Skip to main content

NAB's Databricks and Genie Strategy: Scaling Enterprise Data Intelligence

Summary

  • National Australia Bank unified 180+ data sources into a single lakehouse on the Databricks Data and AI platform, enabling self-service analytics with Genie spaces that allow business users to ask natural language questions of their data.
  • NAB's Ada architecture integrates the lakehouse, Genie, and LLMs to power customer insights, dispute resolution, and fraud detection, with success measured through data product reuse and incremental delivery costs rather than pilot counts.
  • The key execution gap separating banks that see AI value from those that don't is the ability to break down legacy system fragmentation into governed, resilient platforms with consistent data lineage and control.

NAB's Databricks and Genie Strategy: Scaling Enterprise Data Intelligence

Watch: NAB's Databricks and Genie Strategy: Scaling Enterprise Data Intelligence
Enterprise data platforms are critical for banks competing on customer experience and risk management, but legacy system fragmentation, governance complexity, and siloed analytics slow innovation. National Australia Bank's enterprise data strategy with Databricks solves this by centralizing data governance, enabling self-service analytics, and embedding AI across decision-making processes at scale.
Learn how NAB unified 180+ data sources into a single lakehouse, scaled Genie spaces and apps for democratized analytics across the organization, and measured success through data product reuse and incremental delivery costs. Hear from NAB's leadership on data strategy alignment, breaking down legacy complexity into resilient, governed platforms with Unity Catalog, and turning customer insights into business outcomes through Genie, Databricks apps, and AI-powered decision-making.
🤝

Chapters

FAQs

How does NAB use Genie for self-service analytics?

NAB deployed Genie spaces that allow business users to ask natural language questions of their data and receive curated answers based on the organization's data, business semantics, and domain context. The platform also leverages Genie ontology, which auto-generates context from platform usage to improve response quality over time.

What is NAB's Ada architecture on Databricks?

Ada is NAB's enterprise data platform built on the Databricks Data and AI platform, integrating the lakehouse, Genie, and LLMs to support ingestion, governance, and GenAI workloads across the bank. It provides a unified foundation for customer insights, fraud identification, and back-office automation.

Why do some banks fail to realize value from AI investments?

According to this video, the critical gap is execution: legacy systems, siloed solutions, and regulatory requirements prevent teams from scaling across domains, creating gaps in data consistency, lineage, and control that undermine trust in the platform. Banks that succeed establish governed, resilient data foundations before attempting to scale AI deployments.

How does NAB measure the success of its data strategy on Databricks?

NAB measures success through data product reuse—how often data products built by one team are consumed by others—and incremental delivery costs, which capture the efficiency gained as the platform matures. This shift from measuring pilot counts to measuring consumption and reuse reflects a strategic pivot from building to enabling.

Full transcript

[00:08] So, my name is Jon Chee. I'm a solution architect at Databricks, and I've got the privilege to be uh working with some of the NAB team today. So, joining me today, we've got Jon and Nithin, who are both leaders at the National Australia Bank, or NAB. And they're here today to discuss their journey with Databricks and how
[00:24] they they've turned their data into better banking. Uh before we dive into the NAB's journey with Databricks, I'd like to share a little bit more about what we're seeing in the banking industry. As our teams work with that banks both large and small, we're seeing use cases
[00:41] converging across three key themes: driving growth, protecting the firm, and operating more efficiently. As customers progress in their data journey and better utilize their data assets, we see that they're distinguishing themselves from their competitor, whether that's by creating
[00:57] more personalized experiences for their customer base, intelligently identifying and responding to fraud or risk, or through back office automation. One big contributor in the success is a new tooling that you see more and more, especially over these past few days.
[01:14] One of these tools, which you might might have heard of, is Genie agents. So, if you're not familiar by now, uh Genie agents allows your end users and uh business users to ask natural language questions of their data and get a curated response. So, this takes into consideration your organization's data,
[01:31] business semantics, and other idiosyncrasies which your team provides context for. And as you've also heard over these past few days, new capabilities like Genie ontology helps supercharge this paradigm with auto-generated context from just your use of the platform. So, this makes
[01:47] Genie a powerful tool for your business. But just having tools like Genie around isn't enough. By By end of 2026, AI will be ubiquitous with nearly every major bank, insurer, and asset manager having an AI pilot or deployment. Some firms
[02:04] are seeing clear results from the investments. However, others are not. So, the critical gap our teams have identified between those who are seeing value and those who aren't is execution. So, what does this mean?
[02:21] As an industry, financial institutions carry decades of legacy systems, lead solutions, and regulatory requirements. These systems aren't fit for purpose when teams try to scale across domains. They encounter gaps in data consistency, lineage, and control, which undermine
[02:37] trust in the platform. So, firstly, these traditional tools separate storage, governance, modeling, and deployment into distinct tools that really speak the same language. This data fragmentation slows governance, complicates auditing, and
[02:53] repeat enforces repeated rebuilds. As a result of this, platform teams tend to restrict access to the platform. In a highly regulated environment, the risk of providing incorrect information or unauthorized access to data has a real legal and financial implication to
[03:09] the firm. Alongside this risk, it has impacts on innovation. It means teams tend to rely on traditional processes whereby business questions are put into a queue to be addressed by analytics teams, and by the time responses provided, the insights are often out of date.
[03:26] So, while the journey might seem daunting, customers today are reaping the rewards of tackling some of these issues. So, it's with my pleasure I'd like to pass on to John, who'll share more on NAB's experience over the years.
[03:42] Thanks, Sean. Cheers, mate. Thank you. Thank you, everybody for coming along. Thank you, Jin Chee and everybody for actually inviting us to speak, as well. Um what I'm going to talk about I should introduce actually. So, sorry, my name is John Wistner. I run data and analytics for finance, risk, technology,
[03:58] and operations at NAB. So, I'm not on the platform side. I don't do the tech. My guys run the data science and the analytics that then feed off the platform and use some of the tools we're about to talk about. Um, what I'm going to go through is from a strategy perspective, how do you align up your
[04:13] data strategy with your business strategy? Because you've got to keep the two aligned. We'll then talk a bit about the how and what that we've done and that we've delivered. And then also and really importantly, we're going to talk about once you've got the platform done, what is it that you do next? Because that's the key to really unlocking value
[04:31] and actually we're getting return on investment for everything that you've actually put in and all that work that you've done. Um, we're uh, National Australia Bank. We're Australia's biggest business bank and our ambition is to be the most customer-centric company in Australia
[04:47] and New Zealand. Um, we focus on our customers who we want to that we want our customers to choose us because we're the bank that they trust. And we want our colleagues to be truly customer obsessed. We've got about 41,000 colleagues around the world. Um,
[05:03] primarily in Australia and New Zealand. Um, and what we really focus on is we want to be relationship led. So, the Australian economy is really relies on small businesses. Um, and we want to have really deep and rich relationships with all those customers. Uh, we want to be we want to provide
[05:18] exceptional experiences for our customers. So, we want to be fast, we want to be seamless, we want to be friction free. Um, and we want to be there to help them in the moments that matter. And obviously we need to be safe, we need to be resilient. Um, we're obviously a heavily regulated industry. Um, so it's very important that
[05:34] everything we do is safe, secure, um, same any regulatory environment that a bank operates in around the world. Um, so that's kind of our broad um, group-wide strategy that we tried to drive to. Now then, we think about how our data strategy fits into that.
[05:52] If you think about what we need to be relationship led, we need to know the context that our customers operate in. So, we need to know about them, we need to know their businesses, we need to know their markets, um we need to know what's going on in the broader economy. We need to bring all that together so that when our bankers are in those
[06:07] moments with customers, they're equipped with everything they need to be able to have deep and meaningful conversations. Um if we want to provide exceptional experiences, we have to be able to get that data to people as quickly as we can, as promptly as we can, um and through an interface that they're all
[06:23] happy to deal with. And if we want to be safe and resilient, we need to be able to get that in a secure environment that we can tightly control and that we can make sure nobody gets access to anything they're not supposed to get access to. Um our data strategy,
[06:40] which uh lines up with our business strategy. So, we really focus on two things. So, the first thing we focus on is enterprise intelligence, and you can break enterprise intelligence down into two separate components. So, the first part of that is the platform. Our data platform we call Ada.
[06:56] Um I'll get I'll tell you more about Ada's life story in a second. Uh Databricks is at the heart of Ada. Um it's everything we do. And Databricks are on a single enterprise data platform. Um then we focus on decision-ready data. So, you got the platform, you got to get the data in place.
[07:11] Um we've got um about 180 source systems now ingested into our Databricks environment. Um as you can imagine, we get We've got a lot of attributes, got a lot of sources. It's not an easy job to get all that together, but we are getting to a point where we've got a critical mass of data in there and we
[07:27] can start to unlock the value. Um we're taking a view We're taking a position with the decision-ready data where we're going to drive out data products that we want to be as reusable as we can through that middle layer. Um so that we actually can optimize the cost of everything we deliver, and we get into marginal cost of adding value
[07:44] now rather than actually the big cost, which was getting all the data in in the first place. Um once we've got all that data together on the platform, um we need to get that embedded in decision making. When we talk about embedding decision making, we've got two real parts to that. So,
[08:00] we've got AI accelerated analytics, which is where we've got extensive use of Genie. Um and then we've also got customer decisioning. When we talk about customer decisioning, that's really where we're serving things up out of the platform into either a Genie processes or other um decision engines that we
[08:16] have around the bank. And then underpinning all that, and obviously can't underestimate the importance of it, we've got the data literacy part. Um so the data data building data literacy across the organization, building
[08:31] product thinking around data across the organization, we see those as critical enablers to what we're trying to deliver and what we're trying to speed up. Um the transition, I guess, from a legacy data environment where um
[08:47] people looked at the source of the data as being the warehouse as opposed to being the actual source it comes from. Um has been a real challenge that we've had to drive, and we've really uplifted knowledge with people across the enterprise about what data sourcing is and where the data comes from so that
[09:03] they understand the journey we have to go through to get this all in one place, and they appreciate the complexity of building all the data so that we can get the true customer context that we need to feed those uh relationships that we have with our customers. Um and so you can see from that, I
[09:20] think, how when we talk about enterprise intelligence and embedding decision making, what we're truly try what we're really trying to drive to is that um the relationship led the exceptional experiences and the safe and resilient
[09:35] um tenants of the strategy that we operate as a bank as a whole. And so then, when we're talking to executive leaders, when we're talking to the board, when we're communicating within the bank, we can actually link everything back. So, when we're talking to funding maker funding
[09:52] decision makers and other decision makers, we've got one narrative that explains like this is what we're trying to do and this is how it links into your strategy. Now, obviously like strategies look lovely on pages, but there's a lot of hard work that goes behind them. Um and so um
[10:09] you've got to have the why of that strategy, right? So, you've got this is what we want to do, but then you're sitting in a room with somebody who's got millions of dollars at their disposal and you got to persuade them that they're going to give you they've got to give you the money before they give it to somebody else. And so, what we talk about there is like
[10:26] if we want to deliver the relationship-led exceptional experiences, our data has got to be really fast, our processes have to be really simple, and they have to be really resilient. And you look at across a large legacy estate, and the first words that come to mind
[10:42] are probably more complex, slow, and fragile, right? So, you've got lots of different pipelines, lots of different data stores, your models all run off different stacks. Everything takes a long time to change because it's fragile. Anybody comes along with a new
[10:58] initiative or a new request, it has to go through multiple different pipes and approvals. Access to systems is really difficult, and it's also hard to control. But because you're a regulated industry, you have to layer controls on top of these things so that everything's safe. But every time you layer something in,
[11:15] you're building in more cost and you're building in more time. So then, when someone comes along and wants to do something new, not only is it going to be slow, it's going to be challenging for them to get through the process, but eventually what they're going to do is run out of patience, and they're going to build another
[11:31] thing on the end of your existing stack, which makes everything even more complex, and you get into this unvirtuous cycle of making the whole world more difficult. And so, just to to bring that to life, and this links into an example that Nitin's going to run through at the end, um which shows you
[11:46] how we're addressing these challenges with the new platform. Um uh this is Jill. She's an executive uh in one of our businesses. It's not actually Jill, it's a stock picture. Um and so, you can imagine Jill comes along and says she'd like to do some uh analysis and get some predictive
[12:02] forecasts of where we're going to go with disputes, calls about card disputes that are coming into our cost centers uh call centers, sorry. So, the first thing she'd need for that is the streaming data from the call center platforms. So, let's assume she can get that. Let's assume we've got streaming data set up, and it's all
[12:18] coming in, and it's flowing into an environment. What we've then got to do is work with Jill on, well, what is it you really want to see? So, my data teams or somebody else's data team would end up in a cycle of conversations with Jill about, "Shall we do this? Shall we do that?" It'll iterate, take a couple of weeks.
[12:35] Eventually, we'll get to a set of data that we can actually work on. Once they're comfortable with that, we'll start to look for like, "Can we do something predictive?" Which point we'll go and try and get modeling stack. We'll have to say, "Can we find an execution environment to put that model in?" All the times coming through and all this.
[12:51] Um and then finally, once we've actually got that model, we'll have to get somewhere to deliver it to into somebody's workbench. So, that's probably taken at least cycle of months to get there. And at that point, Jill's super excited because she wants to get this thing launched, and it's ready to go. But then we have to go back over the
[13:07] controls, and we have to go back over the security, and we have to do everything else, so we layer in more on top, right? So, everything's getting slower and slower, and this is the why that we need to change because it's a very difficult for humans to operate in these kind of data environments. It's even more difficult for agents to
[13:23] operate in these data environments. So, this is uh this is a a simplified architectural diagram, but this is Ada. This is our system. So, Ada's named in honor of Ada Lovelace, one of the first computer scientists.
[13:38] If If those of you don't know about Ada Lovelace, if you get a chance to look her up on Wikipedia, she's got one of the most fascinating life stories you'll ever read. Um but, we've gone all in with Databricks for our enterprise stack. Um we've got the lakehouse within which
[13:55] we build our data products. That's opened up to Genie, so we we use Genie rooms and Genie code. I can't keep up with the names that have been announced over the last 2 days, so I've just got to I've got to go with what I know. Uh then we've got access out to the LLMs. We've got AutoML and MLOps built
[14:12] into the platform, so we can actually bring everything in together. So, what we're trying to do here is align with our organization organizational incentives for um for speed, resilience, um and scale. And what we find in here is that we can
[14:29] now move a lot faster to deliver the things that we used to struggle to deliver quickly. We've always delivered them safely, but it's been very difficult to deliver them quickly. Um and so, we this is trying to align the data platform back into the data strategy into the business strategy. Um
[14:48] and this isn't an easy exercise, like I said. So, we've got 180 sources now on the platform, ish. Changes every day, going upwards, obviously. It doesn't go downwards. Um and in order to get them on the platform, it's like in in in whatever data platform you choose, it's not a
[15:03] it's not an easy not a cheap experience for people to go through. So, what we what we looked for with our data strategy was we looked for across the organization for where are the things that the organization is trying to do that deeply reliant on data. And what we found at the start of our
[15:19] journey was that our financial crime teams are looking to do a lot of uplift to their data and their data sourcing and we've gone through a whole transformation there of building new platforms. Um we found that our climate teams were looking for um
[15:34] much better access to third-party data. So like the insurance provided data so that they could do more prediction around the impact of client climate on the portfolio. Um and the third one was our marketing uh teams. They were looking at Nynex actually was previously part of the
[15:49] marketing team. Um they were looking for much richer and smoother access to our customer data. So if you think about what that takes, um the the climate guys need better external pipes so we can get more third-party data. The the marketing teams needed better
[16:06] customer data. So we had to fresh out the customer data. And the financial crime teams needed really good transactional data. And so that and that if you think of what a bank does, like and we'll talk more about transactional data in a little while. Um but if you think about what we do, like transactional data is
[16:22] at the beating heart of everything we do. Really is. Um it's like it's how our And you got transactions been the financial transactions. I've gone to an ATM, I've been in a branch. But you've also got like the physical and verbal transactions like I've used internet banking, I've been in I've called the call center. It's just like it's it's
[16:39] it's an incredibly important data set to get right. We've always had it. It was just difficult to get together. It's always been controlled. There was just a lot of layered controls because it was so difficult to get together. But Ada's helped us start to smash
[16:55] through all of that. Um and where this kind of gets you to is you've done this work where you've built a platform and we've largely got all the features we now need on the platform. Obviously we've got a new list of things that people need to
[17:10] deliver following these two days. But we've largely got the features on the platform that we need. We've got the critical mass of data. So, now when we talk to our decision-makers, business leaders, and the rest of the executive groups across the bank, we're saying like, "Look, we've built this
[17:26] um we've built the data platform. We've got a meaningful amount of data in it now. Now your strategy's got to shift from build to consume. Now the question becomes, how do you really unlock the value from it? So, you got to pick your moment in the strategy to pivot
[17:42] from ingestion to consumption and from build to use. Um and we kind of we look at that in two ways. So, the first place is where we say like, we go from ingestion first. So, we've been sponsored by a few material use cases, but we've got a real weight
[17:57] of data on the platform now. And now we need to move to consumption first. I'll talk about a little bit more in a second about how that kind of changes the how what that how that plays out in a changing approach, I guess. Um the second part is where we move from pockets of brilliance to an enterprise
[18:15] impact, right? And so, when you think about those pockets of brilliance, we'd say it's those three things that we've come through. So, finance, client climate sorry, financial crime um climate, and marketing. So, they're the heavy intensive users of the platform,
[18:30] but now we've got to scale that out so we actually get to this enterprise impact across the organization cuz in any kind of I guess transformational program like this, in the early days when you get in the infrastructure in place, there's kind of there's a build it and
[18:46] they will come mantra. Like, if we get this in place, it'll be so good they're going to start to use it. Um and you kind of get to a point where those people actually have to start turning up. And if they don't, it's not going to be a good experience for anybody. Um so, um so that's where we see this pivot and
[19:02] shift from the build and ingest to actually um use and consume. And the way we think about this and so, we talk about scaling intelligence. So, what what I'd focus on again is just going back to the transactional data
[19:18] that we talked about. So, like I said, we've always had it and it's always been controlled. What it hasn't been is easy to get together and accessible for others to use. And so, we've now built a transactional
[19:34] data product that is um in one spot, fed through one pipeline. All models can run from that transactional data set that need it. The The models on it at the moment, to be clear, are specific to financial crime. Um but that's all in place now with one
[19:50] pipe. But if you think about transactional data in banking, we can determine a great deal about customers from their transactional behavior. If you think about your salary getting paid into your bank account, if you think about who your employer is and
[20:05] being able to see where that's come from, if you think about have you suddenly started I was going to use Australian supermarkets as an example there, which is bad idea. Have you have you suddenly stopped shopping at a premium retailer and started shopping at a lower value retailer? Um you can
[20:22] actually then start to think, well, what are my extensible business cases? So, yes, this data set can serve fin serve fin crime. This data set can also serve fraud, can also serve marketing, can also serve credit risk, can also do things like
[20:38] early identification of hardship, can also help us identify which customers are transitioning to digital services and which customers aren't. It can help us identify if customers have moved from one place to another and we need to see if there's anything we can do to help them enable that. It can help us identify if people are maybe
[20:55] beginning to get into some kind of early financial stress, but that's all from that single asset now. And all those examples I've given in the past have worked off one or more different pipes with one or more different models. And again, if you go back to that point, they've had a layer
[21:10] and layer of control on top. And so, as you get to this use of the platform now with a singular data asset and singular sets of pipelines, not only you're delivering better experiences for your customers, not only you're enabling the relationship-led banking to work better,
[21:27] you're also stripping out cost across the organization. You're decommissioning pipelines that are no longer needed. You're removing storage which is no longer needed. Control layers of control that applied to all those other areas and are stripping away. And you actually get back to well, where
[21:44] we freed up, we can now reinvest. And so, as we build the series of data products that we can use repeatedly across the bank similar to transactions. So, I'm dwelling on transactions just because I'm a geek for transactions, but you could do the same with customer. You
[21:59] could do the same with arrangement and account. You could do the same with interactions. And so, that's the kind of logic that we're going through now about well, how do we scale intelligence? And I think if you go back like if you look back to how we how this would have played out before for Jill when she had
[22:14] to do this and we've got like about 150 Jills across Um What what you'd say is like I want to build a model to do X because it will help me enable customer experiences. But what would prohibit Jill from doing X is that she'd have to
[22:31] pay for everything that preceded the model. So, we'd have to get the data together. She'd have to build the pipeline. She'd have to get the software. Now, if Jill comes along, the first check can be well, do we actually already have that data in here? Do we have a model that does similar things? So, Jill's cost to deliver is now an
[22:47] incremental cost rather than the total cost. And it's that kind of play that plays into your strategy outcomes and your strategy conversations with people about driving towards everything needs to be incremental, but you retain one control plane. And so, you're not layering in cost and inefficiency. You can move
[23:03] really fast. You can move really cheaply. Um shouldn't say cheaply, efficiently. But you can move for a lower cost. Um And so that's the scaling intelligence component. Um Keep pressing the wrong button. Um and
[23:19] so scaling intelligence then we look at like democratizing value. I didn't realize how popular the word democratizing was until I got here. It's very It seems to be used in almost every conversation. But we want to democratize the value. So we want to give the value that's in that data to everybody. And the way that we're really thinking about that now is a lot about um how we deploy
[23:36] Genie. So we're looking at Genie in two ways. So one is um the Genie spaces. So developing those. So if you think about what I can give um if I have a transactional data set and obviously um de-identified so you couldn't identify a customer out of it. Um but a
[23:52] de-identified transactional data set if I can build a Genie room on top of that transactional data set, I can service my group economics team who want to know how the general um economy is performing and who's buying what. I can satisfy business bankers who are
[24:08] about to go out and talk to a dentist somewhere in Queensland and they can find out how dentistry is performing in that general area. I can give it to retail bankers who need to know like where are the customers spending in their local areas. So we can suddenly go get great power of data out
[24:25] to the bankers, out to the internal people as quickly as we possibly can. And so we're looking at that's how we're looking at Genie rooms. A lot of our A lot of our usage of Genie at the moment as well as speeding up the analysts' day jobs. So the people in my team who when they got a request from
[24:41] the business they would have to they you know they'd do what anybody does, right? They'd go off, they'd find the data, they'd write the code, blah blah blah, they'd come back. They'd be told it wasn't quite the question that was being asked so they'd do it all again. Um and then eventually maybe that would become a dashboard. Um Um, but maybe it wouldn't. Um
[24:57] but those guys can now work off pre-prepared genie rooms. And so, somebody comes through with a query, it's one question to them. They can check it. They all write SQL and Python, so they can check the code, make sure it's doing the right thing. They know the data sources. Um, so it takes a lot of grunt out of
[25:12] their daily work. Um, and as we scale that out across the organization, again, you're taking out that layered cost of the transactional cost of people working between kind of get X, kind of get Y, kind of get this. You're exposing it up front. Um, and
[25:28] then the other way we're using genies of genie code. So, again, my analysts are heavy users of that in terms of preparing queries a lot more promptly and efficiently. Um, again, they're all very fluent in the languages that genie works in. So, when when they can check the code before
[25:44] they give people outputs, they can check that things work. Um, but the way that we democratize is using those things to move faster and faster. And then we're looking at apps as a way to deploy it. Um, and so that gives us a way to move like with the speed, scale, safety, and simplicity that we're
[26:00] looking for. Um, and this is where I say like so you go through the the stages of the the strategy. You've got the how do I align this with business strategy? You've got this whole why do we need to do it now? You go through the period of the build
[26:16] where you find a bit of sponsorship. And then you get to that point where you've got to transition and say, all right, let's unlock all the value on this. And that's when we start to go down these two pipelines. When you're measuring your strategy, you will obviously shift as well. So, in that first phase, you're looking
[26:31] at like what features have we delivered on the platform? And you know, like a binary count type thing. Um, and then you're also looking at um, how many sources have I ingested? And am I meeting the business outcomes required of the people that we're currently delivering those sources for?
[26:47] Now, we're in an environment where we're looking much more at well, what's the marginal cost of delivering new features and capabilities from the platform? How much have we got self-service enabled across the organization and how much time are people spending producing um data and outputs for analytics?
[27:03] Um And again, if we come back to this just as a kind of started out, right? We want to move quickly, we want to move at scale, but because we're a bank, we've got to move with speed. Uh sorry, we're in with safely as well. We've got speed, scale,
[27:19] safety. Um and so what we found in the platform is that um we've got with the underpinning nature of Unity Catalog, we can now trace anything that moves through the platform. With ML Ops um and auto ML, we can trace
[27:36] the models and we can see what's coming through the models and they're easy to like track the lineage through and see how they're working. Genie again with access control coming from Unity Catalog means that we can actually start to scale Genie safely and secure securely. And then obviously the apps inherit their requirements from
[27:52] Unity as well. And when we're talking to senior management, when we're talking externally about what we do, like obviously with data platforms, everybody likes to see the value. So everybody likes to see the flashy stuff on top. If you can sit with a
[28:07] somebody representing a regulator, somebody representing a board, and you can say here's a model and at one click, I can show you where the data from that model where the data in that model comes from, how that's transformed and where it goes, and I can
[28:23] show you the quality of all of those inputs, like you can knock them off your feet. And like we we obviously work in an industry of trust. You the ability to build that trust with people and your data and in the process that you're doing and that obviously in a in an energetic world, that trust
[28:40] becomes all important. Um the platform gives us an ability to do that. So the the safety bit really can't be overlooked here. It's like it's key to resilience and being able to demonstrate that your data platform is resilient and your data pipes are secure. And so, um
[28:56] I think I'm going to hand over to Nitin now. He's actually going to provide us with an example of how all this stuff comes together. Cheers, mate. Thanks, John. Um My name is Nitin Sachdeva. I head the um AI data science team at NAB. Um
[29:14] and today I'm going to take you through what my team does in building capabilities um data and AI within Databricks. So, I'm going to take you through a on uh Databricks apps, where we have
[29:30] connected the apps to Genie um on the enterprise data, and also to LLMs for contextual reasoning with the data. So, the question that John started with uh in
[29:45] his conversation with the exec, Jill, um on, you know, information, what insights are driving for why customers are calling regarding disputes, um you know, we can see how this is going to help.
[30:02] So, what we have done here is we've connected the um we've built a data product called customer 10 data product, which is modeled based on call center transcripts,
[30:18] and use that data product with uh a natural language uh processing on Databricks apps to help answer some of the questions. So, where it becomes powerful, it moves
[30:33] away from fragmented customer signals to building enterprise solutions and having those insights for the business. Um so if I talk about the example of uh disputes,
[30:50] what we found was the insights that we're generating from the data, it helped us understand the customers who were calling multiple times were not calling because the dispute itself was uh complicated or fraudulent, was
[31:07] because there was an inherent need by our customers to be informed in the process of dispute resolution. And with Genie and LLMs, we were able to explore
[31:23] hundreds of other customer intents that were modeled um on Databricks uh on the unstructured transaction data um to understand where the ambiguity exist and use these insights to shape better
[31:38] notification or clear guidance, and more proactive um service interventions. So using Genie and LLMs in our enterprise um settings, we can reason across
[31:55] customer calls, their sentiments, resolution outcomes, um any behavioral signals to identify what can be some of the avoidable contacts, and what actions will have the greatest impact.
[32:11] So the opportunity here is bigger than just speeding up customer resolution, it's about using data and AI to redesign the experience for our customers. So for data leaders, we are moving away
[32:30] from not only generating insights, but also driving enterprise actions to uh to improve our customer experience. So, if I kind of come back to the strategy and
[32:47] summarize you know, three takeaways. Data strategy creates value when it's tightly coupled with business priorities and you know, we can measure the customer outcomes as a result of it.
[33:03] Now, execution like requires focus but we we need to be agile enough to change the direction when we see new signals, technologies or opportunities that emerge as we've seen
[33:19] in the last couple of days. And third one is platform matters. The the reason we were able to answer or deliver on use cases for our execs like Jill is because
[33:34] we had a data and AI capability within our platform that had the strong foundations to make it fast, scalable, resilient within a governed environment and control.
[33:50] So, with that I will pass it on to Junichi to close us off. Awesome. So, thank you John. Thank you Nin. And thank you everyone for
[34:06] listening. So, that concludes our talk today.

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.