A Japanese Mega-Bank’s Journey to a Modern, GenAI-Powered, Governed Data Platform
Summary
- Global financial institution scale: SMBC, a top 15 global bank with nearly $2 trillion in assets, is modernizing its data foundation by migrating 20+ core banking systems to a multi-tenant Azure/Databricks architecture.
- GenAI-powered governance: This video demonstrates using Databricks' Unity Catalog and delta-share capabilities to govern data by design while enabling global adoption of modern data frameworks.
- Legacy decommissioning: SMBC's initiative reduces data sprawl, decommissions legacy platforms, and creates a unified data-as-a-service approach for loans, deposits, securities and derivatives domains.
A Japanese Mega-Bank’s Journey to a Modern, GenAI-Powered, Governed Data Platform

Talk by Anshul Wadhawan, Principal, Deloitte Consulting LLP ; Gordon Wilson, Chief Information Officer - Americas, Sumitomo Mitsui Banking Corporation
SMBC, a major Japanese multinational financial services institution, has embarked on an initiative to build a GenAI-powered, modern and well-governed cloud data platform on Azure/Databricks. This initiative aims to build an enterprise data foundation encompassing loans, deposits, securities, derivatives, and other data domains. Its primary goals are: To decommission legacy data platforms and reduce data sprawl by migrating 20+ core banking systems to a multi-tenant Azure Databricks architecture To leverage Databrick’s delta-share capabilities to address SMBC’s unique global footprint and data sharing needs To govern data by design using Unity Catalog To achieve global adoption of the frameworks, accelerators, architecture and tool stack to support similar implementations across EMEA Deloitte and SMBC leveraged the Brickbuilder asset “Data as a Service for Banking” to accelerate this highly strategic transformation.
Here's more to explore:
The best data warehouse is a lakehouse: https://www.databricks.com/product/databricks-sql
How to migrate legacy data warehouses: https://www.databricks.com/resources/ebook/migrate-your-legacy-data-warehouse-databricks
See all the product announcements from Data + AI Summit: https://www.databricks.com/events/dataaisummit-2025-announcements
Chapters
0:00Introduction0:53About SMBC2:34Goals3:44PushPull Factors6:12GenAI7:21Global Bank8:37Riskbased approach12:08Architecture15:35Features17:38Roadmap18:54Implementation21:57Accelerators24:36Key takeaways
FAQs
What is SMBC's scale?
SMBC (Sumitomo Mitsui Banking Corporation) is one of the top 15 banks in the world with nearly $2 trillion in assets and a global presence across the Americas, APAC, Europe, and Japan.
Which banking domains did SMBC migrate?
This video covers SMBC's migration of 20+ core banking systems encompassing loans, deposits, securities, derivatives and other domains to the new platform.
How does governance work in the new platform?
The platform leverages Databricks' Unity Catalog for data governance by design and delta-share capabilities to address SMBC's unique global footprint and data sharing needs.
What accelerated this transformation?
Deloitte and SMBC leveraged the Brickbuilder 'Data as a Service for Banking' asset to accelerate this highly strategic transformation of their data platform.
Full transcript
0:00 Introduction
All right. So. I'm the CIO for Sumoto Mitsui Bank for the Americas and I have Ano Ven. Sorry, gave it my best shot with the V's. Um, we're here today to talk about how our bank is working to really power our data journey and using AI, Gen AI. Uh those of you that may have been in a session earlier, we touched on some of it. There was a p a panel discussion with someone from our team, and uh we we did we did touch on this.
This is going to go a little bit more deep, and I think Ancho will go into some very specific uh tools and methods we're leveraging with Deote to help us get on this journey. If I go to this first slide, you're going to see something about SMBC. Oh, wait. We'll go back one. There we go. Double jumped. Um, Sumato Mitsui Banking Corp. at a global level is one of the top 15 banks in the world almost$2 trillion dollars in assets.
0:53 About SMBC
Uh, it's truly a global organization where we sit in the Americas as part of the global banking unit. The global banking unit includes AMIA, the Americas division, APAC, X Japan, and then obviously domestic Japan has a massive uh mega bank that covers all retail uh corporate investment banking, commercial banking in the Americas. is something to note. We have a few um interesting points that we've recently launched a retail digital bank called Genius Bank.
Uh it's part of one of our group company entities, manufacturers bank. We have a global markets uh capital markets, Nikico Securities that's sales and trading and then we really have that core corporate and investment banking. That's kind of the heart of the business. Now, now the aspiration for us is actually to grow even further. And at the GBU level, the global banking unit level, that means really making our stack more modern.
And what what we heard before, I don't know if anyone heard speak earlier, but it was pretty pretty uh energizing. He's always a straight shooter. He actually said data is the is the problem and the solution essentially. He said without data, you can't do anything. And whether that means regulatory compliance and really the growth if you want your front office whatever that product set is to be successful you have to get there.
So SMBC's goal our goal is to build this so we can be more efficient more effective not only internally but for our clients for our customers and with that you see the you know the leverage to grow. Those who saw the last thing it's like they call it a pilot induced oscillation where you start overcontrolling the airplane and that's not good. That's why you need automation like a good autopilot. So if we look at uh the vision for a data powered future um you know we're we're looking again to try and make this as seamless as possible so that end use becomes very trusted but enabled by platforms that can make us more efficient. as you know the objectives we could get into specific business objectives but really the bottom line is that data has to be consistent has to be trusted has to get there on time and with the ability to move that across regions and we have some AMIA partners here I'm not sure if we have someone from our our head office our real goal is to make this at a global scale which is not an easy task somewhat of an advantage we've been brought up regionally so as we develop a strategy like this we can really go like leaprog to a point where we are.
2:34 Goals
We're on that journey. Now, if you look on the right side, you'll see we're trying to optimize everything. Go cloud-based, leverage all the tools. We'll get into some specifics that Anel will go through in a little bit. Okay, let's see if I learn. If you think about um there's push and pull factors and uh you know, I'll highlight a couple of these, but these probably are not surprising to those that are in, you know, financial services and banking.
3:44 PushPull Factors
Bottom line is we have to be compliant and that means a lot of things. That means regulatory compliance. That means compliance from the standpoint of a second line of defense doing proper, you know, KYC. We need to be able to share information globally. It has to be consistent and we have to do that with the appropriate controls. Again, that's leveraging this platform is the way that we're going to do that. And the way we are doing it, we're on that journey.
If you look at a couple other items, they're kind of almost like table stakes like advancing data governance. Okay. Yeah, of course we want to do that. I think the upper left though, if you think about the aspiration of the bank, if we get the rest of it right, if we're governed, we have clean data, we have a nice pipeline, the real goal is to be better at serving our customers, which in the end will result in the revenue growth.
On the right side, the push factors hopefully if someone's figured this one out, uh, you know, please let me know and we can talk after because everyone has legacy applications, le legacy infrastructure. how we have to bring that together. That is a push factor. It's inefficient. It costs a lot. We burn a ton of calories trying to maintain that and then bring that data together. We know that's going to be a multipleyear journey.
You can't always get out of whatever that platform might be. We all have them in our in our own uh tech stack. But the goal is to make that as efficient as possible again. And where you when you have a chance to upgrade then you can do so. I think the other one is if you look at is the uh redundancies and fragmentation. We had and have a lot of data silos still. We have multiple data data warehouses that exist today running the bank.
And the problem is when you have that you get inconsistent as you all know preaching to the choir a bit inconsistent in that end result. So you may have a different version going to a regulatory report, different version going to your front office for a client engagement and then trying to bring it together. no one's speaking the same language necessarily and worst case you're doing something that could put you out of compliance.
So we're going to continue to like you know have these pull and push factors u you know you know driving us really to our decision. Ultimately though that upper left is really the goal of the bank how do you grow and with that you get everything else that you need. So, a little bit here about uh how we're using Gen AI and I think Ancho will go into some detail. Um, but there's like two-pronged approach to this. Our data pipeline that we're filling up our our our lakehouse, our data bricks lakehouse is actually leveraging generative AI to build that pipe and to implement that pipeline.
6:12 GenAI
On the flip side though, and we're early days at this part, is leveraging Gen AI for the end use, for the analytics. So once you have that built, you can then leverage it really for those business insights um and really get our consumers what they need without having to go do that themselves. Many places I know we all have it. People are still chopping away at Excel, they're bringing together, stitching it together. Uh best case, they get it right, but it takes a lot of work.
Worst case is they they do their best, but it's not correct information. And so the goal of this leveraging that pipeline to build the lakehouse and then leveraging genai at the at the tail end is that you are consistent it is trusted and it's common. So whether your use case is for you know regulatory reporting or if it's for onboarding a new customer that same information should be there in the right form that you need.
So this kind of shows you where we are right now uh in our journey. I think a key point this is truly a global bank. We were brought up a bit regionally in the global banking unit. So again that opportunity to cover a very complex set of products. We have as you can see all the CIB type products are in there retail uh all the you know global markets sales trading derivatives really complex landscape and in many ways just like a lot of companies was brought up in different entities and now they're coming together as a holding company in the Americas.
7:21 Global Bank
We have the same thing happening in other regions. So that complexity is really what should concern us, but it's also the opportunity. The left side you see that's not a full list of our systems of record, but these are the critical systems of record that can really help start driving the specific use cases that we're going for. And on the left you see there's various formats. We have, you know, everything and the kitchen sink just like I think every bank out there.
So the goal is to really get rid of that friction, make this as straight through as possible to that end consumer, whatever it might be. And again, going back to the analytics, when we really get good, then we can really start leveraging that for insights instead of working so hard at the basics to get that information into people's hands. So this is what we're actually working on now. So if you think the end use, there's so many out there, but for us, we're doing a riskbased approach essentially. uh BCBS239 anyone it's been around a long time I think everyone is challenged every bank is still challenged to fully comply with that across the board and we take a lot of work it takes a lot of work to do so so we're using our risk team as a key partner and that again is a global program which is I think the way we have to have to look at this uh it's a cross region and with our head office I think not surprisingly risk becomes the next you know next level of detail, reg reporting, areas like that.
8:37 Riskbased approach
And that same sources, those same sources though, once we get them into the pipe, once we get them into the lakehouse, can start being used for other things. And we're act, this is real now. We have the platform in place. I think the MVP, we'll go into a little bit more detail as of last January, bringing uh bringing information in and we're now consuming it for things like counterparty credit risk and it's becoming trusted by our users.
The interesting thing is how do you get gain that trust for people that are used to throwing armies against it that have to reconcile that have to chop through it to make them comfortable. So our goal long term I think they're still you know early days they're still going to do some parallel but once you get them comfortable they'll be able to really essentially use that information and do the analytics do the make the decisions based on it rather than spend a lot of their time just putting it together.
And you can see on the treasury side that's really around like our liquidity management. And so these use cases are really core to just what the bank does. So uh the right side we're not there yet. So kind of the art of possible I guess is the way way we would describe it. Um and again the more we can get bu into that lakehouse from the sources the more we can start going after those art of the possible. And that's really what if you talk to the CXOs and of bank they want that.
They think it's easy, right? And as those in tech like us, we we know that's like not the case. And we usually get a nice blame for things when they don't work. Well, in this case, we want to go at it so that we can make it very easy for them to get that quantitative data or unstructured data that they need to make decisions today. Again, it's an efficiency play that was on the prior page. uh it's people but taking those people and let them use the mind share for things that are actually going to really impact not only our compliance but that upper left on the other page actually grow the bank.
So I'm going to turn it over to my partner here in uh in this effort uh Ancho who's going to go a little bit more depth on some of the stack and then we'll have some time I think for some questions at the end of this right? Yeah I think so. All right well over to you handle that better than I did. I'm going to try. All right so as mentioned I'll quickly introduce myself. I'm with Deote. Um, thank you for your partnership.
We've been working closely with SMBC for the last couple of years on all things AI and data. Uh, but specifically with the lakehouse, we've really begun that in earnest at the early portion of this year. Um, so what has covered so far is what is SMBC as a whole? What's your vision? What are your ambitions as they relate to AI and data? What are your priority use cases? I'm going to jump into a little bit more of the how and the details of it.
Um, we'll cover things such as the architecture we have. We'll cover things such as the approach, so on and so forth as well. Um, and it's going to be fun and of course we'll leave some time for Q&A at the end. So, let's start off with the with the architecture. So, those of you who are familiar with data bricks in general, this should look fairly familiar. Um, it's a pretty standard uh lakehouse medallion architecture.
12:08 Architecture
If I draw your attention to sort of that middle of that page, um you have a bronze layer which is pretty much the raw data which has been technically standardized. You have a silver layer where at SMBC we are following a domain based construct. Domains being commercial loans, retail loans, capital market products etc. uh which tend to be quite different in nature but also treasury finance so on and so forth. Uh and then the gold layer which tends to be much more focused on the actual consumption and the use cases itself.
Um in terms of the architecture as I mentioned it's standard but but let me talk about a few priorities that are very important for SMBC as a whole that we try to service through this particular architecture. One of the things is is speed. Um and as we mentioned we want to get rid of the silos and go with somewhat of a unified data architecture in the middle. But traditionally the problem that you run into doing this is you run into a bottleneck issue um where the speed of delivery tends to slow down quite a bit.
So we're trying to address that very very proactively. So even though we have a unified architecture in the middle, we are federating out the ownership and the responsibility and the ability to bring in the data into the bronze layer and we're also federating out that responsibility from a gold layer perspective for people to be able to make their own joins, make their own data products if you may and consume it themselves.
Um, and that's a that's a journey that we are on and that's important. But you cannot have federation without some rules in place otherwise this this lakehouse will become a swamp. Um so the other thing that we're really doing is setting certain standards in place giving the user certain tools um whether it's standardized ingestion tools etc. Uh having certain policies in place etc in terms of how do you bring in the data how do you combine the data so on and so forth so speed is is very important to us I mean we'll get into details but the goal is the 25 sos that talked about earlier we're going to land all of that data within a year.
Um so that's that's something that's key. The second thing that I want to point to is the delta sharing at the bottom right I guess. Um and talked about the structure of SMBC which is global and also the different banks that reside within it. Manubank, Niko, so on and so forth. And each of these things are quite different right a retail bank operation is very different from commercial banking which is very different from capital markets.
So traditionally each one of these have had their own data warehouses if you may. And for any place where you need to combine data, you're ping ponging uh files back and forth between these different banks. Um with delta sharing that problem is taken care of. Um where you can have again a unified architecture and the data that needs to be shared between these different entities and banks, they can be shared. So that's something that's very important.
Uh and the other thing that Gorder touched upon is governance. Um so you'll see uni unity catalog at the bottom, Calibbra as a system of record for all metadata at the bank. Um while we are building this platform, we're building it as governed by design. So typically when you talk about data governance, it's an afterthought. You build all your data pipelines and you build governance separately. We are absolutely not following that approach here.
Um and as we build this lakehouse, all the data that's coming in is going to be governed by design itself. Okay. So just building on that, I'll go through this relatively quickly. um some of the features that we're utilizing from data bricks currently and then on the next slide I'll cover features that are on the road map. So, Unity catalog extremely important. Building on my point previously, which was we're building everything that's governed by design.
15:35 Features
Now, you cannot build in data bricks without the Unity catalog which is which is great. Um, and the lineage is automatically available. Access control is automatically available out of the box and that is actually being utilized. The other thing to keep in mind is data bricks is not the only platform as it exists today within the bank data platform. there are legacy on-prem ones that are going to continue to exist and there are competing technologies in in various places as well.
So the actual system of record for the metadata is Calibbra and one of the things that we are doing is also using API mechanisms to build connectivity between the Unity catalog and Calibbra as well. So that's another thing to keep in mind. Delta sharing I I touched upon u extremely important very relevant use case for this particular bank. Um the other important thing here is it's global in nature. So while we have enabled a lot of delta sharing within the US and the Americas itself what's coming next next is also sharing across geographies which comes with some implications around localization of data so on and so forth which we are working through but the ambition is that it will also be enabled globally as well.
Um workflows is is tightly embedded with everything that comes with data bricks. Um we have a variety of jobs that need to be set up. Um we have CI CI/CD pipelines to promote code so on and so forth. Um we have jobs that are tend to be end of day, end of week, end of month. But we also have a lot of micro batches which run several times during a day uh for example. So workflows is is very well embedded and integrated integrated with everything data bricks has.
So a key portion of what we're doing what's on the road map marketplace like every other bank I guess worth its salt marketplace is something that's important. Think of it as an amazon. com experience for data scientists data analysts to come and shop for the data. This data could be both internal and external. Um data bricks of course has its offering around marketplace and there are some some other offerings being being evaluated as well at this point of time.
17:38 Roadmap
Um same thing with the other two things Mosaic AI um the intention here is once the data lakehouse has been filled um different different data scientists can come in and be able to use their LLM of choice of course within a governance structure to be able to run the analytics not just LLMs their open source models that are out there etc and being able to access those um Genie one of the key goals ultimately within the next sort of 12 to 18 months is to reduce the reliance on pure engineering to be able to get your business analytics.
So can a business analyst a business user can they go in and can can the data talk to them? Can they talk to the data converse back in simple English and again that's one of the things that's being considered there with Genie but other tools such as PowerBI etc are also being considered and evaluated in this. Okay, let me talk about a little bit about how we're going about this this implementation. We don't have the luxury of starting from scratch.
18:54 Implementation
I think that's the common theme that exists with most companies who are speaking on here especially a bank as storied and um as as SMBC um so a lot of data already exists there are lots of data platforms that exist and while we're trying to do that and switch out sort of the guts of the data infrastructure that is not a trivial problem to solve um there is a lot of skepticism that needs to be overcome from business users Gorder you touched upon that earlier as well you need to convince them to come on this journey with you.
Uh and you also need to maintain existing processes that exist on a day-to-day basis while sort of switching out what what relies underneath. Right? So again the way we started off this should this should not be a surprise. We started off with a blueprinting phase where again depending on the different priorities that we we collected from different business groups and technology groups within the bank, we figured out what are the appropriate tools and technologies to use.
So that was done in the blueprinting phase. Then we went to the MVP phase which is very critical. This is where our attempt was made to bring on certain skeptical stakeholders on board. So we ran an 8week MVP exercise towards the end of last year where we picked a use case related small use case targeted use case related to compliance where we ran the fronttoback use case through all the medallion architecture and provided it to the users and this was quite helpful.
There were performance tests done as well around these things. Uh and we brought on these business stakeholders on the journey because ultimately data is in the service of business. Technology is in the service of business and if you can't solve those needs, the platform is not going to exist for a long time. And now what we have embarked on is this incremental delivery um which is iterative in nature. Um and the challenge that we're trying to solve for along the way is we are making these changes but we want to make sure that the downstream consumers of this data are not impacted by changes that are taking upstream of them.
Um the other thing we need to make sure of is as new use cases are coming up and they're coming up every day in usea new use cases we don't take on tech debt. This is done through two horizontals. There's a technical component to this which involves architecture and engineering. Um and then there's a data management component to it as well. So we are also maturing the operating model as we go along on this journey. So as new data domains are being onboarded identifying who the data owners are for those making sure the federation of those responsibilities are taking place appropriately so on and so forth.
So it's a it's a tricky journey. Um as I said earlier within one year itself the aim is to have all the structured data in the bank or at least the relevant structured data in the bank available in the bronze layer and for the silver layer and the gold layer that's going to continue um over in 2026 and potentially into 2027 as well. Okay. So one of the things I do want to mention is how are we accelerating this journey and talked about the fact that we're using things such as Gen AI to make sure that these things happen at a fast pace.
21:57 Accelerators
If these things occur at a glacial pace again adoption is going to lose momentum. So we want to make sure um that this happens quickly. One of the things we're deploying here and these are some deoid tools that that exist in the marketplace. It's a service called data as a service for banking. Um, and there are a few things in here that I want to quickly highlight. From a banking perspective, as I mentioned, there are certain products which are deposits, retail lending, commercial lending, capital markets products, so on and so forth.
Um, over years, there is a standardized attribute listing that's available for this. This serves as the basis for the silver layer that we talked about earlier, which is a domain based construct. Then there is this concept of smart map. Um one of the things those of you who have worked in in in the banking sector will know that between the system of record which is where the data originates um and your data warehouse data lakehouse whatever might be the case there are series of transformations that need to occur.
Now one of the things is there are some fairly standard core systems and system of records that are out there such as FIS Piserve so on and so forth and what smart map does is it provides out of the box mapping between that attribute listing as well as these various different systems of record. So again that's an accelerator for this journey. What this also does is it captures the transformation between the system of record and that and the target uh in in in in plain English language.
A few things at the bottom and this is where Genai really comes in. Um there is this concept of an automated EDL engine. We talked about standardizing the ingestion process so that it can be federated out of the team out to different teams. The automated ETL engine is how we do that. There is this concept of an intelligent business glossery. Generative AI is is helping uh make data governance a lot easier. Um so instead of having an army of folks who who write business data definitions, you can use Genai for that.
Same thing with data quality. So the net result of that is much faster time to market, much higher data quality coverage, which for a regulated industry such as uh such as banking is extremely important and just greater reliability once you do go live. Um, so all of these things are again helping build credibility as we build this platform, bring on the skeptics on board and ultimately lead to a successful delivery. So I think that covers the main things that we wanted to cover.
Goro, do you want to take what I would say there's a few key takeaways here, but first I'd want to say this, this data problem, I'm going to call it that, is actually a leadership challenge. It's a leadership ownership challenge across the board. And I want to quote again because he said it so insightfully. He said the business is technology and the technology is the business. They have to know each other inherently.
24:36 Key takeaways
And so winning over the hearts and minds of the consumers of this information. If we don't win that battle early, we can't be successful. You cannot. You could build all the tools you want all of this flow until you get them very comfortable with what we're doing. So our real goal here is if if you look at it is how do we get that govern platform? Well govern does a lot of things to different people. Govern means we have all the controls.
We can identify lineage. We can get all that stuff done. That should become table stakes for us. It should be something that everyone knows they can push a button and get it. They want you know there's always everyone just wants things to work. This is a really hard thing to make just work. So the engineering behind it is critical. We have a lot of people working on it and deoid as a partner but it's going to take time but we have to start winning over those hearts and minds early and we've done that early days but we have to continue to do that and if we don't then we start falling back so to me it's leadership it's leadership by tech teams all the way down to the developer the people doing any code and it's leadership all the way to the top of the business our CEO we have to be aligned and without that you can't be successful the time to market of course we want to you know that's our goal um the innov innovation piece is kind of critical.
We always talk about innovation. I think every bank does some invest massive amounts of uh of resources into it. The goal is that we are able to leverage all those innovative tools in a much more simple way from that end end user and you mentioned a lot of the tools that you're you know are coming like marketplace. Imagine a day where you go in as a business user and just type something in and it actually comes up with your analytics just based on plain language.
I mean, that's where we want to go and you can get there. We can get there, but again, it's the data at the source that's going to be key. If we get that wrong, it'll all be wrong. It'll look nice, but it won't it won't be correct. So, that's why that govern platform is key. And then, you know, obviously, we want to, you know, partnering with Deote, we need these accelerators. This is not a problem we can wait on. The drive for the business is incredibly strong to try and grow, to try and be global.
Uh, you know, we talked about trying to share across regions. We can't we can't do that effectively unless we get this right. So, you know, no pressure, but we feel it right. We feel that pressure from our business stakeholders. There's big investment in this. I'm sure you all feel probably the same way if you're in a similar situation. And uh we we want to be successful, not only for us, but for our business and our clients.
So, I'll close it there unless uh you have any further thoughts, an No, I think winning the hearts and minds is a is a great message. Ultimately, technology and data is in the service of business outcomes and business goals and without that we're not going to go anywhere. Yeah, hearts and minds are great though. Once you get a couple people, you have the biggest supporters and they will push for more investment and investment always mean dollars, investment in changing processes.
Uh those legacy platforms usually come with legacy processes that people don't like changing. And when you get them some new way of working then and it and it really does work and they can prove it then you can change things and they get efficient. It's not just about the tech and the and the application itself. So the hearts and minds that's what that's what we're going for. That's right. All right.
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.