both high data precision and low latency across generation, trading, and retail systems. The company also built an executive AI strategic engine in 4 weeks and maintains AI governance through formal AI committees and acceptable use standards.

Full transcript

[00:06] Maybe wait. helps companies take control of their data on the new data intelligence platform. The only platform that brings AI to your data so you can bring AI to the world.
[00:24] Thank you. Good afternoon, everyone. Uh They They still a bit of space here in the middle for the people coming. Um I don't know if you heard it at the back of the video, but uh it says to bring people to bring AI to the world. We're going to be discussing about bringing AI to the world and to the people that actually know where the
[00:40] problems are. So, that's going to be part of uh today's discussion. But, my name is Julien de Bard. I lead the energy and utilities industry at Databricks, and I really want to thank all of you for first making the trip to San Francisco if you're coming from abroad, and as well making it to this forum.
[00:56] You see, together in this room between all our company, we probably represent 5 to 10 trillion dollars worth of energy infrastructure, right? So, the next 90 minutes matter a lot when it comes to delivering the energy to future generations because what we can do all
[01:13] individually will aggregate what the energy system of the future uh will look like. Uh we're going to be talking about the future, but also want to uh talk a little bit about what happened over the past 12 months and recognize some of the company that have either progressed extremely fast in the past year, have
[01:30] done something that is quite remarkable. And the first uh award we are giving away is the energy and utility industry award uh to Ausgrid. You see, Ausgrid has been putting data and AI in the hands of their employees in order to have an AI-powered grid, and the
[01:45] progress they've done over the past 12 months is certainly something we wanted uh to recognize. Uh we have a disruptor award. I kind I kind of uh love the word, right? With Octopus Energy. What's exciting with Octopus Energy is that they've taken
[02:00] 460,000 electric vehicles, 3 gigawatt of capacity, and converted it into a shiftable capacity. What it is is your EV in your garage being plugged to the wall becomes a battery that the grid can rely on when there is peak demand or
[02:17] when there is a a change in frequency, right? These are the kind of solutions looking forward we want to be developing more and more of those. So, congratulation to the Octopus uh team for that uh disruptor award. And then Axpo out of Switzerland developed something they call Axpo.ai. I
[02:34] just said the theme of this forum will be about putting AI in the hands of the people. Well, Axpo AI that they've developed is a way for their teams to have access to all of the data that they need in order to power and deliver electricity in Switzerland. And then finally, uh as an energy and
[02:51] utility partner of the of the year, uh we want to congratulate the Kraken team, right? Kraken has been on Databricks for a few years, but the scale at which they are growing now across 17 countries, 54 million customers, is exactly the kind
[03:06] of impact we want to be having. We want to be impacting the world not one uh place at a time, but across multiple places. So, congratulation to all these teams and we're looking forward to having your company's name on one of those uh awards uh going next year.
[03:21] And then finally, I want to thank the Infosys team that may be in the room. Maybe raise your hand to the Infosys team you're around. We have a few, but uh they are sponsoring this uh forum. We're also working very closely with them on some solution both in the oil and gas upstream, midstream, downstream,
[03:37] and in energy trading. That's something very we are very excited about. Okay. So, to the theme of uh today, I I picked this image. Um you have a field personnel uh standing on a transmission tower here.
[03:52] Uh I was an oil field worker standing on oil rigs in the middle of the ocean. And I hope this picture relates a little bit the sense, the feeling that you get when you are standing on one of these infrastructure that you know every single person in the world depends on
[04:08] every single day of their life. But they have no idea that it exists. They never notice it. They never notice the people that are actually keeping the lights on. And what we asking our field crews to do today is a tremendous task. Right? You
[04:23] see it took 75 years for example to build the grid that we have today in the US, right? What we need to do is to double this in the next 10 years. This generation of field worker has to double what three generations of workers
[04:38] have done over the past 70 years. And what they asking us is to give them the tools to make the impossible possible. I was a field engineer myself 20 years ago I joined the oil field services world. I wanted to change the world. I was very fortunate to live and work
[04:55] across six continents. My goal giving energy to developing countries having cleaner, safer, cheaper energy. And over 20 years I made everything I could to have somewhat of a significant improvement and somewhat scaling my
[05:10] impact. But I grew frustrated for 20 years because I just couldn't do it, right? So 5 years ago I took a leap of faith and I came to the technology world thinking that this is the only technology that applies across the entire value chain of energy and that has the ability to scale
[05:26] to actually make a difference in the way we provide energy today. And I want you to remember this sentence here. The technology that is the most demanding of our energy system today is also the only one powerful enough to save it. And this is why we are in this
[05:43] room together today, right? You represent either a business unit and operation or some data and AI team that are trying to make that happen. I have used this a lot because I believe it captures very well the complexity of our energy system.
[06:00] 225 years of human ingenuity. Every time the world demands more energy, we find a way to either extract or produce more energy. Every time we bring a new source of energy, it comes on top of the existing sources. It never really replaces it.
[06:16] That's why we still have coal today. That's why we will still have oil and gas for a very long time. Every single one of these line on this graph here is a design that we put together, a connection that we built, a handshake between a source and a consumer.
[06:33] Every one of these lines represent where you are working today with your company. Every one of these line represent an opportunity to do things better. It is very complex. And I hear often people using the expression
[06:49] that we need to keep the light on, we need to be uh building the plane as we are flying it, right? Well, our plane looks like a big blue sphere. It is flying in space and there is 8 billion people on board. And we better get it right.
[07:08] Now, the two most powerful industry in the world now depend on each other. You see, by 2027, 40% of the data centers that exist or being put in line are going to be constrained by energy. We need to develop more than 50 million
[07:24] miles or 80 million kilometer of transmission line by 2040. We have 2,500 gigawatt of existing available energy that just can't be used and is being curtailed because the grid or the system can't take it.
[07:41] So, our challenge is not to provide more energy, but it's to operationalize and integrate all these different sources in an energy system that makes sense. Right? The challenge is huge. We all know that. But I believe the opportunities is of
[07:57] the same size, obviously, right? And again, there is no other people than the people in this room that I can I can actually solve uh for this. So, what I'm asking is to get past all the buzzwords, right? And really consider that it's not an
[08:13] energy era, the biggest challenge of our generation. It is not an AI era where everything is going to be era or is going to be AI. Really, what it is it's an operational intelligence era. How do we get it right? How do we combine all those things together? How
[08:28] do we integrate all these sources? How do we move electrons and molecules faster? How do we consume them in a better way? How do we feed all these data centers that we all depend on, right? The winners of the decades, in my opinion, won't be the one that have the most
[08:45] power, that are able to create the most capacity. They are not the one that are going to have the flashiest models and software. Really, they are the one that are able to inject data and give it directly into the hands of the people that know where the problems are and know what needs to happen to fix
[09:01] it. We know how to produce electrons. We know how to move a molecule. We know how to build softwares. That's not the question, right? But as an industry, we are standing at a crossroad where will we show up as independent providers
[09:18] of all these things? Or can we actually work together to be an ecosystem that delivers energy for the rest of the world? True operational intelligence means that we stop treating our silo or our own data
[09:34] as a local product. And we actually share it to create intelligence across the entire ecosystem. And you see our the value chain of energy and utilities is very very large, very complicated. And we always say, "Well, the problem is
[09:50] that our industry is very siloed." It was not siloed by mistake. It was engineered this way. Cuz if you look at everything that needs to happen to power the world, a lot of the matrices, a lot of the activities you spend an entire career to master them, right? Nobody can have the
[10:06] knowledge of everything that is uh part of the entire value chain of energy and utilities. Where we cannot get it wrong is that these organizational silos, these group cannot translate into data silos.
[10:23] We need to be able to take our operational knowledge, our operational data, feed it into a unified data platform, and forget about having our seismic data in one certain system and owned by a certain group of people, having our OMS
[10:38] events stored in another place, having weather events stored in a separate place. What we need is all feed our data, our operational data, into an intelligent platform that can return intelligence for everyone.
[10:58] But that is a stark picture of where we are today. The reality The reality is still three in four AI initiatives are stuck at the pilot phase. And really, I see two main reasons for that. The first one is we've asked data and AI
[11:13] teams to develop the actual operation solution that is supposed to save the world. They are not facing the problems every day. They are not facing the operations every day. They don't see the opportunity the way their field crew see
[11:29] it. So, they're doing their best. They are developing solution as they think they should be developed. But very often what they developed is very much substandard to what operations people actually need. The second thing we've done is we say, "Okay, that's the problem. We're going to ask our operations
[11:46] people, our field crews to develop the solutions." But when you are in operations, all that matters is the equipment you're in charge of. So what you're going to develop is a very local solution. And if we don't give them the tools so that local solution that they develop
[12:01] and create value where they at can actually be scaled at enterprise level afterwards, then the incremental impact that we having on the industry is very minimum. We need to be able to take their knowledge, their local experience, their local wins, and very quickly scale them
[12:19] at enterprise level when they've demonstrated value. So how do we operationalize AI? I'm not going to go ahead on the fact that you need to have trusted governed data for people to operate on the single source of truth. But I will say this,
[12:36] the customer that we see being the most successful today at having real impact on their business, on their top and bottom lines, are the customer that have created a unified data platform and that are very authoritative in saying, "If you want to build something new, you need to use
[12:53] your data from that single source of truth." It deals with a couple of things. First, obviously, people are developing something on data that you can trust. But the other thing is that if the data is not there, they're going to start screaming. And a lot of my a lot of our customers when we meet with you, they you tell us,
[13:10] "But what data set should I prioritize? Which one should I bring on first, right?" Well, the best way is to have your team telling you, "If you don't put it there, I can't work." Right? So we've seen very successful customer just using this concept to create the right backlog. But what we do next is where it's very interesting.
[13:27] You need to empower the front line. You need to empower the people that are facing the problems every day because they know what they need, right? And like I just said, they're going to develop that local solution and all you want to do is make it that they develop it in a
[13:43] place where it can be taken at scale. I will take a very concrete example. If someone needs to do maintenance on a certain equipment somewhere, you know that maintenance is due, you need to find what are the spare parts and tool that I'm going to be using. At a local level, it is very easy to go
[13:59] and look at your maintenance manual, pull those part numbers of these equipment that you need, and go and schedule the work, right? It's going to work for them. But when you start to when you try to take these at enterprise scale, it doesn't work anymore. What I'm saying is
[14:14] that these people that are developing these solution locally, force them to use to use AI to pass through these maintenance manual, to pass through these operations manual, to use AI to be interrogating your inventory to know if these spare parts are here, right? Cuz that way when they demonstrate value
[14:30] locally, you can immediately take it to scale, right? And that's what we're all trying to do, right? We're trying to multiply the pace of value where what someone does in a corner can be expanded to enterprise scale, right? So here, that's how I would like to
[14:46] offer as a definition of operational AI, operational intelligence. Enabling and empowering the people who know the problems to actually build the solutions. It is not easy,
[15:02] but if you look at what we've been doing with Databricks over the past 2 years, we talked about democratizing data and AI. It's about making people be able to use data and AI in natural language, query your data in natural language, build your apps in natural language, build
[15:17] your agents in natural language. You don't necessarily have to know all the tech stacks that's under it. What you want is it to give you the right answers and to develop the thing that you want. That's what we call a democratization of data and AI. And that brings me to another point
[15:34] still related to people. And that is the workforce crisis that we are facing. And I'll take only on three points here, but one in four energy workers is going to retire in the next decade.
[15:49] You see, the 23-year-old Julian on an offshore rig with his full 1 year of seniority facing a problem that is not in the manual that he read 3 weeks before going to that rig was doing this, calling down.
[16:05] Talking to the 55-year-old guys that's been there and say, "Joe, have you had that happening before? What should I be doing?" Right? Well, the new generation, the 55-year-old Joe, won't be there. And we need to use AI to capture that knowledge, to codify that knowledge
[16:22] before this expertise retire. We don't have a lot of time, so I hope you understand a bit of the urgency here that I'm talking about. One in four people working in our industry will retire in the next decade. That is tremendous.
[16:38] The second piece, and I'll talk about GDP and energy interchangeably here. You see, um I'm not blind to the fact that a lot of people are worried that AI will replace jobs. Right? Truth be told, we need AI to augment
[16:54] every single active worker in the world right now. Cuz the way the population is going, we have more and more people retiring and living very long. Right? In 2030 or or in 1950 we had eight and a half workers for
[17:10] every person retired. Right? In 2015 or 16, I don't remember what's on the slide here, It was 3 and 1/2 to 1. Right? By 2040 it's going to be 2 to 1. Which means that both from a GDP standpoint and from an energy production
[17:27] standpoint, either every single active worker produces a lot more or we're not going to be able to sustain all the people that are enjoying, rightly so, a longer retirement. This is just the way the world is going, right?
[17:42] So, again, we should not be afraid of AI replacing jobs. We need AI to augment every single active worker today in order to be able to keep the world going the way we're going. The last one is uh our energy and and and utilities
[17:58] industry not very sexy. People don't really want to come and work in the energy and utilities space today if you ask people outside of the of the university. But still, 60% of Gen Z and younger people say that they want their career
[18:13] to mean something. Right? They want to have a purpose-driven career. There's no better place than energy and utilities to have these kind of career. And if we combine that with the fact that everybody wants to work in tech, once we put tech inside energy and
[18:29] utilities, we should be able to attract the entire all the talent in the world, right? This is one of the biggest challenge when I meet with CDOs, CIOs, and people that are really trying to make the change. They tell me, "Julien, I can't hire people that are good enough
[18:47] to make things happen, right? Because they've all gone to a very deep tech type of things. We need to be much better at explaining how our industry can help you as a have a very purpose-driven career, right? And this is our role, all of us,
[19:02] to go and talk to everyone about you can save the world for the future generations. So, I've presented all this and you're probably great. Nice slides. Um is it happening? Well, for us it's happening with 1,400 customers and I'll only put a few names
[19:19] here on this slide, at least for the one that are public. But this is already happening. Our customers are building this future where data and AI is obviously demanding from our energy system, but also the solution to solve for our
[19:35] problem. And today I'm super happy that we're going to be hearing from our customers. You're going to be hearing directly from the people that are leading the way in using data and AI to create real value inside their company.
[19:50] So, we'll hear from Chevron, we'll hear from Shell, we'll hear from Constellation and we'll hear from Alinta Energy. But before I leave the stage to them, I just want to leave you with two key things that I want you to remember.
[20:07] The two most powerful industries depend on each other. The only technology that is demanding of our energy system, but is also the only one powerful enough to save it, is what we do in data and AI with your IP on top of it. This is what
[20:24] you're building on top of the tools that we are giving you. This is a technology I'm talking about here. The technology that you are developing with your IP on top of our own tools. And operationalizing AI to have a real
[20:40] impact on the world, it's about giving people who understand the problems the ability to build the solutions. Thank you very much for your time and I look forward to further discussion. And in the meantime and
[20:56] without further ado, I'll invite Anne-Marie to come and join me on stage. Anne-Marie Chevron. Thank you. Hi folks, so bear with me. I'm going to take a selfie so I have evidence to show
[21:12] my children that people actually do listen to me. So thank you for that cuz they roll their eyes a lot and I won't be able to see you rolling your eyes so that's fine. Um so let me get my slide. Here we go. So building on what Julian shared, a lot of
[21:28] what we're working here is about building on actionable and trusted data. So my my talk is going to focus a lot on how are we building trust in data? Um so a little bit of background on Chevron. You're all probably familiar but over 145 years old. I think we're coming up
[21:44] on 150 um this year. So we started in 1876 I believe. So um 145 year old company. We've got over 45,000-ish employees and operational assets in 180 countries.
[22:00] Um we've got a simple purpose. Our mission in life is to deliver affordable, reliable, ever cleaner energy to power and or enable human progress. Um so that's kind of what we all rally around. The segment of the business I support and what we'll talk about is on
[22:17] the downstream, the refining side of the business. Um fun piece of Chevron history. So Chevron is a California based company. We started as Pacific Coast Oil Company back in 1876. Transitioned into Standard Oil California. Um so Socal is what it was
[22:32] called in about 1900. Um and then transitioning eventually when that Southern the Standard Oil Companies uh broke up, we became Chevron. Um so we are a California company at our core and uh the one of the refining assets, our oldest refining asset is right here in
[22:48] the Bay Area in Richmond just across in the East Bay. Um and then the other thing Standard Oil, we we still have one Standard Oil branded gas station and it's right here in San Francisco. So kind of fun. We're definitely rooted here in California. All right. So at our core again, we are
[23:04] an energy company, but at our core we're a technology company and I know Julian, you kind of hinted at that as well. Um and across our refineries, we operate a vast array of uh technology sensors and devices and systems that feed us just an amazing amount of data. Um these are
[23:21] being fed to both local teams that are refining, but also in our central orgs and now also offshore um teams as well to best operate and keep our systems running safely and efficiently. Um so, this is a big opportunity. Tons and tons of data.
[23:36] Why does this matter? Cuz as you can well imagine, every decision in refining matters. You know, this is under immense amount of pressure. Our operators are making dozens and and often in in a high consequence event, you know, they're making so many decisions under this high pressure and often they have so much
[23:54] data, they don't know what to do with it. And so, we're trying to figure out how to make that easy for them. Uh so, having that right information at the right time is really it's not a nice to have, it's essential. So, what's the problem? Here's the challenge. We don't suffer from a lack of data. That is not our problem. What
[24:10] we suffer is from a lack of trust in data. And so, when people don't trust the data, they don't act. They often will fall back to old patterns and what they know. They don't always have the right information and so, that is can get us into real problems with some high consequences.
[24:28] So, why does the trust break down? What happens? So, we found is it you know, the trust in data erodes often because our metrics don't match. So, we're using the same words in situations, but often we're actually meaning different things. And so, through this journey we've been on, we'll have dashboards, but we're not
[24:44] always comparing apples to apples across the system or even across units in a single refinery. And so, we're that that erodes trust in the data. And people say, "Well, that data is not right because it doesn't take into this account or it doesn't take this fact into account. And those that drives the
[25:00] a lot of effort for our analysts and others to kind of build that trust. So, that's the journey we've been on here within within refining right now in Chevron. Um So, where do we start? So, the team my team's called digital engineering for manufacturing. Um we start with a
[25:15] people-first mindset, and Jillian you talked about that, too, putting the data into the front line. The goal isn't more dashboards, it isn't about driving tech, it's really about driving useful data into the hands those front-line workers, and building it in a way that they actually work. So, it makes sense to
[25:30] them, it's being served to them in the way it's the right way, and and and they believe it. So, technology is an enabler, it's not the solution itself. So, we're not doing tech for the sake of tech. Um design data is designed around those front-line workflows. And we stay focused on usability,
[25:47] clarity, and actionability. So, today we're pulling millions or billions of data points across our all five refineries. One of the missions are each of these refineries they're disparate across the US. Um app to
[26:04] operated independently. We're really trying to operate more as a system. How do we actually do better together? And so, bringing all this data together can be a bit overwhelming, as you can imagine. This includes data like equipment history, some work orders, operational data from our historians,
[26:20] performance logs, sensor readings, all of this data, all the support that real-time decision-making. The objective is to bring our five refineries into one connected digital ecosystem. This is a shared view across a common view across all of our systems. So, our
[26:35] our operational managers at every site has the same dashboard, the same information. Sharing common language one of our biggest challenges our taxonomy, right? We how we label things in PI is different from site to site. It's the same pump perhaps, but we call it something different, and that creates
[26:52] challenges on consuming data. Uh standardizing those asset hierarchies, um naming conventions, which I just talked about, and delivering the ability to deliver on competitive performance objectives consistently across our system.
[27:08] All right, so we're going to talk through a couple of examples, and I have long speaker notes, and I know the good stuff is at the end, and I don't want to miss it, so bear with me. The first initiative is is near and dear to my heart. This is my team solution. This is our kind of first dip into this world. Um we've labeled it work transparency. We call it DVD.
[27:24] Um so imagine you walk into work. I'm sure you're all going to resonate with this. You walk into work, and you've probably already checked your phone, and you know you've got 15 emails requiring some sort of response. You've probably got Teams or Slack messages waiting for you. You've got uh your calendar's packed. If you're lucky, you've got a
[27:41] few minutes slots to maybe get some work done. You've got a one-on-one with your manager. You get into that meeting, and you're you're prepared from what you kind of recall. Um but they ask a question that's something from 2 weeks ago, and you don't have the most recent updated information because you're working with
[27:57] a million different things, right? So you've got this vast amount of team. You're juggling so many priorities, it's very difficult to be clear on anything. You start listing off the things you remembered, but you don't have the full answer. I mean, we've all been there, right? So this what we realized, this isn't a personal productivity issue.
[28:13] This is really a system problem. It's just too much information. As we researched this, we learned that our teams So this is my team directly. This is my control systems, uh process control was under me at the time, and our IT teams, they get they get requests work requests from at least 11 different
[28:30] systems. Um and a lot of shoulder tapping. We know all that happens. So it was it was really difficult to manage those expectations, prioritize everyone's short Everyone's feeling overwhelmed and short-staffed. How do you best prioritize? So that was our big gap we were trying to resolve. Go back a little bit here.
[28:47] Here we go. Um So, we realized that something very simple, right? It wasn't a So, one of the first things we thought about is, "Oh, let's get everyone on the same system. Instead of having 11 systems to drive work, let's get everyone on the same system." Well, that's not realistic. You're not going to get your ops team to start using maybe ServiceNow
[29:03] or ADO or some other system that an IT team wants, and you're not going to get your your facility engineering teams to use something else. So, that was not going to happen. So, we decided, "Hey, let's bring all that data into a single pane of glass." And that's really what this solution was. So, using Databricks, we pulled data from all these
[29:18] disconnected tools um into a single file unif- unified view. We layered reporting on top of it. And this showed the full scope of work on any given day. Really helped en- enable our supervisors to work with their frontline workers to really talk about what's the most important thing today and what isn't going to get done
[29:35] today. So, it helped us better manage stakeholder engagement and manage expectations. So, now imagine a different morning. You log in, you get into work, you probably check your email. There's still stuff happening in there. But you get in and you can see on our DVD or our work transparency tool, you can see exactly
[29:52] what's in there, what's aged, what's new, what new requests have come your way that maybe is marked urgent, and it really enables folks to have a better, more uh uh tactical or or strategic view of their day. And instead of debating on me trying to call four different people to
[30:07] say, "Hey, what's the status of this?" I can log in and see where things are. Enables me to have better conversations with my leadership and better manage those expectations. So, again, we're really working on building um moving stress into focus, making activity into prioritization,
[30:24] and uh busy work into actual real progress. All right. So, our next solution is our unified KPI dashboard. So, this is another one that this is used across our manufacturing system. So, imagine you're part of a leadership team and you're actually
[30:39] managing your multi-billion-dollar refining system using a spreadsheet. Which we've all done, I know. I love Excel and you will take it from my cold dead hands before you remove it from me. I love it. But, we realize that is not effective, right? We have all these KPIs and the majority of the meeting is
[30:55] debating, is that number right? How come this one changed? What happened over here? And we don't have the full picture. This is the waste of everyone's time, right? So, this is not an effective way that spending more time debating the actual data analysts spending probably weeks building this report monthly. So, it's very dated and
[31:11] and not real time. Critical decisions are delayed while analysts are manually getting those updated data sets and um trying to pull data from the disparate ecosystems. Everyone's working hard, but no one has a complete picture and certainly not timely. The challenge
[31:28] wasn't a lack of data. The challenge was really about the thousands points of data being turned into actual actionable information. So, no one of our common challenges we didn't have a common KPI definition. We used the same words, but we actually meant different things. So, that was one of our first obstacles to
[31:44] overcome. Um we calculated them, like I said, in different ways in different sites. Limited visibility into leading indicators. We always had a lot of after-the-fact data. Well, that's hard to fix later. Um significant effort producing these reports. Low confidence in the number. Once you
[31:59] get it wrong, you've lost that trust. Um and as a result, the leadership time is consumed debating the data rather than actually making decisions and acting on it. So, our biggest breakthrough in this journey was really moving from this is a reporting problem to actually this is a workflow standardization problem. So, it
[32:16] was a lot of effort working with key stakeholders around these critical um performance indicators to align on what's the definition of that KPI, how are we going to all calculate it the same way and agree that that's the number and it's okay, building the right data pipelines to automate that so it's
[32:32] not a manual exercise anymore, um and establishing those common owners and governance around those KPIs. So now in the leadership team, the new new focus now is everyone's looking at they all agree that data is the data and everyone owns it and they move jump into
[32:49] what are we going to do to fix this? What are we going to do? We see this trailing off. We don't like where we're landing on a utilization number. What are we doing about that? So a lot less time debating, a lot more time actioning. All right, let me go.
[33:07] I have long meeting notes on this one so I want to make sure I get it right. All right, so final example, this is what we call it's a margin inside tool. We call it Nova. Um, so this was one that was developed by our advanced process control team. So our APC engineers, these are process chemical
[33:22] engineers, very very bright folks, making a lot of money for the refinery. But they're often trapped in data data collection, right? So they're pulling data. So advanced process control is one that optimizes the refining system. Think of a refinery as a giant chemistry set and the whole magic is you're
[33:38] heating things up, you're breaking down your molecules, you're adding hydrogen to make my higher value products and it's it is a lot of tweaking. So if you can make this pressure higher, you're going to get better outcomes or So that's what our our advanced process control team does. Um, but they would spend
[33:53] days trying to pull their view together to understand why maybe this unit is not delivering on for some reason, but they can't really figure out why. It's is it the controller's off because the operator doesn't like something about it or some valve is faulty and that's why it's not optimized or there's some
[34:08] sensor broken or there's something else at play, right? And so they would spend days pulling together from OSI Pi, SharePoint, some spreadsheet, maybe some SQL database. They're pulling all this data to build their story, spend hours and hours and hours on it. They get excited because they figure out the
[34:24] problem and they bring it to ops to influence some change. We don't believe that number. That's where ops is. We don't really buy it. And so this is super frustrating and obviously value is slipping through our fingers as we're debating the data. So this is a really common problem and we're losing margin because we weren't optimized like we
[34:40] should be. So it wasn't because the controller weren't working, it's because they were sub-optimizing and we couldn't influence the right people to make the changes needed to drive the right outcomes. So let me go here and make sure I got my notes because we have stats at the end of this one. So reporting was super manual and so it
[34:56] took a long time. Uh data was lived across just in a disconnected systems. Often how we calculated the value of that margin loss varied from site to site. Um our engineers were spending hours reconciling this and trying to optimize or influence right outcomes.
[35:13] So we asked a simple question. The problem isn't the controllers, it's about visibility and agreement on what you know what this value means. And so we put together this project this Nova or this margin insight tool to really pull this data live, it's real time and it shows you what your unit that's
[35:30] underutilized or under optimized and what value that equates. So you can flip it to say what unit do we really need to focus on because that's the one that's losing us the most money from a margin capture target. So it really changed the focus and so you're not working it it held it made everyone accountable of
[35:46] like every day this unit sub-optimized we've left some money on the table. And so it really did shift from a monthly to a real time conversation uh fragmented to unified a lot of a lot of effort to bring everyone together and agree that this is how we're going to calculate it. And much more uh move it from reactive
[36:03] to a much more proactive approach. Um so a couple of things we did see with this one. Um we pull a lot of data from OSI Pi, we pull a ton of data from also our BCO org and around real time costing so we can actually put the real dollars right not
[36:18] a an estimated dollar cost so we understood and could make really good decisions. So the impact we saw was the reporting from our uh time from our APC engineers dropped by 44%. The amount of time they were just pulling their story together dropped significantly. The margin
[36:34] capture also improved dramatically as we started utilizing this tool. And the miss value strength significantly. So instead of trying to convince someone for 2 weeks to take some action, you could give them data that everyone agreed is right and then they we can decide what to do about it. So but the real shift, we made APC
[36:51] visible. We made it trusted. And we turned it into a real-time decision solution. All right. So the biggest lesson we learned through this whole journey and this has been a journey we've been ongoing. These are just three examples we've shared. Um is that data quality is
[37:06] as much as is as much a people problem as it is a data problem. Um in a high-risk environment where every decision matters, we're really working to drive insights in from this data so that people can do those in real time. Our digital engineering team focuses on simplifying, standardizing, and
[37:23] designing data around the way our front frontline workers are working. This is a human problem, not a technical problem at its core. So Databricks gave us that technical foundation uh to build trust, to drive adoption, and behavior change so that we could
[37:39] move from designing the solution the foundation around the people the way the people actually work. In a high-risk environment, people that trust data will uh they know that and they know what's the single source of truth, they're much more inclined to take action on that data.
[37:58] All right. So trusted data drives um not more data. So again, it's not we don't have a data problem. It's not a volume problem. Trusted data drives more data and therefore or the trusted data drive and not more data drives action. In a high-risk environment, this sheer volume of data uh doesn't help if they don't trust it
[38:14] or can't make sense of it, honestly. Um hesitation from low confidence costs us real dollars um both either safety, reliability, and certainly and profitability outcomes. Data trust erodes when definitions are misaligned or not standardized, and that
[38:29] was a real problem. It drove a lot of conflict. Inconsistent KPI definitions across sites, fragmented tools, and lack of ownership also drove a lot of bad patterns and old patterns coming back and just relying on
[38:44] what you know. So, Chevron's manufacturing digital energy team takes that few people first approach. Technology is an enabler, not the solution. Platforms are designed around how people actually work, versus or emphasizing on usability, clarity,
[39:00] and actionability. In Databricks, these Databricks driven initiatives have really drawn a real lot of value rebuilding that trust into the system that people really know what the data they're getting they can trust. So, I talked about our purpose. Our vision
[39:17] is to be the global energy company most admired for its people, performance, and partnership. And when people trust the data, they act. And that is how performance is improved. That is the future we are building. Thank you.
[39:42] Thank Thank you, Ann Marie. And I guess you you've realized how the narrative has changed compared to previous years. We didn't talk about data lakehouse and all those things. We're talking about use case that are being solved today. Cuz AI is no hype anymore. Our customer and the leading ones are already
[39:57] creating a ton of value. And thank you again, Ann Marie, for this presentation sharing with us what your team is doing and enabling human progress. Um one more round of applause for Ann Marie, please.
[40:16] Ne- Next up, I'm super excited about this story as well and I'll invite uh Doug and John to come on stage from Shell sharing how we are changing the way we do energy trading. Thank you. Thank you very much, Julian.
[40:34] All right. Good afternoon, everyone. My name is uh John O'Brien. I'm from IT. I'm a portfolio uh manager for data analytics for Shell Energy Americas. I'm based in Houston. You probably picked from my accent I'm actually from Australia. Uh I've been working with been lucky
[40:51] enough in Shell to be a pathfinder on the Databricks Lakehouse since 2023. A couple of years ago I got to present to you what we're doing in governance. Um but I'm excited to share how we've been replicating what we did in Australia in the Americas and actually share a journey about how we partner between IT
[41:08] and our commercial business. And with that, Doug. All right. Thank you, John. All right. Uh yeah, so Doug Way with Shell Energy. I lead our West natural gas and power trading business. And I have a confession to make. I've
[41:24] been in the industry for about 30 years. And when I started in the business, uh data came to us through phone calls, fax machine, and uh TV networks for weather, I believe. So, uh again, I'm very pleased to be on the
[41:40] stage talking to everybody about how far data has evolved in the course of my career. But maybe just a little bit about my team. So, I've got uh natural gas and power trading desks underneath me. And we are trading all the way from 1-hour increments in power
[41:57] to 1-day in natural gas all the way out to say 5 years uh for liquid trading horizons and then up to 15 years for our complex structured transactions. So, you can imagine the decisions that need to be made by
[42:12] all of these groups are very difficult with the different timelines that they've got to actually understand. And so, Shell has actually seen the value in data and understanding fundamentals as we think about them in our business for a very long time. So,
[42:27] for at least 20 years, we've had a designated fundamentals team that has been in charge of bringing in the data, transforming the data, and giving insights to help our traders make decisions day-to-day on how we're
[42:44] going to actually uh take advantage of our understanding of the market. So, again, the goal today here again is to um talk about uh the uh importance of data to Shell and
[42:59] how Databricks is helping to transform. So, the usual Shell blurb, cautionary note about future statements, but we'll just carry on from there. Um So, for uh
[43:15] this room here, I'm not exactly sure how many people understand how the physical natural gas markets necessarily trade. It's not like an equity market, so maybe I'd just spend a little time with uh a natural gas 101 trading just to give
[43:30] people a little bit of a basing. So, the natural gas market in North America is a beautiful thing. It's a bunch of pipelines that look like a giant spaghetti if you see it on a map, and it's connecting producers with end
[43:46] users. And those end users take so many different forms. Uh but when we think about the opportunity for uh understanding fundamental information, we have to think about it like this. There's 200 different pricing hubs in North America, and all of them
[44:03] have different supply and demand fundamentals on any given day. And so, on the right here of the chart, you'll see um all of the things that are sort of impacting what will end up being the marginal price at any location.
[44:18] But to just refer referred a few, right? We've got uh production again, obviously obvious driver. But then we have demand. And when we think about demand, that's residential usage, industrial usage. We've got LNG exports. We've got electric generation.
[44:36] So, the the amount of things that are changing on a day-to-day, hour-to-hour basis is a lot. And so, when we think about how this is actually done, there's 300 market participants, by the way, that are also involved in trying to
[44:53] actually have the best trading around these products as possible. So, it's a very optimized system. And it's truly a feature of the US market that allows uh low-cost, competitively priced energy to be delivered at the right time.
[45:08] Uh but most importantly, this happens every day in terms of this example from 5:30 to 7:30. So, this day-ahead natural gas trading. It's over in the blink of an eye. And if you don't have the data in the right time, you've missed it.
[45:25] So, why don't we carry on? Let's see. So, coming back to why data matters. Data matters um because
[45:40] um it's obviously a very complex market in order to deal, right? And so, when we think about this complexity, I hopefully that's been conveyed a little bit in the past discussion. Uh there's there's uh
[45:55] an ability that can be gained as an edge if you have your fundamentals right. So, fundamentals matter. And so, fundamentals matter not like the stock trading markets because equities don't necessarily need to correlate to fundamentals all that well.
[46:11] But, in these markets here where the supply demand balance is so important, you have to balance every single day. The market needs have to be met every single day. And so, fundamentals matter. Speed matters. I think we've actually heard that from a few of the speakers. I
[46:27] mean, this is of course. Right? You cannot be doing this well if you don't actually have a good handle on uh getting this done in time. And then finally, confidence. I think we've also heard that theme come through. If my traders don't have confidence in
[46:43] what the model is putting out, then ultimately, they're not going to be relying on it and decision quality is going to suffer. So again, this is where uh you know, opportunities to transform our data usage through products like Databricks is an actual competitive advantage.
[46:59] But maybe just to have uh John speak a little bit about the IT contribution. Yeah, thanks, Doug. Um a highlight what Julian said is trading is just like an amazing place to work if you're into data and analytics. As Doug showed, you
[47:15] know, 800 data sets, we're going to grow that to 2 and 1/2 thousand data sets. But one thing that we've learned in this journey is that actually IT, we're only good at certain things. And so, having this relationship with the business where we can say what we concentrate on
[47:30] just rolling out the Databricks platform and making that really secure. Let's look at setting up the ingestion pipelines for you. But actually, can we teach these fundamental analysts, these traders, how to build their own silver layers, their own gold layers
[47:46] themselves, build their own data products, share it amongst the teams, and give them that ability to innovate. And so, that's probably my key message and my sort of takeaway for everybody jumping on board um this journey is to partner with your business. We're lucky
[48:01] we sit our our data engineers sit next to the traders on the trading floor. So we can hear what's going on. We can partner with them and not be a I'll not be a bottleneck. Actually be somebody who can help them. We can give them more RAM on their clusters. We can show them how to use Genie.
[48:18] And this isn't something that we've done overnight. I'll uh click the next slide. This is This has been a journey for us. It's not instant. As I mentioned, we we kicked this off in Australia in 2023. We've had a really strong relationship with Databricks back in the machine learning data science days. But we
[48:34] really went all in in in uh in Australia to be able to roll this and try this try this out. Where we really partner with the business and allow them to build and innovate on the platform. And as we as we saw this ramp up and that's advanced users, the person next to them sees
[48:49] what's going on, they want in. As this ramps up, we then look to replicate this um around the world. And we've started with sort of 6 months into the journey with the Americas team. Something changed really significantly in about February this year with the release of Genie code. I just want to say this as
[49:06] well. It takes me about 8 minutes now to onboard a new person to Databricks. Because I show them and we'll show a small demo at the end, you know, how to get to things what they're familiar with and then how to use Genie to teach them everything else. So what value does this give uh to your
[49:23] team, Doug? Phenomenal amounts of value. Okay. Um Yeah, so when I think about empowering the business, what does that mean? So I I talked about the investment that Shell has made for such a long time in having fundamental analysts, right? But when I
[49:40] think about how our fundamental analysts have behaved, they've been the ingesters of all of the data for so long. They've been doing all of the manipulations to the data and then they're putting it back out into the community. So, there's there's only so many people
[49:56] that you can have doing that. And so, you hit this congestion level. And so, when I think about what we can actually do to empower the business, well, first of all, we can get a lot of our systems off of Excel or whatever else we are using, although we love Excel, right? Um
[50:12] but yeah, there's there's an opportunity here for us to get things into a solution where so we're not spending so much time ingesting data, manipulating data, and then doing something else with it. And so, the the vision or the the the fantastic thing that we really are
[50:28] starting to see is that you can actually sort of debottleneck people. Right? And you can actually have people that want to run their own models start actually doing this rather than having people that are running their own models on the side. So, again, this this shared platform
[50:44] here, I think there's an opportunity. The probably the other thing to really touch on though again is this standard thing. We've got people in the South gas team. We've got people in the South power team. We've got all these different groups that are not not necessarily in my group.
[51:00] Everybody's built out their fundamental analysis slightly different. Right? And so, it makes it very difficult if somebody leaves or goes on vacation for anybody to jump in and actually pick up any of that work. So, again, I I think there's um a massive opportunity here to again
[51:18] supersize what we've got going on with traders getting involved um and then having that common platform. So, extremely excited about that. What if we have a a look at what that kind of what that looks like for an analyst or a trader um on the desk? And so, we've got a little
[51:34] demonstration of of of what this platform looks like. So, you can imagine you jump into our catalog and you can find a whole bunch of data. This is S&P Global allows through Delta Sharing for us to instantly get all of our North American gas data, all catalog, good
[51:50] descriptions. So even from the catalog, you can just ask a question in natural language. Uh to just immediately dive in, it's going to show you the sequel that's being produced um thanks to Genie, and be able to give you an instant answer, you know, directly in your catalog. This
[52:06] is a This is a, you know, entry point. A lot of our teams will then build dashboards to be able to share with the business. And so this is now capturing those high-level metrics. So this is that uh S&P Global data set. Multiple pages.
[52:22] But of course, some of, you know, some of what you're asking may not be on that dashboard. So seamlessly integrated, again, is Genie. So let's ask a more complicated question from that data. And for it to be able to have access to that context.
[52:37] You're able to think through, come back after a few minutes, come up with an answer. You can even save that out to a PDF. You're able to share that with the team. But we go beyond that. These fundamentals teams work in notebooks. They're quite proficient with Python or R. So we use Genie code. How do we make
[52:53] that safe for our business? So we've been using instructions. So that we can sort of tailor exactly how we want them to make use of Genie code, and skills. We've been encouraging the business to create skills. We've actually got a handful of reusable
[53:08] skills that we deploy in the workspace to make their job even easier. So in here we can run an advanced um uh um a model on that same data. It's able to fetch it, preview it. Be able to help the business create the model, and be able to do much more
[53:24] bespoke custom visualizations to be able to answer those questions. The teams themselves can schedule these jobs. And lastly, we love the idea of apps. You know, break away from what you're constrained in the in these uh AI BI
[53:40] dashboards if you want a bespoke interface with interactivity really high quality uh interactivity to deliver to your business. We've even made a shell design skill so that all of these fit the design off-shell, look beautiful, and these
[53:56] analysts can concentrate on that actual data actions um be able to deliver value to the business. So, I just wanted to have a big sort of thank you to Doug as we as we uh wrap up on that because of the the opportunity to be able to take what we learned from Australia and then
[54:11] roll this out to to the US and be able to sit with your team has been amazing, Doug. Thank you, John. So, yeah, maybe just um a couple of final reflections really to to to bring it home. So, hopefully I've made the case for why data is so important to a trading organization.
[54:28] Obviously, if you get it right, it is a competitive advantage. It's a It's just a matter of time though. I I think all of these things as time moves on, you go from the fax machine to what's better This is just table stakes. You have to be keeping up and keeping that
[54:43] competitive advantage. And so, really the goal here again is to be taking our fundamentals investment and actually supersizing that. And so, really if we can get additional productivity then this is going to be a phenomenal thing. Now, I I think as John has touched on, he's already had that success in Australia.
[54:59] He's already seen his teams be able to have the amount of time it takes to analyze data shrink and the amount that they can actually analyze grow. So, um you know, I think it's a phenomenal opportunity. We are sort of just getting through this journey here. We're already
[55:15] starting to see the benefits. And I absolutely love for I look forward to trying to get on that uh awards list next year. Is that right, Julian? Anyways, thank you very much. Thank you.
[55:31] Talk Talk about IT OT convergence, right? We just started happening here on stage. Again, I hope you notice that this is real stuff, right? This is what traders are using every day. And I hope Doug won't mind me saying that traders are not easy to earn the
[55:48] trust from. They are very suspicious of what are you trying to do? And I see some I hear some laughs. People have been on trading desk. I love that. But again, this is real use cases, right? The again, the AI is no hype anymore in that space.
[56:04] And to be honest, being able to bring all this data that is required to take the right decision requires a data platform that can really scale and take all these different data sets. So, again, please a round of applause for Joan and Doug and for doing a live demo.
[56:20] Uh that was awesome. Next up, I want to uh welcome Vinit uh from Constellations uh that will be sharing how Constellation is actually using data and AI to impact the world. Thank you, Vinit. Thank you.
[56:48] I can get the slide to work. All right. Well, thank you so much. Good evening. Uh my name is Vinit Vijayakumar. I I am the vice president for the business management office of IT at Constellation Energy. It's a really fancy term for a very simple job. I make sure that the technology investments we we make are
[57:04] thought thoughtful. I'm making sure uh my team make sure that the architecture decisions we make uh are meaningful for how we build and make sure that the data we build and the business uses is trustworthy, which was the common theme from what you all said.
[57:20] I'm not a data engineer. I don't have demos to show. Um I cannot write Spark jobs. But what I have to do is tell my CIO at the end of every week, what are we spending time and money on and why don't we have the business answers just yet. And that's the lens I'm bringing to this discussion
[57:36] right here. Want to start with a little bit of who we are, who Constellation Energy is. And it matters to the story that I'm here to share with all of you. Constellation is the largest producer of clean energy in the United States of America.
[57:51] We have the largest nuclear fleet fleet in America. 12 stations, 22 nuclear plants. We're a Fortune 200 that can generate 20 that can power 27 million homes. We serve 2.5 million customers in the
[58:07] United States, and 80 of the 100 Fortune 100 companies buy energy from us. When Microsoft needed a partner to build its AI data center, they reached out to us. We're restarting the Three Mile Island nuclear plant under a 20-year power purchase agreement.
[58:26] And here is why this matters to the story verse I'm here to share with you today. Behind all of this are thousands of sensors. Thousands of sensors in our nuclear reactor, thousands of sensors that track every asset through our asset management system and supply chain systems, and sensors and data that we're
[58:42] using in our trading floor. A bad data decision in our world is not just an audit finding or a missed quarterly target. It is a safety event for us. If I rewind just 6 months ago, we closed the acquisition of Calpine Corporation.
[58:59] $26.6 billion, one of the largest gas producer gas power generators in the United States. Overnight, we absorbed 79 gas plants in over 22 states. We added 27 gigawatts of generating
[59:15] capacity. And an entirely separate technology stack. Different trading platforms, different data warehouses, and different way of how we architect solution and build data solutions.
[59:30] We did not just add a business unit, we subsumed a whole organization that's at our size. What was challenging through all of this and is challenging through all of this is we all have a completely dif- different way of defining things. I'll talk a little bit more about this, but what
[59:46] the retail team says customer is is very different from another team and it's a very different meaning from what the gas team or or the Calpine team says. And when uh this question landed on our desk, the primary thing that our team asked is
[01:00:01] can we can our data foundation and what we build absorb this? Can we build towards it or are we back to ground zero? And that is the reason I'm standing before you today. Here's the thing that is most important context for all of you to realize. Constellation is really good at data.
[01:00:19] We're not bad. We have built the best solutions and they solve problems that our business needs. We have machine learning models that predict weather and pricing based on near real-time market conditions. We have world-class nuclear system, our commercial system uses sophisticated
[01:00:35] analytics. We have systems for finance, supply chain, HR. They all work and they all work perfectly. But none of it was designed to connect with each other. We had plenty of data. What we did not have was a knowledge
[01:00:51] base. We had the facts in dozens of systems, but no shared understanding of what those facts meant. And that's a point that I heard from some of my colleagues here. So, when we ask the question how many customers do we have, it really depended on which team you asked it.
[01:01:07] If you asked the retail team, it was based on the billing meter. If you asked the wholesale team, it was a counterparty. If you went into the CRM system, it could be an account or a use. None of these are wrong answers. In some cases, they overlap,
[01:01:23] but they don't all connect. They connect. They all have a different meaning. Now, multiply that same problem by the acquisition we just made, and you can appreciate the problem we were grappling with. And that is a shift we're making. And we're still in the middle of it as
[01:01:38] we're building our enterprise data platform. I want to be really honest about what's the hard part of building the enterprise data platform, which is based on Databricks. The platform decision, it took us weeks.
[01:01:56] Databricks was the right choice for us because it gave us room to experiment, to grow, to mature as we were on this journey. And it has been an awesome platform to build on. The ability to interoperate with our architecture was a selling point.
[01:02:12] But the platform is not the hard part. Getting the business to agree on what a customer means, we're still working on that. Every domain had a reason why their definition was correct. And they're usually not wrong. You're not correcting people, you're
[01:02:28] asking smart people to adopt a shared understanding of what that means for another team, and make it work for the enterprise. And that is a super hard journey to take. We call this the shared meaning. We're working with business leads to define this definition.
[01:02:45] Very much like Doug and John are probably working together, we're trying to replicate it across our whole enterprise. And it's a hard journey that we're on. Then we answered the question that kills most platform, who owns that? Which is where the governance piece comes in.
[01:03:01] And I know Julian talked about governance, you know, Unity Catalog and all of this all of those technical concepts, but to us it was more than that. It is how does the business own meeting meaning, and how does IT support the platform and security so we it can be
[01:03:17] trusted. We're building one governed data set on top of another. It's a hard journey, but we're staying put with it. I can talk about our architecture all day long. Um but the only question that matters is
[01:03:33] is it working? Are we having any early success? And there's three use cases that I'll talk about. NOAH AI is the first use case. It's an assistant for our nuclear engineers. Now imagine yourself being one of these technicians walking into the nuclear
[01:03:49] plant. There's 40 years of history in that plant. Decades of operating manuals manuals that describe how each knob works. Finding an information at the critical moment should take minutes, not weeks. And that is a problem problem that NOAH
[01:04:05] AI is solving for us. But before I go to the next two use cases, I think the one thing I want to sit what want to take what for you all to take and sit with is the fact that we're storing nuclear data in Databricks. This is export controlled information,
[01:04:23] equipment condition indicators, inspection records, and RC compliance documents. A governance failure on this data is not just an audit finding, it's a national security risk. When we had to prove that Unity Catalog had the access control requirements to
[01:04:39] meet the needs of supporting ECI data, it took us months. We got the approval from nuclear and legal to actually go ahead with it, and that gave us the confidence that we could build on it. And that was the most critical achievement that we've had.
[01:04:55] The fact that the nuclear leadership has trusted this has enabled us to deploy more use cases that we're building on. Which brings me to the second use case. For those who don't work in the nuclear industry, it might be hard to imagine what outage planning in a nuclear reactor looks like.
[01:05:11] For somebody who's gone to a college town, it it's kind of like the football game that happens where everybody congregates and the population of the town expands by 10x. A lot of people come together and they're all on a tight schedule to make sure everything works exactly
[01:05:27] as it is supposed to so that we can change the spent fuels, we can do an audit, or we can make sure there's a maintenance thing that is completely tackled. The whole process is called outage management and it is super complicated. It's a process that we've perfected over the 40 years of nuclear industry and it runs on a 28-week rigid
[01:05:45] cycle. One step after another and everything happens in certain specific ways. Scheduling in one system, work execution in another system, and asset health monitoring in another system. And every planned downtime costs millions of dollars.
[01:06:01] What the team is doing is pulling together all the data and data bricks to drive that and using AI search to drive insights on what is the next best action to take. What is the next plan to take? How do we update our schedule? The opportunity we're targeting is 20 to 50 million in just uptime savings.
[01:06:18] More critical than that and what's not on the slide is the fact that Julian mentioned. A lot of the people in our industry are retiring. We need to find ways to empower our people to have that support to be able to do more with their job. And that's the bet we're making.
[01:06:36] Another use case that we're doing, this is in our sales space. Our sales reps spend hours before every customer conversation just assembling data. Pricing from one system, usage from another system, the contracts system has another data. And half the time the numbers tell
[01:06:52] different stories. What we're doing now, the AI surfaces highest value leads and it recommends the next best action for each of these. All of this built on the same governed data set in a a workspace. We already delivered $7 million dollars
[01:07:10] gross margin. We see early signs of tremendous value and we're doubling down on this effort. So, we're scaling the pilots that we've deployed. Shared our use cases and our story and there's a lot of boxes on this
[01:07:26] slide, but if there's three things I want you all to take away from Constellation journey, here's that. One, the semantic conversation is the real work or the context that as we've talked about over the last 2 days.
[01:07:41] Nobody wants to sit in the room with the business arguing for 4 hours on what exactly a customer means in different contexts. But, that is the most important conversation that you'll have, which won't feel like it when you're having that. Second, you cannot mandate adoption.
[01:07:57] We had teams build platforms over many years and they work well and they're generally proud of them. Telling them that you just have to move system and start using ADP is not a memo you send. You have to earn that trust. And that means building and showing that
[01:08:13] there's more value that's created with every use case that you put on it. And the third and the last one I'd say is we lost momentum early on trying to create the perfect solution. If I could go back and tell myself and our team, it would be just put it just get the small wins out
[01:08:29] there. Let's be imperfect. Let's get more people on there and let's let let's roll it out. So, where do we go from here? Our biggest test ahead of us is the integration of Calpine.
[01:08:45] We have a way of how to ingest all of this data into our systems, but we have work ahead of us to prove it. We're not further along because of a bigger team or a bigger budget or because we are in a tear because of all the AI demand.
[01:09:00] We're making progress because we are shifting how people work. We're partnering with business. And there's no finish line that we see ahead of us. We stopped asking the question of who can build can IT build me this dashboard and we're starting to ask the question of what does this mean for the
[01:09:15] enterprise? What do you need from the dashboard? Explain the meaning. That's the question we're asking. And I know most of you in this room are getting asked about AI. Here's what I have learned. AI does not fail because of models. Models are commoditizing by the month as
[01:09:30] you all know. AI fails because the data underneath it is not governed and that's the journey we're on to fix. The shift from pipeline thinking to meaning thinking is the shift we're trying to make. Databricks has given us the great platform to run this experiment on but
[01:09:46] the experiment is ours to run. We understand our business. We are partnering with our business counterparts to really define this meeting and this is not something you can just outsource to a vendor. You have to get in there and get your hands dirty. We're not at the end of this journey. We're at the middle of it and some days
[01:10:02] it feels like the beginning but the foundation is holding and we're learning something new with every challenging day. Finally, the last thing I'll say a takeaway for this room is without context, data is just noise. Thank you.
[01:10:24] Thank Thank Thank you, Vinit and I love when Vinit goes on stage and numbers are always impressive, right? 55 gigawatt, $27 billion acquisition. But I'm sure you're hearing the word that keeps on repeating through these different presentation and that is trust, right? And knowing that the
[01:10:39] nuclear team is trusting Databricks to have their data in there was a big step for all of us, right? And and I'm very excited about that. I'm super excited to invite last but not least Brad Walker from Alinta on stage and I have a uh, confidence to to share,
[01:10:57] right? I go to Australia twice a year to meet with our customers and every time I go, Brad is always offering his time to be sharing with all the other customers how he's been taking his journey with Databricks a long time ago and how he's been um, really helping his business,
[01:11:14] right? So, I've always gone there and every time Brad is with me through my journey in Sydney, Melbourne, and Brisbane, and so on. I was always wondering when can I get Brad to see many more customer in sharing his story. So, I'm excited about having him today. Thank you, Brad.
[01:11:29] Thank you very much. Well, I uh, I've got to thank uh, Julian for the opportunity to speak to you all today
[01:11:45] and it's it's really a pleasure. And one of the things that struck me and I'll I'll I'll use the benefit of being at the at the back of the conversation is Julian gave us no mandate on what we needed to talk about or themes or insights, which is a little bit challenging. We would have liked a bit
[01:12:00] of a hand, but uh, what I'm astounded by is how all of us from such different companies and different backgrounds have touched on so many of the same themes. So, what I would encourage you to today, if you hear a theme that rings true to you and you've heard it from more than
[01:12:15] one of us, know this, it's an actual reality of what we're all facing and what we're doing. So, I think it's a really interesting litmus test for what's really happening. Now, the second bone I need to pick with Databricks is I should have learned. I've been on the front of the of their way for the last 7 years and every time
[01:12:33] we think about something, they release exactly what we want. And we've got to pedal a little harder to actually keep up and keep it up to date. So, there's actually a slide in here that I'm going to have to uh, do a little uh, freestyle on because uh,
[01:12:48] there's an amazing new capability that's coming out that really goes to John and Doug's comment about the value of speed and the value of trading. And uh and that uh that L tap and that's that's looking like some really uh that's some good juice. So, anyway, let's get let's
[01:13:04] get into it. Um was going to do a bit of a standard transformation journey because one of the things that's unique about Alinta is that we've been doing this for 7 years and we have uh we have uh managed to put one data platform under our entire value
[01:13:19] chain. Uh but instead of uh the usual glitzy transformation journey story, I thought we'd just talk about what we did. So, who's Alinta? I don't have a lot of time, so I'm going
[01:13:36] to skip the full company brief and uh and just get stuck in and there's not many architectural slides here either. So, we'll uh we'll get into the content. But Alinta Energy is a gentailer, and in Australia, that means we generate, we trade, we retail. And so, that gives us
[01:13:53] the complexity uh that is absolutely a challenge to deal with from a data platform perspective. We also have one of the densest energy regulatory markets in the world, and we operate in that, and that is one of the really unique challenges
[01:14:10] for people who are trying to supply data. So, some stats cuz everyone likes a few stats. So, my platform, the Alinta Data Hub, I didn't name it. It's not the most imaginative. But my platform, the Alinta Data Hub, has 36,000 daily consumption
[01:14:27] queries that's been hit from every part of the business. We have over 2,400 source files, streams, APIs, objects. And we've very much got to a point where we will take the data any way the source systems will
[01:14:43] give it to us. And so, there's a whole raft of capabilities there. But, that last point, 90 plus source systems. So, we're bringing in data from a very rich and a very complex environment. So, uh that that's that's what the system is.
[01:15:00] But, the really important part is what what it does. It It supports trading. It supports generation in our power station. It supports customer intelligent intelligence, regulatory reporting, and the most exciting part is it now supports agentic AI for our executive
[01:15:17] leadership team. Back in 2018, Alinta Energy invested in what I think was a phenomenal uh strategy for IT. And the the most
[01:15:33] important part of that strategy, it didn't tell us what technology to buy. It told us what technology needed to become if Alinta was going to achieve its goals. So, when Alinta chose to hire me in early 2019, I
[01:15:49] was faced with what many of us were at the time, uh which was an aged on-premise data warehouse, uh a new uh cloud data warehouse that might have been struggling a little to scale with the demands. Um and so,
[01:16:05] we turned around and we created a data analytics strategy. But, the fundamental thing that we built out of the data analytics strategy was the idea of the team that was going to support it. So, before we bought any technology and before we hired any vendors,
[01:16:20] what we did was design the team that would carry the platform forward once we got underway. And that team has is fully integrated. It has delivery, it has engineering, it has DevOps, it has data science, it has machine learning, and now we're having the joy of adding AI
[01:16:37] engineers. So, uh so, that's what it did, and then we had a triple mandate. We to design the platform of the future, we had to stabilize the legacy systems and hopefully turn them off. Uh and then we had to develop the operating model that was going to turn
[01:16:53] data into value. So, one of the important things that came out when we were developing the strategy was not that our ambition for innovation was going to get us across the line, but we absolutely had that. But it was the
[01:17:09] understanding that this platform would be a fundamental for this business. And so, when you create a data platform that becomes a critical asset of the business, you're not just losing a report if it goes down, you're losing the ability to operate. So, our CIO at
[01:17:26] the time understood that and through his support we were able to get the establishment of our capability not about innovation, but about risk and value creation. So, it was quite it was quite a good journey and we were we were quite excited about the fact that we'd we'd
[01:17:43] run a lot of workshops, we've got a lot of support. Then this funny little thing happened. Uh I'm sure you all noticed it, but in 2020 COVID struck. And with it, it didn't just take our freedom. For those of you who know about Victorians, we lost a few weeks of our lives. Um
[01:18:01] it took our transformation funding. And that was a real blow cuz we'd just started to get some momentum. But we were very fortunate the business actually chose to partner with us and so every major project the business were doing, the CFO and the executive team with the
[01:18:19] support decided that they would carry the cost of the extension of our platform along with those business-driven projects. And I think one of the really interesting parts about that is that what then looked like a constraint actually now has become the operating
[01:18:35] model because they became our sponsors, they became our advocates. And when you've got the businesses your advocates, it's a lot easier to get investment in your platform.
[01:18:51] Now, I feel like I might be treading on some ground that was already covered well by Doug and John, so I'll do this bit quickly. Um I think you mentioned an hour for for some of your trading cadence. Uh we did a big project not long ago called 5-minute settlement. So, we now dispatch
[01:19:07] and settle our energy in 5 minutes. And that went from 30 minutes. So, for those of you who can do the math, and I'd say that's most in the room, our data grew enormously as a result of that. And so, one of the things that we
[01:19:22] needed to do was to make our platform be able to do that. But, uh one of the things that was really interesting is that uh when you've only got 5 minutes, time is seconds are critical. So, uh
[01:19:38] one of my colleagues, one of my peers that I work with, Chris Pratt, who's the head of our energy supply technology team and manages our trading systems, uh he and his team did some very good work to compress the amount of time that it took for them to get their critical data
[01:19:53] into their system so that they can give back as much time to the traders. Um and one of the things that we did uh was to integrate those team those two systems very closely. Um and cuz Chris was always telling me that you
[01:20:09] know, when when you've got 5 minutes to settle, the seconds matter. So, you know, we we looked at that and we said, "Okay, how can we work this?" So, then Chris and I worked really hard to make sure that his trading system, which back in the day made a lot of sense to
[01:20:25] separate from your analytics platform, uh because you never wanted a big query to take down your ability to trade. Uh Chris uh Chris and I worked really hard to get the two systems integrated and his we feed his decision support in the
[01:20:40] trading models and we get his data back so our traders have one view. Uh the challenge this L tap thing I think might just give us a bit of opportunity to make a tweak or two of this and I'd hope to come back here next year, maybe the year after and
[01:20:57] say we don't have two boxes here, we've got one. So uh yeah, thanks for the thanks for the task. The other thing that we did that was really exciting is that we had an old MATLAB forecasting model and we were able to bring that model into data
[01:21:12] bricks and we're able to bring that in with advanced data science practices and and machine learning standards and so now we've got advanced machine learning forecasting and enabling us to hit and any small improvement in that drives
[01:21:29] a massive impact to our P&L. Uh so you know, those sort of things have been quite positive. Um and I'll move on seeing as we had a good talk about that. One of the things that seems to be the hot topic is trust and trust for us
[01:21:47] really came to the surface when we started talking about deploying AI in a heavily regulated industry. Um so we did some simple things. We built an AI subcommittee to our architectural group. We build a very simple approach to
[01:22:02] evaluating AI across a buy, build, and blend and then we created an acceptable use standard that was plain English and that the business could feel confident in understanding what they were and weren't allowed to do. And what was amazing is that through
[01:22:20] initiatives like our AI tiger team, we saw AI start to really spring up and and and hit its straps and that came from the fact that once people trusted what they're allowed to do, it it really turbocharges what they would focus on.
[01:22:40] I thought this was a reasonable way to talk about where Databricks plays a role in our company. And it is literally from our power station to our boardroom. One platform, all of our reporting, analytics, data science, machine
[01:22:56] learning, and AI. Uh so we we go from our one of our major generators, Loy Yang B, where all the site maintenance is managed off our platform. That's a challenge. Jobs have to be up by 6:00 a.m. every morning or tools don't get picked up. So, we know
[01:23:12] about it if that's ever a problem. We've talked about trading. Our retail teams, our retail regulatory reporting, our retail analytics, our campaign analytics, all of that comes off the platform. Uh we've got uh we've got some of our best and brightest
[01:23:29] uh happening in other areas in our commercial business. Um and the exciting one was finance. Finance were one of our later teams to come on board, but they've come on hard and they're one of our fastest growing teams. So, uh
[01:23:44] then the exciting part is what happened recently and that was the uh emergence of agentic AI into the boardroom. Now, we've heard a lot today and I think this is another part that'll get a bit of a turbo charge, but we've uh we've
[01:24:01] deployed uh agentic AI at the request of the CEO and the CIO. And uh and we've managed to do one thing that uh was really a challenge for the executive leadership team when they came together for strategy days or offsites.
[01:24:18] Uh when you think about the complexity of what they they have to manage. Uh they're dealing with structured data coming from metrics and KPIs, but they're also dealing with a huge amount of complexity in board reports, in market research, in strategy
[01:24:34] papers, in consulting decks. And so, one of the things that Jeff asked us to do was to create a capability so he and his team could work faster and get more clarity. So, we were able to build a complex
[01:24:49] genie and agent-based agentic platform, and it's got 9,000 pages of our most crucial business documentation and 500 metrics from our data warehouse. And that now serves the executive
[01:25:05] leadership team. And uh one of the things is when you're working in the in the back room on data projects, that's uh quite nice to know from time to time is that uh the work you do actually makes a difference.
[01:25:20] So, we uh we had a combined team of Databricks and Alinta people that built this, and we built this system that we call ACE, the Alinta intelligent strategic engine. And we built it in 4 weeks with 70 person days. And we were able to do that because of
[01:25:36] the access to the data that was already in our core platform. But the great thing was because it was a combined team and the Databricks team were working with us. Um the CIO, Nick Smith, dropped us a note and said,
[01:25:51] "After demonstrating and handing over the system to the executive team, uh he said that was one of the most engaging executive leadership offsites that that has in some time." And then he asked us to uh then he asked us to um
[01:26:09] get ready for the next wave because each one of the EDs was selecting people from their teams to build agents in their world. So, one of the challenges uh of uh doing well is they will give you more work.
[01:26:27] So, in summary, um what might look like a constraint, and we've talked about the challenges and the trials and the tribulations of 7 years on Databricks, but it's the reality that in the energy transition, data is exploding. Everything's getting more complex. Thankfully, Databricks are adding a few
[01:26:45] features that are making life just a little bit easier for the data engineers, but genuinely, I think what every energy operator will need by 2030 is this kind of capability. So, we just happened to get there a little earlier than some. Thank you very much.
[01:27:11] Awesome. I I hope you really picked on what Brad said at the end, right? They are not doing Alinta is not doing an AI strategy. They are doing their strategy by AI, right? AI is actually helping with the with the strategy. So, that's that's awesome. And again, about the trust. Look, first, thank you very much,
[01:27:28] and another round of applause for all our speakers today, please. Um You every one of these presentation, there was a themes about IT and OT connected, right? That's the operational
[01:27:43] intelligence I want to leave you with, right? It is about putting the tools that are powerful enough into the end of the people that understand the challenges to actually build the solutions. And the only way this happen is when our IT data and AI teams work in close collaboration
[01:28:00] with the people in operations. That's will conclude our talk today. I hope you really appreciated it. Really appreciated our speakers to come and present what they actually doing with data and AI, and I invite all of you to join us at the reception we having at the lobby of the Marriott Hotel in the
[01:28:17] next few minutes. Again, thank you very much everyone.

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.