Sponsored by: Deloitte | Future-Proofing Finance: Modernizing Banking Data Platforms

Summary

Navy Federal Credit Union partnered with Deloitte to consolidate over 80 data applications into a unified Databricks Lakehouse platform with standardized ingestion patterns and Unity Catalog lineage-enabled governance.
The finance data product blueprint embeds ICFR compliance controls and data quality standards from the start, building the auditability and regulatory confidence required for financial reporting at a major credit union.
Agent Bricks is the next horizon in this video's roadmap—enabling agentic automation for the office of the CFO through multi-agent workflows that operate over the governed finance data foundation already in production.

Chapters

FAQs

What problem was Navy Federal Credit Union trying to solve with this modernization?

Navy Federal's finance organization had over 80 applications connected through point-to-point data pipelines with no central catalog, making it data-rich but insights-poor. The goal was to consolidate these fragmented systems into a unified Databricks Lakehouse with standardized ingestion patterns, lineage, and governance.

How does the finance data product blueprint address regulatory compliance?

The blueprint embeds ICFR (Internal Controls over Financial Reporting) compliance controls and data quality standards by design, rather than adding them after the fact. This governance-by-design approach gives finance leadership and auditors the transparency and auditability required for regulatory reporting at a major financial institution.

What was the first use case delivered on the new Databricks platform at Navy Federal?

The first production use case was loan securitization analytics on Databricks, delivered as an MVP to demonstrate immediate ROI and build buy-in for the broader modernization journey. This video also covers data lineage for loan origination and risk review workflows as early examples of governed finance data products.

What is the next step for AI-enabled finance operations at Navy Federal?

The next phase is deploying Agent Bricks to build agentic multi-agent workflows for the office of the CFO, operating over the governed finance data foundation now in production. This video also covers Genie access for finance business users as an intermediate step toward full agentic automation.

Full transcript

[00:08] Hi everyone. Thank you for coming today. Uh I'd like to introduce this is John and Tom and today they're going to be talking about future proofing finance. Thank you. >> All right. Awesome. Well, thank you all
[00:20] for coming. Uh appreciate you coming to listen to us. But just do some quick introductions. So my name is John While. I'm one of the finance AI and data leaders in Deote and uh you know been working with Tom and team for a while on
[00:33] our our journey around how we're building out the finance data product and Tom introduce. >> I'm Tom Hughes from Navy Federal Credit Union. I'm the senior vice president responsible for our foundational
[00:46] technologies and platforms which includes most of our data and AI engineering capabilities. So thank you for attending. >> Awesome. So we're going to talk through a few
[00:58] things today. I think one we'll start with what was the problem we were trying to solve. What are some of the challenges that the finance organization faced? We'll talk a bit about the finance data product and I think focusing more on some of the novel
[01:11] aspects of what we've brought in. How do we do it faster, better, differently? Uh and then the expedite value. What did they actually get out of it? How did we show that relatively quickly? And how did that drive a lot of the buyin to the
[01:23] journey we were taking? uh and then we'll as it says shift a bit talk about some of the barriers we are trying to break down around getting AI ready data for finance uh and then the future of
[01:35] our AI enabled finance operations but with that Tom I'll hand it to you if you want to talk about that the start >> sure um so our finance organization um pretty much covers all the full raft
[01:49] u of corporate finance even though we're a bank um it's traditional finance organization. We support the office of the CFO uh directly. Um we've been on a transformational journey in terms of
[02:02] data and I AI for about the past five or six years or so. Out of necessity, our finance team um was not participating in the initial components of our data
[02:16] transformation mostly because we were reverse engineering all of the anchor platforms that support finance. So we felt it was best to get that out of the way first. Um so now that we're turning our attention back to the finance team.
[02:30] Um it's an organization a classic example of an organization that is datari but insights poor. Um because there a multitude of systems that
[02:43] connect to support the finance team. I think there's well over 80 different applications that are in the portfolio. There's a tremendous amount of pointto-point manual data movement that
[02:57] that is going is going on. Um as well as um a fair amount of of data duplication and also data inconsistencies
[03:09] that they deal with on a daily basis. Um the there's also a lack of fundamental governance controls around data. Some of the systems in there are are fairly well
[03:22] um fairly robust in terms of data management controls, but others lack basic data stewardship capabilities. And more importantly, there is no really de
[03:34] facto standard catalog of information that the finance team can utilize as a digital storefront to interacting with with their data. Um and last but not least here um analytics is extremely
[03:49] um compromised. Um there are multiple silos or stovepipe analytic capabilities none of which provide the finance team and the office of the CFO with a concise
[04:02] holistic picture of how our company is performing to our strategic objectives. Uh so the data initiative that JJ JW is about to go into a little bit more is
[04:14] really about how we've addressed these three primary areas of concern for finance and moving to them towards a data product centric uh delivery model. So with that I'll let JW.
[04:42] such that they can have some consistency and accuracy and comfort and what the reporting meet all the needs of finance controls and ICFR and and all the different compliance components they needed. Um, but we wanted to think about
[04:54] as well like start with the end in mind. So one of the first things we did was this data product blueprint. And so when you think about this, this alone was kind of a shift for finance of they were very used to I need to go address a
[05:07] specific use case. I need to get a specific report. And they would go and figure out what are the 20 different places I can get data. Am I scraping from screens? Am I getting emails from other people in the org? and and how do
[05:20] we get them oriented around you need to own and be accountable for building and designing your data products. So that was one of the first places but also getting them to think about the needs across the finance organization. So
[05:32] whether it was you know controllership and FPNA and tax bringing all of that together to say what does this ultimately look like and this the key piece coming out of this as well is that allowed us to start figuring out what is
[05:45] conceptually the model that needs to look like the underlying data model to support this such that all these data products are going to be interoperable. They're going to be able to navigate this themselves think about enabling self-service capabilities. So this was
[05:58] one of the fundamental building blocks that was critical to this journey of laying out the the ideal end state of what we wanted to work towards. I think the second piece of this was you know again within finance there's such a
[06:11] critical reliance upon the governance around the data being able to meet the necessary controls and have confidence in whether it's accounted for results whether it's directional planning and projections and so thinking about this
[06:25] concept of governance by design in our approach and building that in to as we are building these data products it's not an afterthought it's not a retroactive let's go figure out what the data means and and how we're pulling it
[06:37] together. It's doing all the metadata definitions kind of in you know in time as we're doing discovery around the products. It's coming up with you know how does the data need to flow? How do we ensure we have the collective lineage
[06:50] captured in Unity catalog and also then thinking about you know what are the standards around data quality in the different types of rules we need both from a technical data quality perspective to ensure what's flowing into the data lake is pristine but even
[07:04] from a business data quality perspective to make sure we're enabling different use cases on the you know the output side and and
[07:17] we're learning. So this was one of the the novel aspects of how did we do this then with AI. And so instead of legacy ways of people going into source systems and you know trying to find old data
[07:30] dictionaries that someone had locked in their closet 20 years ago from mainframe you know we were able to basically use agents to go through scan through metadata look at the profile of data and come up with contextually aware informed
[07:43] definitions for how that was all defined through the process as well as then informing what are the different types of data quality rules we need what are the thresholds need to look like around ensuring ing that quality is is living
[07:56] up to the to standards expected as well. But so this was a key pl uh a key component of the approach that not only accelerated that effort but is is giving confidence to the finance team in the data that they're getting exposed to as
[08:09] well.
[08:23] know, defect fixes and and debugging to basically use those suggestions and help to refine some of the frameworks we're using and expedite that process further even as to how we're doing more frequent
[08:35] iterative production releases. so that every time we're moving through kind of a a program increment, if you will, we're able to push code to production, get new data to the finance team, and tied in with all the pieces we're
[08:47] running in this governance process, they have the confidence and awareness of of what exactly that looks like. And so, one of the the key pillars to success, I think, in this program was the fact that we were able to very quickly showcase
[09:01] value through this. And so one of the first things we did from understanding that blueprint was we said what are what's you know one of the core use cases that we can you know take from start to finish to deliver. And
[09:13] so for here there was a a use case around loan securization which you know in the banking world a key component of how how does the credit union unlock additional funding so they can go and issue more loans. And in a 12-week
[09:26] period, we were we were able to look at, you know, six different systems of record, 150 data attributes, bring all of this through the architecture, prove out the approach, prove out the foundations of the data product, how
[09:39] does that structure actually physicalizing, you know, instantiating the entities supporting that. And you know, this really did a couple things. Is one the platform itself proving that out and showcasing, yes, the
[09:52] architecture is working. We're able to do this in speed and time uh as expected with with the quality needed. And then secondly for finance, this is you know a 50 to$und00 million process per month in
[10:05] terms of that funding process. And so there's an immediate impact of you know moving the dial 2 to 5% on additional funding which directly flows back into the business. And you know Tom it would be great to get your perspective. How
[10:17] did you see, you know, the delivery of this this initial MVP kind of resonating across the organization? >> Uh, yeah. Um, I have to say that the our finance partners really dove in head
[10:30] first on this and they've been actually phenomenal partners with us throughout this. Um, I just want to touch on one thing that JW mentioned is this concept of accelerators.
[10:42] Um, we didn't start this, you know, from scratch. We had a series of accelerators, some of which were kind of industry standard data models that we were able to actually bring in to the
[10:55] design sessions with our business stakeholders that helped them establish a common business language around the different data components that we were designing and building for them um on
[11:07] their behalf. Uh in addition to that, there were also some analytics accelerators we took advantage of as well. So we started off with a little bit of an outofthe-box approach to some of these things and with the input from
[11:19] our business stakeholders we were able to tweak them and give them pretty much exactly what they were looking for. Um the enthusiasm um that they expressed throughout the process um was was
[11:32] phenomenal and as JW highlighted we were actually able to deliver a significant product in I would say a record amount of time and it was actually a we were
[11:44] actually able to demonstrate very clear and immediate value once we released this first critical MVP. Um, and it was wellreceived all the way up through the office of of the CFO. Um, and again, the
[11:57] model that we're now using is becoming a benchmark for how we're approaching this data product concept across the credit union. >> Awesome. Yeah. And I think you know the pillars
[12:10] of what we delivered in that like one of the capabilities when we rolled out some of the dashboards is is is now even evolving how people interact with that data and like talking to data and all those fun concepts. But I think one of
[12:22] the you know there's kind of a construct we've seen for the finance organization of how do we get them more comfortable with starting to adopt AI. And so kind of three key areas we've seen there. So
[12:35] the first being this concept of trust. Um, and so I think a lot of this, as Tom mentioned, starts with the the data product, making that understandable where they can look and they can see,
[12:48] you know, at a a data attribute, data element level, there's defined industry standard definitions, things that resonate with them. And that approach is something that helped them step into the shoes as a product owner and taking that
[13:01] accountability so that they were able to translate this and say, "Okay, I can see where this is going. I know how to stitch this together and I'm comfortable to say that I will actually own this and and define what what needs to be there.
[13:15] I think the second pillar here we've seen is around the validity and this is one of the one of the aspects of getting finance comfortable with using probabilistic models that are going to help you get towards a deterministic
[13:28] auditable result. And you know we've been doing machine learning for years. We have explanability and traceability and all these concepts. I think one of the things for them is showing them a bit of the journey of how can we use
[13:41] this though to accelerate the way that we are building but still get back to I can document this is auditable. I can create all the artifacts needed to you know prove out what we need in terms of audit in terms of the traceability of
[13:54] logic and
[14:06] alluded to, Tom alluded to, starting with how do we structure that model and thinking about what are the different ways to break that down and aligning that back to the business architecture to say if we're thinking about the
[14:18] lending portfolio, how do you originate a loan? How does it go through risk reviews, how does that get serviced? In bringing that to life for the finance organization in a way that those semantics are consistently defined, it
[14:32] aligns to how the data is going to flow and operate in the system and in the architecture. But then allowing that to be available now when we put AI on top of that such that that has the context awareness,
[14:46] it's actually designing and building things the way that the organization talks about it. And so when someone goes into a dashboard and says you know I want to look at this you know specific metric you know for a region I want to
[14:58] understand my gain on sale the you know the AI is well informed as to okay here are the 10 components of gain on sale as to how the organization defines that and it can appropriately reflect that uh in
[15:11] a way that resonates with the finance organization. So I think there's a few things here then that
[15:23] towards AI enabled finance operations and I'd say a a few key pillars here we're looking at so the first one is really around like some of extending what we're doing today around the lakehouse thinking about how we can
[15:36] further you
[15:49] at how do we bring in more agentic flow to say I can now have a pod that's probably covering, you know, 8 to 12 different data products in parallel. And so that's one of the things where we're able to see you know we can have
[16:02] significant efficiency increases and then you know coupling that with as well thinking about analytical uh the operational use cases on top of analytical and thinking about like OLTP type capabilities and lakebased and
[16:15] thinking about how that extends further to to enable that for finance. I think the the second piece here is around this concept of the digital storefront. So we've we've given bits and tastes of
[16:27] some of the capabilities that you know can be enabled around the AI insights but I think starting to think about how do you expose the holistic marketplace and as the data products expand and grow
[16:39] how do you make that more available and enabled via unity catalog for the the broader finance organization and even you know potentially enterprise users and then I think the journey with generative BI and thinking about Genie
[16:51] is one of the others where we already have finance user users who have access and are going in and starting to explore and building dashboards and and and you know starting to really you know bring that to life but I think thinking about
[17:04] how do you govern and orchestrate that and and really make sure that they're using it in the right way and having some reliability and consistency in how they're doing that. Uh and then you know lastly the intelligent finance workflows
[17:16] and really thinking about how do we start to take processes they run like loan document audits for example and
[17:29] exists in finance today and to really do a full you know process transformation type of mindset of how do we redesign and rebuild this with AI. Uh and with that I mean Tom it would be great for
[17:42] you to talk a bit about the broader kind of AI strategy and and how you all are approaching this as well. >> Sure. Um so um data bricks is a foundational platform at Navy Federal.
[17:55] It's an one of our anchor platform capabilities. Um and our AI strategy really embraces three key pieces. everyday AI which is AI for the masses, enterprise AI which is the proc code
[18:08] environment and the anchor platform AI as well. So within the finance community what we've what we've somewhat landed on the fact is that we've built this incredible foundation on top of data
[18:20] bricks building the lakehouse data product out for them. So it's become a natural opportunity for us to then continue to expand as we branch out into our AI journey to use the the native
[18:33] components that are within the within the datab bricks platform. One of the things that JW mentioned was the Unity catalog and how important that has become to this particular initiative. Um
[18:48] if you recall one of the challenges that we faced is that the finance team lacked solid governance controls and stewardship and and quality management controls. With the implementation of Unity catalog that is giving us the
[19:02] ability to capture all the data lineage, all the business rules around that data that we're bringing into our the data ecosystem. And then it's also giving us the ability to publish that information
[19:14] into our enterprise um data catalog, our our data navigator as we call it, so that we have a simplified business consumption layer around it as well. And that's how we're approaching this
[19:26] digital storefront. that digital storefront is actually now a going to be a key enabler to how we open up the generative and agentic AI capabilities um that we'll be on you know implementing over the next 6 to 12
[19:40] months for the finance group um and then you know building on agentive agent agent agentic AI and work and intelligent workflows um one of the
[19:53] things I highlighted early on is that there's all these pointto-point interfaces of manual data movements. So our objective over the next you know eight months will be to go back and take a look at those and start to using the
[20:06] data bricks agent bricks platform start building agentic AI capabilities within the ecosystem to essentially alleviate those pointto-oint making those processes
[20:19] agentic AI enabled. Um and then the third piece and JW touched on a little bit is this concept of operational data. Um while we're doing all this data
[20:32] transformation and finance transformation, we're also basically re-engineering our core banking platform as well which has you know significant um interactions intersections with our
[20:46] finance systems of record. So those those agentic AI capabilities will actually help us build those those platforms and capabilities out as well. >> Thank you very much. We really
[20:59] appreciate it. >> Thank you. 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.