Blueprint to Breakthrough: Activating Banking Analytics, AI, and Agentic Use Cases

Summary

First Horizon Bank, an $83 billion regional bank based in Memphis, spent 2025 building a unified data foundation and is now activating use cases through a governed Data Marketplace and agentic AI capabilities on the Databricks Data and AI platform.
Databricks Genie is being used to accelerate development of business-ready analytics solutions through conversational and developer-enabled experiences on top of the governed gold data layer.
The bank's roadmap through 2027 includes embedding advanced analytics and AI—including multi-agent systems—into decisioning and workflows with strong governance, security, and controls.

Blueprint to Breakthrough: Activating Banking Analytics, AI, and Agentic Use Cases

this video continues the story of how a Regional Bank is modernizing its AI and analytics journey, evolving from building the foundation to activating value through measurable business outcomes. Powered by Databricks and Deloitte’s industry experience, the Enterprise Data and Analytics Hub is a one-stop shop for line-of-business and enterprise consumption, anchored by a governed Data Marketplace that simplifies discovery and reuse of trusted data products. As part of this evolution, Databricks serves as the knowledge base for enterprise agentic and AI capabilities, and Databricks Genie is being leveraged to accelerate development and delivery of business-ready solutions through conversational and developer-enabled experiences. The session will highlight how the bank is operationalizing these capabilities through a prioritized pipeline of high-impact use cases by accelerating speed to insight, improving consistency across reporting and analytics, and realizing value at scale. Discover how the bank is embedding advanced analytics and AI, including agentic and multi-agent systems, into decisioning and workflows with strong governance, security, and controls.
Talk By: Mohan Sankararaman, Chief Information Officer, First Horizon Bank ; Nikhil Kaushik, Senior Manager - AI & Engineering, Deloitte ;

Chapters

FAQs

Who is First Horizon Bank and what are they building on Databricks?

First Horizon is an $83 billion regional bank based in Memphis, Tennessee, with over 160 years of history. They partnered with Deloitte and Databricks to build a unified data foundation in 2025 and are now activating advanced analytics and agentic AI use cases in 2026.

What is the governed data marketplace at First Horizon Bank?

The Enterprise Data and Analytics Hub includes a governed Data Marketplace that simplifies discovery and reuse of trusted data products for line-of-business users. It features an access request and stewardship workflow so teams can find and consume data without bypassing governance controls.

How is Databricks Genie being used at First Horizon Bank?

Genie is being used to accelerate the development and delivery of business-ready analytics solutions through natural language, conversational interfaces. It allows non-technical business users to query the governed gold data layer without requiring SQL expertise.

What is the trajectory for First Horizon Bank's AI program through 2027?

The bank plans to embed advanced analytics and AI—including agentic and multi-agent systems—into decisioning workflows while maintaining strong governance, security, and controls. The goal is an AI-first banking analytics capability, with Databricks Genie Ontology giving agents business context.

Full transcript

[00:07] Good morning everyone. My name is Mo Hayne. I'm the CIO at First Horizon Bank. I'm going to tell you more about who we are. Uh joining me today, you also have Sriram and Nikhil here. Uh they are from Deloitte and we're going to walk you through the journey we
[00:19] have been on, our data journey, along with uh Deloitte and certainly with uh Databricks as well. You can move these slides.
[00:32] There are four things I want to walk you through today. The first one is what have we done? Where did we start? I was here last year. I shared our journey uh in 2025. I'm going to now catch you up on uh what we have done in the last 12 months and more importantly uh where we
[00:46] are headed uh with a uh focus on what are we doing with AI, the marketplace, and more. And then finally, I'm going to uh wrap up the discussion with what have we done? What's real? Uh show you some metrics, and then show
[00:59] you the trajectory towards 2027. So, we're a regional bank uh based in Memphis, Tennessee. We're about 83
[01:11] billion in asset size. We've been around since the 1860s, more than uh 160 years in existence. Uh we we use this phrase. This is the tagline we use inside the bank, the one
[01:24] that you see at the bottom. Big bank muscle and small bank hustle. So, we focus on uh providing capabilities, providing products and services that the super regionals have. But inside the bank, we operate like a
[01:36] small community bank. Easy decision-making, the agility, the nimbleness. And that's how uh our technology agenda is also influenced inside the company. So, in 2025, we started our data
[01:49] journey. We started building the foundation. Our landscape was no different from any institution you probably are very familiar with, very fragmented data uh warehouses, repositories, whatever you want to call them.
[02:01] We started building a unified data repository in 2025, and then 2026 is all about activating use cases and realizing our value from our investment. Next slide, please.
[02:13] This is a good visual representation of where we were prior to 2025, where we are today, and where we are headed. So, prior to 2025, as I said, a very siloed uh organization, right? Repositories
[02:27] around the company. Uh we couldn't uh gain any efficiencies. And if you were in the analyst team, you had to spend a lot of time uh sourcing data, curating data, fixing quality. And you probably didn't have uh very
[02:40] much time to go do what you really wanted to do, which was to run the analytics, to provide the insights to the leadership, right? Because you were spending so much data uh time on sourcing, curation, quality fixing, and so on. We wanted to fix that. So, 2025,
[02:55] we started building what we call the enterprise data hub on the Databricks platform. And I'm going to show you what the architecture also looks like. In 2026, now we're more focused on use case activation, right? So, that we can
[03:08] start realizing value out of our investments. What use cases do we have? What role is AI playing? What are we doing with agent care? All those together is what our journey in 2026 is
[03:20] all about. And then more in 2027 as the evolution of AI uh continues, as the evolution of the platform Databricks platform continues, we're going to more move into AI first,
[03:32] whether it's data sourcing, and you know, the work we do on the platform is going to become more and more AI first. That's our agenda for 2027. Next slide, please.
[03:44] This is the architecture we have been building, will continue to evolve, right? In the middle, what you see is the classic medallion architecture. Uh Um, you can see there's the bronze layer, the silver layer, the gold layer. I'm not going to bore you with all that. It's a very well-known industry
[03:57] reference architecture. What I will tell you is we have also now added evolved this architecture to add what's called the platinum layer. So, if you think about reg reporting, if you want another more locked-down layer, we have added
[04:09] that. So, the real consumption of this data comes from the gold layer. That's the uh data sets that the analyst and others inside the company need. But, to arrive there, you travel through the bronze and
[04:21] silver layers so that data is curated, data is governed, uh any quality issues can be fixed so that you can move to the right and make a good uh use of the data set. The marketplace is where everything comes
[04:34] together. So, if you're an analyst in the company, you're an executive in the company, you're a leader in the company, how do you get to this golden layer data sets that we have built? That's where the marketplace comes in. So,
[04:46] think of uh the simple example I can think of is like an app store, right? So, this is a data app store where we have listed all the data sets that are available to you, right? There's documentation behind it, the metadata is
[04:59] described, the glossary is there, the lineage is there. Everything you need to know about it is there in one place. So, you can cruise through the uh portal for lack of a better term, pick the ones that you want. And if there's something
[05:11] that meets your needs, then you make a request for access. Of course, then that has to go through a process. Not everyone can get to every data set. So, you have to make a request, and if there's a good valid reason, access will be granted to you. And all that is
[05:24] integrated into this, and you're going to see a demo of this uh later in the session as well. On the left-hand side is where you have the producers. These are system of records. Think of in a bank, the deposits, the loan platforms, and more.
[05:37] That's where the data starts feeding into the enterprise data hub in the middle. On to the right side, you have the consumers of this data. It could be the risk teams inside the organization, could be reg reporting people, could be
[05:50] analytical people that are helping the business leaders draw insights and conclusions and much more. Then in the bottom is what the AI layer is that we're building. And then as the data platform from Databricks also
[06:03] evolves and they roll out more AI capability, we'll adopt and merge these, but we have also stood up our Agent-ic AI platform that can sit on top of this data and make good use of
[06:15] the data foundation we have built. So, if you think about the ontology we need, those are the types of things now we're thinking and starting to build out. At the bottom is the governance layer that's naturally provided by the Databricks platform itself so that
[06:30] when this is fully operational, we already have all the We have more than I'll show you this, too. We have more than 35 data sources that are onboarded onto this platform. The quality is there. So, now our analysts inside the company don't have to spend time
[06:43] sourcing, curating data. They can focus more of their energy on drawing insights and helping our leadership make their own decisions and conclusions.
[06:55] So, this is a good 2-second overview of the architecture we're trying to implement. We started this in '25. We're continuing this. It'll continue to evolve and grow into '27. Next slide, please.
[07:11] The marketplace. I like to draw If there's one thing I like for you to walk away from this session today is it's the marketplace. Who are we making all this wonderful work available to people inside the company that need this? So, if you look at the number two, the
[07:23] personas, those are the real people that are going to be using this. The business executives that need access. With the evolution of AI, maybe they can talk to their data. Maybe they don't need pre-canned dashboards, right? The second
[07:36] one is the business managers, people that have to make day-to-day decisions, operational dashboards, insights into the data. The last one is the reporting analyst that are in the firm that are helping some of these other leaders
[07:48] inside the firm, right? These are the three major personas that I can think of and there's many more, but I thought I'll highlight the top three for you. On the left-hand side is the
[08:00] the data sets that are coming out of the golden layer that people are going to be using. These personas are going to be consuming data from the marketplace that's coming out of the golden layer. And then at the bottom is the patterns
[08:12] of consumption. If I'm in a business executive, maybe I want to talk to my data. If I'm a reporting analyst, I need direct access to the data so that I can build my dashboards and visualize the data for my leaders. If I'm an operating
[08:25] manager, I need ready-made pre-canned dashboards that are built or stood up on this platform. So, you can see all these the the personas, the patterns, and the marketplace have to come
[08:37] together to deliver on our vision, and that's what we're zeroing in right now. So, with this, I'd like to pause here for a minute, invite my partners here to show you what a real marketplace looks like, what have we done, how does it
[08:49] feel, what's the value of it, and then I'll come back to wrap the session up to show you real metrics on what have we really done, what our pipeline looks like, what have we sourced, what have we stood up, and then what do we think about 2027.
[09:02] >> Thanks, Mohan. Am I audible? Okay. So, I mean, we stood on the same stage a year before when, you know, this was a vision that was a vision that Mohan set
[09:15] set for us saying, "Let's just go stand up this data platform, set up a marketplace where all the user personas can come and consume this data set." That was the vision that was set for us last year,
[09:27] which we achieved, you know, since we spoke on the same stage a year year before. Now, what we have stood up in the marketplace is the the foremost thing what Mont spoke about
[09:40] is a data data product, right? I mean, the gold layer, you know, is where everything comes together where what we have done to the left side of it, you know, the benefit of that everything comes together in the gold layer as a data product. Now,
[09:53] when we started doing this right from the word get-go, there are few things that we made sure that okay, these are non-negotiables as we are setting up our data products. Now, because you cannot make these things as an afterthought.
[10:05] Setting up our data contract. So, what I mean by setting up a data contract is do we understand where the data is coming from? Do we know who owns the data? Do we know at what frequency the data gets refreshed? What data quality controls
[10:17] that need to be put in place for the data set that we are curating? If someone a business user or an analyst or an executive wants to understand how this particular data set has been derived. Do they have an end-to-end lineage they can walk back
[10:31] at not just at the entity level, but at an attribute level. Can I walk it back to clearly get the confidence that okay, now I understand how my data product is being derived, I understand the attributes that being involved, I understand the calculations that being
[10:43] involved. All of that is in a day in a in a think of like as a user-friendly manner. Can we provide all of that in a market place? So, that was was like the non-negotiables that we know that we set
[10:55] for ourselves when we set up the market place. So, it just doesn't become an afterthought saying oh, we curated all this data, we collected all this data. Now, let's just start applying the governance layer. Versus like right from the word get-go, let's make sure that this is the vision for our market place.
[11:08] And that's how we ended up doing it and we'll show the demo shortly. Now, what that has helped us to achieve and that will also help us to going forward. I'm not sure a lot of us would have seen the keynote. The Genie ontology and and some of the
[11:20] Genie demos, you know, that we already started incorporating into the market place, you know, my colleague Nikhil will show in shortly. But this foundation of having the glossary, the data quality controls, the data contracts
[11:33] that is now will help us to set up the context for, you know, Genie ontology. That will help us to even accelerate this even further. What was not possible think of like 10 months before, you know, before these agents have even
[11:45] become even more smarter, now we have the complete foundation of having this entire marketplace with all the corresponding building blocks scaffolding available for the agents. Then to Mohan's point, the vision for
[11:57] 2027 and going forward from this point onward which is how do we now establish an ontology ontology at the right side? How do we make sure that these agents and now can harness this information that we have set up and even provide insights much
[12:09] faster. I mean, I know all of you would have heard it, you know, the current era of AI, if somebody is writing code manually, that is no longer acceptable. Similarly,
[12:21] if we as an organization we're going to have a set of dashboards and visualizations that are being produced manually, I mean, that is just not a good use of that particular, you know, resources productive productive
[12:34] time. Rather, you want the agents to do the heavy lifting. You want your business community to focus their energy on what new products they can serve to the customer. How can they harness this data to take better better decisions
[12:46] rather focusing their time in creating visualizations and dashboards because now those things can easily be automated using these agents sitting on the right side. So, that's that's
[12:58] journey is like now enabling these agents to the right side and some of this is already enabled, you know, as we go through in our data demos as we as we are talking through. So, you have to switch to them.
[13:14] So, while he is pulling it up, the other vision that we are setting ourselves, you know, in the in the coming months, you also heard about this, you know, zero ops, you know, during the, you know, keynote. Now, we have a complete end-to-end data platform up and running.
[13:26] And Mohan spoke about the multi-agent framework in the architecture on the right side. That's also up and running. So, we have the complete end-to-end data platform. We have the multi-agent system running on top of it because we have all the ingredients we need.
[13:38] We are thinking of setting up the ontology layer to the right side so that we can get to the insights much faster. Now, to make this whole platform up and running, you know, you know, and then you have any data pipeline that breaks,
[13:50] any maintenance issue that happens, our next logical thing is to have agents ops running on top of this platform. That will That will allow us to, you know, uh help us go faster.
[14:04] Okay. So, this is the real marketplace that's on live, you know, um this is now up and running. We came here last last last time the same time. We provided a vision for this. And that time this is not up and running. Uh we demoed saying, "Okay, our vision is to
[14:16] produce this marketplace." Now, uh fast forward 10 months, you know, here we are. We have this marketplace and maybe we'll just click one of the data products. Like you Okay. The fundamental building blocks of this
[14:29] marketplace. Now, this is running on top of, you know, Delta Lake on Databricks. And you also might have heard about the data apps, you know, during the keynote, right? This is running on This is built on data apps. So, this is built on Databricks data
[14:41] apps. It's completely integrated with Databricks Unity Catalog. You have your um data lineage being pulled from uh um Unity Catalog. You have your data quality controls coming from there. You
[14:54] have this data contract that is now established. Every user that is being onboarded to the marketplace has a data contract. They understand where the data is coming from, who publishes the data set, what kind of data quality controls are being run on this data product so
[15:07] then they can trust the data they're looking at. I mean, this is a regulated industry, right? I mean, given a bank, you know, you understand the regulations that goes behind it. The lineage of this data set, if we can just show you the lineage. And this is all business friendly.
[15:20] Someone wants to click further, they can drill down to the every individual attribute. They can go through every single attribute of, you know, where it is coming from, who derived what's the logic behind it. Both the technical metadata and also the business metadata
[15:33] behind it. So, all being harnessed from the entire Databricks platform. Um Bringing to the play to the point where me as an end user, I have complete trust in what I'm looking at.
[15:45] I have complete information on on the data set that I'm looking at. And I have quality controls that are in place. And I can also read through the data contract what I'm adhering to when I'm consuming this data
[15:57] set because in a regulated industry, this becomes extremely critical because as you scale up a bank, as you scale up these data platforms, I mean, I'm sure a lot of us would have experienced this.
[16:09] You start realizing that, you know, the proliferation becomes, you know, unmanageable. You don't know who's producing the data set. You don't know what kind of sort of controls are you know, that exist. It all starts becoming an after thought and then it just starts becoming like, you
[16:22] know, all sorts of regulatory actions start coming in. You start spending time in trying to fix an issue rather than trying to have something productive on this data set and start using it for your business advantage.
[16:35] So, that's the difference of having this establishing these data contracts, establishing the data lineage, establishing data quality in place. All of that stuff come what make this marketplace, you know, a real thing versus I'm curating data sets. I'm just
[16:48] asking users to come and consume like. Now, coming to some of the agentic capabilities, right? I mean, um maybe Nikil, if you can talk to the talk to my data stuff. We started this like, you
[17:01] know, 6 months back even before Genie ontology was released, even before, you know, the Omni agent was released, which is now making us even even, you know, feel even more excited because now you have ontology Genie ontology. You
[17:13] have Omni agents that being harnessed is being released. We already have all the building blocks of, you know, the multi-agent framework. We have already set up the entire glossary, all the scaffolding that is needed. It's going to make us go even faster in
[17:26] enabling these things. And maybe Nikhil can just talk to the genie stuff. >> Am I audible? I didn't start. Am I audible enough? Enough. All right. This, taking over from from what Sriram
[17:38] said, the talk-to-my-data capability is what you're seeing here. Um if you open it, this goes on to DataBricks Genie. Um
[17:50] All right. So, this is what you typically see if you are the In this case, let's assume you are a business executive somewhere in the deposit line of business. Um you know, and you want to be able to see what are my Summarize my deposit activity and my transactions and give me
[18:03] a summary of my deposits, right? Um so, here's an example of a simple prompt. Please summarize deposit accounts and their status, right? So, what you see is
[18:15] a is a in a quite a literal sense literal sense, talk to my data, where you could see your accounts by status, how many are active, inactive, closed, dormant. You could see that in a bar chart form. You can also see your total
[18:27] principal balance by account status. Um and then you would also be able to see your account distribution type, right? How many are certificate of deposits, how many of these are standard checking accounts, how many are money
[18:39] market accounts, so on and so forth. So, I think this kind of illustrates to Sriram's point that all of the ecosystem which you saw earlier, the bronze layer, the silver, the gold,
[18:52] the governance aspect, all of that, the culmination of that, one of the consumption patterns is this talk-to-my-data. There's three key pieces I want to highlight before I turn it back. One is
[19:05] the LLM models have grown really powerful over the last few months. And the more context you're throwing at them, be it in terms of data, metadata sitting in Unity catalog, etc., the
[19:17] responses are more powerful, more accurate, and more faster. That is one. Second is data is only as good as the trust the business user has in the data. So, governance, having data contract set
[19:31] up right, having data quality set up right, knowing the lineage, having a business glossary to interpret the data, all of that is key for the trust and usability. And third is the access management, which we didn't talk
[19:43] about, but that is also something which is incorporated up up front, where marketplace is not in a silo, but ties into your broader security framework, which, you know, uh the bank has, given
[19:55] Sriram's point on how regulations and how regulated the the bank is. So, few key pieces, all of the culmination of that is this is an example of a simple one, but Sriram. >> Sure.
[20:07] So, the the next logical question is, you know, uh where do we go from here, right? So, last year when we were here, we set up the foundation, we set up the all the medallion layers, we set up the data products,
[20:19] we have set up the marketplace, the marketplace is up and running. We also have some genie capabilities already up and running, where I can talk to, you know, talk to my data kind of capabilities. Now, what hopefully by this you all will be
[20:31] here next year, and you know, what we'll do between now and next year is what we are embarking on now. So, this is where we are now taking a genetic first thinking, every single aspect of how we going to
[20:45] what we're doing at the bank, how we are either sourcing new data set, building new data products, or building new governance layer, building in more, you know, uh
[20:59] silver and gold data products and platinum data products and enabling use cases to the right side, building visualizations for my now know, my end user. Can all that be done as an agentic AI native way? So, that's our you know,
[21:13] that's a goal that you know, Mohan and Krishna and his team has given us for the next year. Can we have an agentic AI first thinking? Every single aspect, I mean, you know, think of like SDLC on data
[21:25] side, data development life cycle. Can every single aspect of data development life cycle, can it be agentified? Why should we be be writing you know, a mapping and social dollar and mappings and developing ETL pipelines in a manual way?
[21:38] Can I have my agent that has a complete context of my data platform that I've built? It can harness the context from my source system, from my you know, product manuals, from my operating manuals. Can it harness all that context? Can we
[21:51] have agents build these things faster for us? Then the you can apply the ontology to the right side. So, think of like this is tech for tech, right? So, Mohan's team is building this data platform, you know, they're now the vision is to use
[22:04] agents to build this faster. Then the ontology layer comes to the right side. Can every single business user in the bank can they have and think of like an insights agent at their disposal 24/7?
[22:18] So, that any visualization, any insights that they need to draw I don't have to go through put through an IT queue and then wait for 6 months to have that team's bandwidth available for me to build the insights for me. Rather,
[22:30] I can use that team's that team's size usually limited, right? I don't have infinite capacity in my bank who can keep creating dashboards and visualizations and you know, and they have very limited amount of resources that I have to use it for my advantage.
[22:42] Can that team become my SME team that becomes the brain of the bank that can come up with ideas and have my agents create visualizations and insights at scale for every single business executive out there.
[22:54] So, that is our vision for this year, agentifying the entire journey from left to right. So, that's what we have started. Hopefully by this time we all here next time, you know, we can demo that as at at scale. So, with that I'll
[23:07] just hand it over back to Mohan. >> Next slide, please. Here's where we are, right? On the left-hand side, you have some hard numbers that tell you what have we done, how many sources have we onboarded, what
[23:20] do we have in the pipeline, uh what can people do with the data and more. On the right-hand side is what uh we're now focused on. More agentic use case enablement, you heard from Sriram on how
[23:32] we're thinking. The more important thing for me is I'm no longer uh keen on data as a service. We've moved past that, right? Uh our view now is data as a product, so that there's a life cycle to it, there's an
[23:45] owner to it. Uh we can keep it alive all the time, we can revive it, we can refresh it. Uh so, data as a product is how we're starting to deploy these. So, the next thing is uh
[23:57] how can AI help me accelerate what I'm doing, how I'm building this uh platform, how fast the pipeline moves, so that uh we'll have human in the loop, but more automation using AI agents, so
[24:10] that we can fast-track the buildout of our vision. And the last thing is what I touched on is data products. Uh can we view all this as a product? Products typically have an owner, a product owner, they have a life cycle to it,
[24:23] they have investments into it. So, that's how uh we're starting to view this, so that we can build out our analytical uh capability. You saw a few demos on can I talk to my data? If
[24:35] appropriate, maybe we'll have some pre-canned uh dashboards and whatnot, but this is where we're headed as we move past 2026 into 2027.

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.