Etihad Airways Transforms Data with AI-Powered Self-Service on Databricks

Summary

Etihad Airways consolidated 100+ disparate data systems into a unified Databricks lakehouse, migrating over 80% of company data through a wave-based strategy with coexistence—not a big-bang cutover—from Oracle Exadata, AWS, and Cloudera.
Five platform properties guided the strategy: AI readiness at scale, simplicity, governance, cost-efficiency, and portability, with Unity Catalog enforcing trust and access control as the enabling foundation for all downstream analytics and AI use cases.
A catering pilot reduced analysis from hours to 2 minutes using Genie-powered self-service BI, and a six-month finance transformation demonstrated how governance-by-design and a data-first approach unlock rapid value at enterprise scale.

Etihad Airways Transforms Data with AI-Powered Self-Service on Databricks

Modern airlines need more than dashboards; they need intelligence at speed. Etihad Airways transformed its analytics landscape by consolidating 100+ disparate systems into a unified Databricks Lakehouse. Starting with the data foundation rather than dashboards, Etihad migrated through waves, not big bang, moving over 80% of company data into one platform while maintaining coexistence with legacy systems. The strategy centered on five properties: AI readiness at scale, simplicity, governance, cost-efficiency, and portability. Unity Catalog enforced trust and access control across all tiers.
this video covers the migration journey from Oracle Exadata, AWS, and Cloudera to Databricks, with practical lessons on building federated autonomy in business units while maintaining central control. You'll learn how Etihad monetized the foundation through analytics, Genie-powered self-service BI, and AI agents, from a December catering pilot that reduced analysis from hours to 2 minutes, to finance automation handling fraud detection and AP processes. The six-month finance transformation demonstrates how governance-by-design and data-first thinking unlock rapid value at enterprise scale.
🤝

Chapters

FAQs

How did Etihad Airways approach migrating 100+ data systems to Databricks?

Etihad used a wave-based migration approach rather than a big-bang cutover, maintaining coexistence with legacy systems including Oracle Exadata, AWS, and Cloudera while incrementally moving workloads. This strategy allowed more than 80% of company data to be consolidated without disrupting live operations.

What are the five platform properties Etihad built its data strategy around?

The five properties are AI readiness at scale, simplicity, governance, cost-efficiency, and portability, chosen to ensure the platform could support a broad range of use cases from self-service analytics to autonomous AI agents. Unity Catalog was central to the governance and trust properties, enforcing access control across all data tiers.

How does Etihad use Genie for self-service analytics?

Etihad deployed Genie-powered self-service BI to put analytics in the hands of both technical and non-technical business users, enabling natural language queries without requiring SQL knowledge. A catering pilot using Genie-powered AI/BI reduced analysis time from hours to just 2 minutes.

What did Etihad's six-month finance transformation achieve?

The finance transformation moved financial data from legacy systems onto the Databricks Data and AI platform, introducing AI/BI for finance use cases including fraud detection and accounts payable automation. The CFO's vision and organizational change management were cited as key factors in enabling the rapid six-month timeline.

Full transcript

[00:18] Thank you. It's not until what I'm on anyways. Okay, I think we're good. Hopefully everyone can hear me. Um, sorry for the wait, some technical
[00:34] difficulties. Um, so thank you for attending the session. So we do the first slide here for me. Cool. So, our session here is going to be split into two parts. First part, I'm going to talk a bit about strategy here at Etihad um, Airways and then the
[00:51] second bit I'm going to hand over to Ketan. We'll do a little swap of swap of the laptops um, because of technical issues. So I'm really honored to be here um, representing Etihad Airways, the first company in the Middle East to be speaking at a Data and AI Summit event.
[01:08] My name is Evan Digney. I am head of enterprise data um, platform strategy um, at Etihad Airways and my the vast majority of my career has been in London in financial services and 18 months ago
[01:23] I joined Etihad Airways and was very fortunate to inherit a uh, pretty mature strategy uh, within the company. Um, and then I was brought in to look at the next phase of our strategy over the
[01:39] next 18 months that will end at the end of the year and you'll see some key themes coming out of this presentation hopefully and I thought one of the key themes here was keeping it simple for us and that will be one of the things that hopefully will be a
[01:55] thread throughout this presentation. And the Steve Jobs quote I thought was quite um, um apt here. So, simple can be harder than complex. You have to work hard to get your thinking clean and make it simple. And so, here we go. Too far.
[02:11] So, so with Etihad, we set out to ask ourselves some very simple questions um as an organization. So, how does Etihad Airways ensure that its data is trusted? How do we ensure that it's discoverable, it's accessed um
[02:28] and governed across the organization? Second question we asked ourselves is, how do we ensure that we when we're in a position that we are running hundreds of agents across the organization, we can deal with that and we anticipate that coming. And the third question we asked ourselves is, how do we put the power of
[02:45] AI in every single business user's hand, whether they're technical or non-technical? And within that, we have three very live examples in airline here and relating to these three pictures here. So, if you think about it from from an Etihad point of view,
[03:01] we have one uh example whereby we have a storm, we have aircraft uh that have been grounded. How do we use AI and how do we use analytics to look at how do we re-accommodate potentially those planes? How do we look at the passenger impact on there? How do
[03:18] we make decisions in real time um to be able to resolve that situation? The second use case is that we have a very big cargo business. How do we make sure that we're using AI and analytics um that we are getting maximum yields from all of that cargo space, whether it
[03:34] be on an actual cargo flight or whether it's a passenger um jet where we're putting cargo packet passengers on it. And the third one here is looking at guest experience. How do we make sure that we understand the guest experience? We understand how they're feeling and we
[03:50] can put the power of, you know, the AI into the cabin crew's hand to enhance that experience for our Etihad guests. And what we've seen as an industry is that we are we are moving, right? So, business leaders now, um they don't want to know what happened,
[04:05] they want to know why it happened, they want to know what action I can take as a result of this. And they also want to know what's coming next. And BI is changing. We've gone from the no longer the manual static dashboards, we're moving to something that is, you know, less passive and more active and
[04:22] something you can converse with as a leader. And the platform we built has to service those two things. So, today's analytics and always-on intelligence for us. Um and so, how do you build for that? And part of the the the thinking here at
[04:39] Etihad is that we stop treating um readiness as an outcome and we treat it as a strategy here. And so, we have five properties and one foundation. So, the left-hand side of this, actually the my left-hand side of this,
[04:55] um is how we are doing those things and then the right-hand side is um how we're staying honest about it. So, the first one here is AI readiness at scale. So, we want to make sure that we are building a platform and we're building it not of what we have today, but in the
[05:11] future as well. If we don't think about it in that way, um we're going to be looking at retrofitting a lot of the stuff that we do uh in years to come as a result. Keeping it simple, again, how do we make sure that we keep everything simple cuz
[05:27] every tool is a contract, a skill set, a blueprint. You know, simplicity here is for a choice and the more complex we make things here at Etihad, the difficulty is going to become, whether that's versioning between different systems or when we're running agents, we have those problems. So, keeping it
[05:44] simple is key for for We want to make sure it's governed. Um how do we make sure that we have really good governance um across the platform? How do we make sure that when we are having agents making decisions, for example, they are
[06:00] trusted? It takes, you know, everybody here, data engineers and analytics people will know that it takes a long time to build trust, and it takes 30 seconds to remove that trust from the business. And cost-effective. We don't treat cost as an afterthought. We are very
[06:17] um mature in making sure that we think about cost as an underlying foundation for when we build things. Cost can't be a second choice for us of what we're doing. And so we are very cost-effective. Third one is um port- so the third one this side is portable. So, how do we
[06:32] make sure that we are the masters of our own destiny? We are bigger than any one hyperscaler here. We are currently built on Microsoft Azure today, but we might want to change that tomorrow. And how do we make sure that we're able to do that with our infrastructure?
[06:47] And so here is how the lakehouse deliver these four things in those four layers. So, open storage, governance through Unity Catalog, one query engine, uh consumption in tools that people already use. And this is what Databricks gives us, essentially. All those five things
[07:03] at once into one platform. And so it all starts with the data. And as everybody knows, AI is only as good as the data that can reach. And so this is what we uh started with on the left-hand side.
[07:19] We had lots and lots of tools, really good tools, but they were disparate, they were spread out across in different places. We had many copies of data in silos. And so over the last couple of years, we've been looking at consolidating that view to make sure
[07:35] that we remove as many copies as possible, to make sure that we are um as slick as possible in our thinking. And so the right hand side of this is where we are moving to. Um we are the vast majority of the way through that journey, but we are removing as many copies as possible. We
[07:52] are making sure that we are organizing our data a lot cleaner. We are making sure that when we ingest that data, um it is the the authoritative source of our data, right? And so, how did we do this? Well, we
[08:08] didn't start with um dashboards. We started with the actual data itself. And we wanted to optimize our pipelines and not optimize the business intelligence first on this. And so, we went about it migrating in
[08:24] waves, and we didn't try and boil the ocean all at the same time. We've done this in a logical, methodical order and going by business units in order to do this. Again, AI is only good as the data it can reach, and if it's trapped in silos, then your AI is going to be trapped in
[08:40] silos as well. Um now we have everything in one place. We think about how we end up bringing that to life, um and how do you make sure that um we have hundreds of people operating our platform and it doesn't collapse in
[08:57] on itself. And that's how we go to the architecture side of things here. So, the top here, if we look at it, we have many different sources of data. We are taking advantage of some of the Databricks platform through Lake Flow,
[09:14] but we think about a couple of different scenarios here. So, the first scenario is we only ingest our most important data into our platform where it makes sense. The second one is that we will do zero copy where it makes sense as well, where we don't physically need to copy the
[09:30] data. SAP is a good example of that. We are using S/4HANA and business BDC, and we will do zero copy delta sharing into the platform in that regard. And the third one is it gives us also flexibility to directly connect with the systems, the source systems, as long as they are
[09:46] governed under Unity Catalog. And so, by only don't ingesting that high demand data, it will land in our enterprise data platform. This data platform now contains over 80% of the company data in one place. It's
[10:02] going to approaching 4 petabytes worth of data by the end of the year. And so, when it lands there, it will go through the medallion architecture. We go from bronze to silver to gold. And when we get to gold, that enables us to produce reusable, trusted data
[10:20] products. And those data products are built once, governed once, used everywhere across the organization. And so, before I move on to the next bit, the the key bit here is is actually how we physically set ourselves up as an
[10:37] organization as well. It's very important that we have really strong product management and those product managers deliver these products into the business units that they serve. We create that deep SME knowledge within it.
[10:52] Within hand-in-hand with that, we have engineering chapter standards as well. We want to create deep SME knowledge. We want to be flexible as an organization, but we don't want our engineers moving about every 10 minutes to different places and working on different
[11:08] products. And so, what we end up doing is we create standards. We make sure they hit those standards and therefore when we do need to flex some of our resource, our engineers find themselves in familiar territory. It's not domain specific.
[11:24] It's the same standards. It's the same governance in the same place. It's the same language that they speak. It's just different topics, whether it's cargo or finance or HR, the underlying principles are still the same for us. And so,
[11:39] this federated autonomy at the edge and governed reusable products at the core here. What we can then see is our business units here. So, we've got commercial, cargo, finance. They are using these data products into their own workspaces
[11:56] and they are building their own AI BI. And they are also have the ability to build their own agents themselves. And so, we have agents operating now that are, you know, AP automation, so accounts payable automation ones. We have agents looking
[12:13] at fuel optimization, conversational booking agents. And so, these are federated across the business units. But the principle here is that they are using those governed data products.
[12:30] Okay. So, key thing here I haven't talked about is Unity Catalog here and how do we make sure that we are in this world, governance becomes key. Scale this very, you know, a lot of data products. We have a lot of data to deal with. Trust is very
[12:45] important to us. And a scale without trust is dangerous. And so, when we talk about how do we create trust in an organization, we put it through three different things in the
[13:00] three different ways we think about it. So, Unity Unity Catalog for our lineage, discoverability, and access. We've got Collibra for our quality rules and stewardship. And then we're using AI Gateway for cost audit and agent access.
[13:16] And if I think beyond just the BI element here of it, creating central places where we have catalogs where you can go and discover our agents for the business, what's the purpose of that agent, who created it, who owned
[13:31] it. This is all done through one of our platforms within Collibra. We are looking at a cost control through AI gateway. What do agents have access to? How much are they, you know, costing us? Have they got the right guardrails
[13:46] in place? And then Unity Catalog for our ABAC controls, RBAC controls to make sure that they are governed. We don't want to be able to serve up information to the the right question but to the wrong user. We deal with various different data types from personal information
[14:04] here and we don't want to be able to give that to someone that shouldn't have access to that data. The governance here is is going to be key for us. It's the differentiator at the moment between when we take development from prototype
[14:19] and then we put it into production. And that I think a lot of organizations are going to struggle in the future that when we build things, whether it be AI, BI dashboards or we build agents, making sure that that is well governed is extremely important to us. And so when
[14:35] we put it through those three layers, we give ourselves really good confidence at Attihad that we're able to deliver it on time very quickly in an agile manner to those business units. And so every query, every dashboard
[14:50] enforces the same rules. You cannot govern what you can't trust and you can't trust what you can't govern. And so a lot of this now is happening today for us. So we have MCP integrations with
[15:07] SAP, with Collibra, some of our internal systems. And this isn't a test for us at the moment. It's a real underneath working layer. So Mosaic AI, Databricks Genie, vector search, all sitting on Unity Catalog. And what makes this possible is that AI
[15:23] is becoming a first-class citizen on the platform, and we're not treating it as a separate bolt-on here. So, language model functions running natively in SQL, vector search lives in the right the same vector search lives in the right data under the same
[15:39] governance as everything else. And that's why data that's why agents can reach the whole enterprise and not just on the data platform. And I'm going to give you an example of one agent that we built in Azure Synapse, and this is a a very live
[15:54] example for us. So, we have thousands of pipelines running on our enterprise data platform. And some of those pipelines kick off jobs at 2:00 a.m. in the morning, and this is one of the examples of a job that kicked off at 2:00 a.m. in the morning. And what used to happen is that
[16:10] we used to have people that the pipeline would break. We would have people at 2:00 a.m. Someone would come into work at 9:00 a.m., and they would find out that pipeline would break, and we would have to essentially wait all day until the next run to get access to that data again. Now, your
[16:25] option is to buy your way out of this problem. We could just employ people to monitor that pipeline. We could do some pager duty on that pipeline to do some alerting when it breaks. However, we built an agent to agentically monitor
[16:40] pipelines through a an agent called Osiris. And what that agent will do is on this specific example, it's an ADF pipeline that we've got. When that agent will monitor the pipeline, it will do two things. Number one, it will do forecasting and predicting when it thinks a pipeline is
[16:56] going to fail. The second thing it will do is when a pipeline does fail, we have given it a series of actions that it can take to agentically fix that. So, it will read the ADF error logs. It will then make a decision on what it wants to do, whether that be as simple as just restarting the
[17:12] pipeline or making some more fundamental changes. We're increasing the expandability of that. And so now, what we end up seeing is that pipelines that would end up taking 7-8 hours for us to realize it was fixed sometimes come down
[17:28] to 20 seconds for us. Pipeline fails, Osiris jumps in, fixes the pipeline, remediated, and we never know about it now. And so this is what we are This is how we think about things at Eddy Had to make sure that we are solving problems
[17:43] all the time through through AI, and we're embracing this technology. And so what's next for us here, right? I don't want to come here and say that we have done everything and everything is absolutely perfect. We are part We are in the good way through this journey. We
[18:00] built a solid foundation of what we're doing. We need to add depth. Over the remainder of this year, we need to make sure that we are adding that depth and so we can enable all the business units um to build and to work on data products. We are currently seeing that
[18:16] the demand for data products on our platform is outstripping supply at the moment. And so you can see that you that Katen's going to talk about in a minute that the business are getting out of it is very much front and center for us as an organization. And so the three key takeaways that I
[18:33] think was very key for us here is we wanted to start with that data product and think of the use and not the use case. If you start thinking about a use case only, you will build something inadvertently just to service that use case. We try and think about not just
[18:49] what it's serving now, but what questions maybe want to answer in the in the future, and how do we make sure that that data product can be reused in several different places and scenarios. How does it interlink with the rest of our estate? The next one is govern from day one.
[19:06] We want to make sure here um as a company that we retrofitting trust is 10 times harder than building it in. So, again, very, very easy to lose trust. Especially, you know, those of you that have used Genie, is that if it's providing information that's accurate on wrong data, trust will be
[19:22] lost potentially very quickly here for an organization. The third one here is build on the platform and not besides it. So, every workspace that lives outside the platform is debt accruing interest here. And so, we don't want to make sure that we have things bolted on. We are
[19:38] building them at source. And so, the next bit I'm going to hand over to Ketan in a minute. Ketan is transforming our finance division. Proverbially, we have built the runway here at Etihad, and we're going to talk about Ketan's going to talk about how he
[19:54] is taking off with the plane now. Ketan.
[20:13] I just noticed my name. It said Ketan and Katherine and all that on the on the auto queue. You know who is that? I'm um It's my fault that we're running a little bit late because I'm trying to do too much, and we'll see how this works out. And this is why this is guy is
[20:29] here. I'm going to try and do a live demo a little bit later on. And let's see if it works or not. Um I'm also knackered. It's It's gone midnight in London. It's even later in Abu Dhabi. And I am I I tried to use a What do you call it? Um lots of coffee.
[20:44] But so, if I say something crazy, I apologize in advance. Okay, can I do my He's He's uh What I will say before my slide start, um Young Gavin is being very modest. He
[21:00] started at Etihad what, beginning of last year? So, everything that he's doing there is within a year. It's It's not, you know, he's he he gave in. I I can say things that maybe he can't, which is, you know, it wasn't great. Um
[21:16] and he's done a great job. So, it it's really interesting what he's doing. Um I need my slides first, not this. Sorry. Oh. This is the demo bit, which is going to come in in a little while.
[21:32] I'll start by saying in September last year, I I came in. I'm uh independent. We We've got a company out in London. We were hired to do um to set up finance, finance analytics. So, brand new function. So, when he talked about 80% of the data that was
[21:48] there, that 20% was was the data I needed. So, that wasn't there. And what I'm going to tell you is a story we've gone through since September going through and what we've done. Thank you. I'm a pain all right.
[22:10] Um so, there there we are. So, September 2025, end of September, that's when I flew in to to Abu Dhabi. Um it was pretty interesting. Um they'd already started work on their first dashboard in you know, we're using Databricks. It was a brand new workspace that young Evan had set up for finance,
[22:26] which is great. Uh leadership had a hundred or so priorities of all all kind of set out nicely. And I was like, you know, this is really cool. Lots of good stuff. Sounds great. And then I started looking under the hood. And I'm thinking, okay, what's going on here? So,
[22:42] my workspace had one schema, no data. Um let's say it wasn't the ideal configuration in finance that was going on. The dashboard they were trying to build was using Synapse. Um and it was using a let's say a Tableau
[22:58] server, which was in someone else's space. So, it's pretty interesting. Um like they had legacy Power BI. Um that was very, very interesting in terms of how they put those together and what they were using. Um they had an army of RPA bots, which was just moving
[23:14] data around. So, that's all going on. I'm thinking, "Hmm, something there's no quick wins." So, I thought, "Let's do a control alt delete instead. Let's just reset very, very quickly." The other thing was that knowledge was wasn't quite there. Finance, you
[23:31] know, they've got a vision um and they wanted to do all this stuff. It was part of a three-tier team and and and objective that they had. Um but it was it wasn't there. And if you imagine I've got in in September and you're thinking I've got to redo all this stuff,
[23:48] frustration was coming in. So, if you think of the CFO, the VPs, and those guys are thinking, "What's this guy doing? I'm not got any any developments or anything going on." So, then comes December and there was a priority around catering. They didn't have everything that they wanted um in terms of insight.
[24:06] The CFO was getting worried about how it the costs were escalating, etc. So, I picked on that and we had a look at it. By the way, this isn't and and and you can speak to Nadia about this. This isn't a holiday slide, which is what the feedback I got off um Kyle, I think his name was
[24:22] um at Databricks. It when he was saying he thought this was a holiday slide, it's not. Um anyway, so we did catering. What happened here is you go from having hardly anything in
[24:37] invoices, etc. to having a AI BI dashboard and the Genie space that itemized every 10,000 food items across their whole network and being able to ask them very, very quickly,
[24:56] "What's costing me more?" And then you could actually turn around and say, "Actually, over the business classes, if you say, I don't know, let's say it was beef. You go from economy class, business class to first class, what are the fluctuations in price? You'd expect it to be big cuz it's more quality. But
[25:11] actually, how do that compare across different um flights that we're operating, whether it was London or anywhere else? Cuz they've got it like hundreds and hundreds of caterers that are supplying all this food." It was really, really interesting some of the stuff they were doing. What was taking them
[25:26] hours to do the analysis on, it's taking them 2 minutes. It It It's really quick. What they're able to do and how they're able to actually bring costs down It's just amazing. So, that's what they did. Next 6 months, what have we been up to? So, in finance, um
[25:43] there's a bunch of stuff there. I called it Fin X because it wasn't just analytics, it's also process, it's also technology that we're trying to build it in the finance team and where they're trying to move move things going forward. We've built a bunch of AI BA dashboards, Genie spaces, we've even
[25:58] built rag applications. So, I've got a fleet contracts, 10,000 contracts come through um vector search, which is now called AI search, I believe, from Databricks. No doubt it's going to be called Genie search soon. Um Training for the business. So, we've had the tech guys have have training. We've
[26:14] had um all the business users have training. We've had Databricks do a bunch of training. We've done some in-house. We've had Ari Kaplan come over and do do do some stuff as well, which is really, really awesome. Um I'll cover some of the core stuff shortly in the middle, the green things.
[26:30] Um some of the other things we're doing in the background, so ML data science. So, I have all the cash transactions uh you know, there's there's hundreds of accounts that an airline uses all around the world. Um I've had uh that ML and data science working over
[26:47] since January. It is actually going in and saying, "Actually, fraud detection, doing all the kinds of stuff." It's really cool. I'd love to show you, but I'm not allowed to show that kind of stuff. Um we've also got um Genie code. It's really cool what I'm doing on Genie code. I think it's cool. Again, I can't
[27:04] show you, but we're using it and we're using markdown files and assistant files to actually train it and put the guardrails in. So, I'm building um playbooks. So, it's a very small team. What we're doing is we're carrying a playbooks that actually can rapid
[27:21] prototype um AIBI dashboards, Genie spaces, and that kind of stuff. It's It's It's It's It's really nice what's going on. There's a There's a load of stuff going on. How do we do it? So, going from September to now, very quickly with nothing in there, and then
[27:38] and get and get into the space we're getting into, for me, it's a house of cards, you know, and and as you know with a house of cards, you pull one of those cards, it all falls apart. So, for me, it's just the usual stuff. It doesn't matter that we're using, you know, AI right now. Actually, it's all about the data
[27:54] framework first, and I know a lot of people have been saying that. Um it's not just about all the big technologies that we're using. It's actually how do you create that model in your workspace? And it's it's fundamentally so important, but people just I I'm seeing it all the time
[28:10] because kept it in me in business for 20 years because people keep getting it wrong. The The other bit is semantics. So, just the meaning stuff. And that that's, you know, back in the day there was data dictionaries, all the bits and bobs, and now you've got no choice but to do it because the AI is not going to work properly if you don't
[28:25] do that. Um governance, they were talking about it a lot this morning, so I won't cover it off anymore. Um and adoption, because if they're not using it in the right way, it's not going to work. There's um one really, really important bit that
[28:41] people don't talk about, and I call it enterprise orchestration, and that's actually how do do get a whole organization to shift in a different way of working. So, you've got the the the IT guys, you know, technically know that know their stuff. Some of them do, some of them don't.
[28:57] They they're talking all the techy stuff and just blindsiding everyone. Then you got the business side who know their business but don't know basically the techy side and then you got to bring all that together and then in between you've got everyone else. And then there's a bit on data which
[29:13] where everyone knows data. Every I I mean I look I look into LinkedIn now and everyone's got uh is a leader in in data. They they know data inside out and I blame Microsoft because when they released Excel, they were the first ones to democratize data. I truly believe that
[29:30] and it was really cool. However, it also caused me lots of pain because everyone thinks they know everything about it. Um you do all those things and you get the opportunity to excel and that's the business end users. They're the ones who are going to do stuff. They're the ones who will bring the value, the ROI and
[29:45] that kind of stuff and our job is to enable them. That's it. That's my job and that's what I'm doing at finance. Um I do have another video but you're not going to hear it and I'm going to tell you what he's basically saying is
[30:01] there's if you imagine this guy is the VP of treasury um and investor relations. He also is my key sponsor when it comes to analytics and that kind of stuff. If you imagine what he was talking about at the beginning when I got there which
[30:16] was um he'd like to be involved in how these dashboards look and this kind of those kind of stuff that that the basic stuff. To him, what he's talking about here is actually data is important. How we put that data together is more important than any anything else and
[30:32] he's he's basically turning around and saying, "If you don't do the data properly, the AI doesn't work." And and that's his message and I wish you could hear it. Um I'm really disappointed. We've got another another video later on which I really want you to hear, but we're not going to be able to. So, that's what
[30:48] that's what Marco said. And you know what? He's understanding. It takes some time. You move them across. But eventually, that they they do their own research. He's excited. They're doing loads and loads of stuff. Um, so if I go into the fundamentals and
[31:03] just go actually um, my piece on this little bit. AI is not going to sell sell broken strategies. And that's why you need people like Evan because as a as a whole company, you need to have that strategy. I see it so many times where
[31:19] um, what they were doing before is not going to work. You're just going to battle. That's what that That's that's what happens. Um, and the ones that win in the AI BI era are the ones that do the basics properly as far as I'm concerned. You do all that really really well, you will move
[31:35] forward. Um, and when it comes to Databricks, I would say the great thing is is that you can do most of the stuff in one place. That's pretty cool. Um, and it means you can actually do things with small a small amounts of people, just good people. And
[31:52] you can do exceptional things. And just so you know, um, most of my clients they're financial services in England. Um, that weren't using Databricks. This is the first time I ever used Databricks um, in September. Um, the team is very small in finance.
[32:08] It's the smallest team in the whole of the yard when it comes to analytics. And yet we're still doing stuff. So, now we're going to do the the change to the um, live demo. Let's see if that works. Um, while he's going to switch, you can see some of the uh, the the genie spaces
[32:25] that we've already put together. So, we've been a bit creative rather whilst we're doing like very structured in terms of some of the the bits that are around there. I've also got a metadata one which actually looks at all the metadata in the in the whole company um and moves that forward and and we can
[32:42] go and explore so you can see what data's about about if you want to create a new new insight and things like that. Um we also have a self-service they load their own data um and we give them universal tables that they can play with it's really cool that they can use um and then and it's
[32:59] to increase adoption we've also used some marketplace data which will actually um allow them to play so we use some flight information and other bits and bobs. Here we go.
[33:18] Perfect. There we go. So I can't show you um obviously confidential company information what I can show you about the stuff that I think about um and so I and I think every analytics or data team should have their own metrics KPIs that
[33:35] kind of stuff. So these are the things that I think about so if you think about what I've built behind the scenes you can see what we've been doing across here and this is our uh live dashboard these this is how every dashboard looks in terms of the notes and things like that. We look at
[33:51] user adoption or I do um which tells you where we're at and I'll mention this in a bit. You can see all the people in there so I've got you know Elton David who were you know champions the multi champion you know
[34:08] things. You can go through and see what they're doing there. I I B I dashboards you look at my catalog catalog health what I'm looking at there is what the scheme is I've got it governed the PII that's classified we use medallion um so
[34:24] we don't talk about it we do medallion properly um all in there governance and compliance I also look at pipeline and and freshness around there because actually the engineers should be held accountable in terms of what's going on. And while we're developing it's really really important. I also look at cost
[34:41] and efficiency because obviously I want us to run really really really well. So you go through there and actually I can see who's who's costing me more per query. So these are some of the things I look at. This dashboard by the way cuz I'm
[34:56] playing cuz it's internal for me was all built built with Genie code. So these are all the tables that we've gone through. It took me a few days and gone through through and do that. I did just before we came ask it uh I used you know the the Genie space and
[35:13] and said asked it a question. I said I'm doing a live demo at Databricks showcasing the story. Provide an honest review on the progress I've made and see what it does in a bit of swagger. So this is what he came back with on on on earth uh
[35:30] peace over the air. So one is saying welcome to Databricks. We've built a production grade AI beta platform. It goes through. I'll let you read it as it goes through. Flagship use case flight gain cost. So that was the first one we came out the box. The interactions, the top you you know AI use cases that
[35:46] we've got through. Sustained engagement. The platform maintains strong monthly engagement. We've gone through a user adopt adoption and maturity. And the bottom line 66 users active in the last 30 days running queries at
[36:02] scale across a portfolio of live AI applications total performance platform cost. There which is a you know at this scale this is what production AI looks like. Not not a prototype, not a proof of concept, live scaled and delivering. That's not me, that's Genie. Saying that
[36:18] looking at our data and and doing this stuff. And that's within what 6 months going through. We would switch back cuz actually all I wanted to do and end on was a video from our CFO who talks about his journey in there, but I don't unless you got
[36:34] guys got the voice the sound there, I don't think that's going to work. So, that's not going to work. So, what what What Raffi says is basically he talks about his vision and that's what I asked him to do on a video. He talks about his vision and what he wanted from this and AI and his
[36:49] investment in there. He wants to reduce cost. But he wants to use that to reinvest in the customer experience and that's his big goal. And then he talks about December when we released catering. He was on a flight from Abu Dhabi to to Europe somewhere. He was
[37:07] on the plane and there's not much to do but but he's got high speed internet or or you know it's it's to look at his stuff. We released it to him. And it I wish you could see it and if I see you later and if you want to you want to hear the video, I will show you. The guy just sat around and says that
[37:22] that that flight passed by by 1 minute. He was just amazed about what he could see for the first time in his life using this kind of technology. He was just totally blown away. He's just totally excited. He's he's loving
[37:39] that we're doing so much more in terms of you know cash all the other bits and all the other cost cost payments with him. He also talked about how investment in people is not is he going to reduce head count and he was like no
[37:55] and he gives an example actually of the business partners that that that the catering guys in in operations and they said they'd come back to him and said with all this insight with all this stuff we can do, we actually need more people. And his message is to say actually AI is not going to reduce head
[38:10] count. It's actually going to sustain or even keep or even let's say have even more people because there's more to do. There's more efficiency. There's more more performance to be had, but people need
[38:25] to change the way they do things. And that was his message in there. And he then did it and said he just loved the journey. He's loved everything that he's gone through. And I wish I could show you, but I can't. And that's us. I I mean it it's where we've got to and it's a great journey.
[38:41] We'll show you more if you if you want to talk to us after this, but we're out of time, I think. Thank you.

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.