From Demos to Deployments: How U.S. Bank Is Industrializing AI

0:00And most importantly, I think a business model that can turn these use cases into measurable outcomes. And that's kind of really what we've been focused on.

0:09Peter High: Welcome to Tech Innovation. I'm Peter High, President of Metastrategy. My guest today is Dilip Venkatachari. Dilip is the Senior Executive Vice President and Chief Information and Technology Officer of U.S. Bank Corp., the parent company of U.S. Bank and one of the largest financial institutions in the U.S. U.S. Bank Corp. generated nearly$43 billion in revenue in 2025, with businesses spanning consumer and business banking, payments, wealth management, and corporate and institutional banking. Dillip joined U.S. Bank in 2018 and leads a broad portfolio spanning the company's global technology infrastructure, cybersecurity, technology products and services, and digital and technology transformation.

0:46Peter High: During his tenure, he's helped modernize the bank's technology foundation while expanding automation, AI, and emerging technologies across the enterprise. Most recently, his work has included building agendic AI and digital assets infrastructure to improve how the bank operates and serves its clients. Dilip brings an unusual combination of experiences to the role. Prior to his current role, he held senior positions at Google, PayPal, and eBay, advised major financial institutions on digital transformation at McKinsey, and co-founded and led multiple technology companies. He also holds more than a dozen patents spanning analytics, financial transactions, risk management, mobile, and customer interaction technologies.

1:23Peter High: In today's conversation, we'll discuss U.S. Bank's technology transformation, the evolution of its technology foundation, how AI and automation are changing the way work gets done, and Dilip's perspective on translating rapidly advancing technologies into business and customer value. Dilip, welcome to Technovation. It's great to speak with you today. My pleasure. Thank you. Today's episode is brought to you by Tynes 3B. Organizations are pushing teams to move faster with AI, but that's creating a wave of AI-generated code and unsanctioned automations, leading to rising costs, security risks, and huge IT burdens.

1:56Peter High: Meet Tynes 3B, a single, secure environment that enables employees to build with AI while giving IT and security complete control to run and govern that work. 3B enables effortless building. Describe what you need in plain English, and 3B builds a fully functioning workflow. It uses IT-authorized connectors, meaning credentials stay protected, and non-technical teams can innovate without taking on unseen risk. Also, with full visibility across every workflow, IT gets total governance, while features like Auto-Tune and Auto-Fix ensure workflows improve and self-heal continuously over time. Employees get the freedom to build, while IT gets the control to govern.

2:33Peter High: No shadow AI, no data leaks. TINES 3B moves IT and security from blockers to enablers, giving teams the speed to innovate without compromising control. Visit the TINES website at tines.com, that's T-I-N-E-S dot com, to learn more and try 3B for yourself. And now on to the interview. Well, Delph, let's begin with the institution you help lead, U.S. Bank. Take a moment if you would and describe the business in greater detail if you would. Sure. U.S. Bank is a over 160-year-old bank. We are the fifth largest in the country, headquartered out of Minneapolis. And we have a kind of a very well-diversified kind of set of businesses ranging from retail to kind of wealth management, small business, institutional.

3:26What's very interesting about the bank in particular is in addition to our traditional banking businesses that we also have a kind of a fairly sizable payments business that kind of work, you know, among other things, also supports about a thousand other financial institutions with their payments capabilities on a private label basis. And we also have quite a sizable merchant payments capability as well. So based in Minneapolis, as I mentioned, and mostly focused in the U.S. and parts of Europe.

4:03Peter High: And describe your purview, if you would, as Chief Information Technology Officer of the organization. Sure. I'm responsible for all things technology. That basically means, you know, from a functional standpoint, it includes security. It includes all of the vertical chief information officer roles that support each of our lines of businesses. And then the horizontal technologies for enterprise architecture, AI data, our data centers, our cloud programs and the like. Broadly, the way I would think about this is basically enabling all of the kind of technology-based kind of services and offerings for our clients and for our employees and supporting kind of this very important aspect of transformation towards AI and initially digital and now AI as well.

4:55Peter High: And talk a bit about how your team is organized. You've described how the business is organized and some of the areas like AI and EA and cloud that are within your purview. Talk a bit about how the team then folds into the structure you've described. Sure. We are, as a bank, we are very functionally organized, which basically means we have, my peers are kind of the leaders of the lines of businesses. You know, one would be leading wealth management, one would be leading institutional consumer banking and so on. and all of the technology organizations are kind of, you know, part of my team. And so the way we are structured is we have chief information officers, CIOs for each of the vertical kind of lines of business, the CISO, which is kind of the security organization, and then what we call chief technology officer organizations, which are basically the horizontal technology.

5:47One of them is basically AI. and the AI function actually, interestingly enough, is frankly quite federated. There are portions of that that have to do with central technology enablement, things like developer tools and so on. And then we have moved AI to, and I can speak more about this later, to as close to the customer, to as close to the business as possible to enable much more rapid outcomes and delivery. And so that basically means each of our CIO teams also has AI capabilities and kind of going to work closely with the businesses.

6:25Peter High: Fascinating. Well, one of the key ingredients, certainly, Dilip, to effective artificial intelligence is data. I can only imagine across the various lines of business you mentioned, in addition, of course, to the payments business that you noted, supporting a thousand institutions, I believe you said, you must have an extraordinary amount of data and therefore potential insights you can derive from it. Can you talk a bit about how you organize your data to be able to derive better insights? Sure. As you pointed out, a bank has millions of customer interactions, thousands of employees, and huge, huge, huge amounts of data that support highly complex processes.

7:03And the data, interestingly enough, is kind of used for a variety of things, ranging from, frankly, online transaction processing. You know, when you have to see your bank balances, your transaction balances, the things of that nature kind of in real time, those tend to be kind of supported by systems that are, you know, basically designed for that high performance, high latency, and kind of immediacy of kind of customer transactions. Then there are a variety of kind of, you know, databases that support, you know, all of the AI foundations. The important thing with data and AI is, frankly, having a very good set of kind of, if you will, definitions and clarity on what the data is, uniformity across the different businesses.

7:49So when a data field means something, it means the same thing to every business and being able to have each of them available to use at the point when a customer transaction happens so that we can apply our models to them in real time. And so having both the ability to use the data, have it mean the same thing and have it be accessible in an AI standpoint is really kind of among the most critical of things.

8:13Peter High: Fascinating. And as you have developed this AI competency within the organization, I'd love to understand first and foremost, how have you thought about training your team, developing the skill sets of relevance, ensuring the technical and non-technical resources across the bank have a common, at least foundation associated with the topic such that they might derive value from it and do so responsibly. No, no, no, absolutely. The way we think about it, you know, as we set our priorities, maybe I'll kind of, you know, provide the backdrop, you know, the way we think about the first use case usually takes a village.

8:49It takes a lot of people coming in. The 10th use case should take an afternoon, right? If we haven't done that, we have not built a platform. We just have a bunch of good demos. And that is where, you know, all of the kind of the capabilities, kind of the both the training as well as kind of common platforms and kind of systems come into play. So scaling basically requires, so there's a huge difference. The way I would think about it is when we first started out, we were in a prove it phase, right? We were basically saying, you know, how does AI work? What were the kind of the components? Are we able to extract value?

9:26Essentially, making sure we as a bank are able to put all of our regulatory kind of guardrails and controls and still see kind of customer value coming out of it. Now we are moving from a prove it phase to a scale it phase, right? Which is kind of, we initially started proving that the hard part here isn't proving that AI works, I really think. The hard part is industrializing it. And how do we move from individual experiments towards something like a common technology foundation with strong governance, you know, reusable capabilities. and most importantly, I think, a business model that can turn these use cases into measurable outcomes.

10:06And that's kind of really what we've been focused on. And what's unique about, I would say, scaling in a bank is banks have to scale both innovation and control at the same time. This is quite unusual because security is very important to us as is privacy, model risk management, regulatory requirements. And all of these are, we don't see them as barriers to AI as much as these are our designed requirements, things that we will need to make sure that we always kind of apply. And so that's been kind of our fundamental issue. Then as we looked at it, what became clear is challenge for us from a technology standpoint are not the models.

10:51It's the plumbing. It's all the data. It's the identity integration of kind of the different components. observability, how do you make sure you know what things are kind of developing, and then the ability to deploy at scale. These are kind of the critical components to kind of how you actually get to value. So I don't know if that addresses your question fully, but to do all of this, we engaged in quite a bit of, we focused a lot on two things. Training, yes, of course, you need to kind of help people kind of enable new skills. But as much as training is important, it is essentially helping people feel more comfortable with the notion of learning continuously.

11:38Because what you learn about AI is certainly moving so fast, evolving so fast, that the tools that we thought about were the right ones, as recently as four months ago, have been superseded by something else. And so the answer is not someone who knows everything. The answer is having people who are comfortable, that they're open, that they want to learn and can constantly explore kind of question, you know, what they've, you know, how we do things so that we can kind of constantly improve. That has been kind of the critical part of our, you know, what we call our learning and development process.

12:12Took a lot of effort. And I think very, very pleased that, you know, our teams are, you know, adopted, embraced it, you know, quite, quite well.

12:20Peter High: Well, you highlight therein a conundrum, certainly, Dilip, which is the pace of change is such that after a few months, as you mentioned, the state of the art advances so tremendously. In years past, perhaps a training curriculum might be developed and remain relevant for years. Today, that may be less the case. So as you think about continuous learning and finding the training, as you say, to help enable the new skills, how does the curriculum and the training evolve to remain relevant given that moving state of the art? It's a very good question. As you pointed out, kind of the state of art, you know, as it may be, keeps changing very, very rapidly.

13:04and frankly it's very hard or probably even impossible to keep up training content at that pace. So the approach we've tried, and I believe this can have kind of a lot of impact over time, is basically a few things. There are some fundamentals. What are the first principles that you don't really change as much that we need to kind of train people on? That's kind of the basics. Second, to engender more openness to change, we are trying two very different things. First is making sure we have a lot of things like hackathons, a lot of hands-on internal events where people get a chance to try and test new things.

13:45And it's not what you do in a hackathon, but the fact that you're open to experimentation. it is that that kind of creates more space for learning. Along with that, we're also spending more time with some of our critical technology players, whether it's the LLM in the frontier model companies or the key strategic partners that we work with for various scaffolding and other areas. We have a lot more hands-on sessions with them because they're also adapting and learning very fast. And so the more we're essentially using those mechanisms as ways to kind of train our teams in that. Lastly, we have kind of a sort of champions.

14:31And then we have like kind of teams, train the trainer kind of rollout, cascading from that. And so those are some of the things we have done to kind of energize kind of the teams. And it's increasingly gone beyond just technology into business teams as well, because there's a lot of interest, a lot of eagerness there.

14:55Peter High: Earlier, you noted the importance of identifying reusability and what a fascinating topic that is. And I can imagine some challenges across an organization as vast as yours. And how do you think about the balance between allowing different lines of business to experiment and develop solutions relevant to their particular business versus those areas where the need identified perhaps in one line of business has broader applicability and therefore ought to be shared? It's a very good question. We are trying to avoid a thousand AI experiments beginning a thousand technology stacks. And the answer really is a common AI platform with reusable services, guardrails, and capabilities so that each new use case gets to be faster and cheaper to deploy.

15:40At a large bank, you don't want every team independently solving kind of the same problems around models, around data security, evaluation, governance, and so on. And so we are trying to create reusable capabilities. And the way we have done that is in a couple of ways. First is we have an internal AI marketplace where we let people kind of share things in so that everyone is aware of what the different pieces exist there. Second is we're internally providing social mechanisms to encourage reuse, not forcing it. because forcing it sometimes could have the opposite effect. You could use something that is not really as good as effective.

16:23So you want the innovation kind of stops. So we're trying to, in a sense, encourage people, almost like research papers and citations. You see yours being kind of cited. It means there are more people who find it useful. So you may actually go in and fix it a little bit and you say, look, there's some new things I discovered. So it gives you kind of encouragement, the subtle kind of social status kind of value, You're keeping on having your models, your tools rather reuse. So trying a lot of different techniques that are trying to ensure reuse, but do so without a heavy hand of controls to kind of limit its kind of innovation.

17:05Peter High: Dilip, we've talked about in the past the importance of change management. And certainly a lot of what we're talking about here is a necessity for change and an ongoing footing and hopefully an increased comfort with change as well. But talk a bit about change management as a topic that's a key ingredient in ensuring all that you've described is successful. You know, AI isn't replacing the importance of people, right? It's changing where people spend their time. It's very important for us to be able to articulate that, to be able to, in a manner in which people understand that to be kind of genuine and legitimate and kind of as an opportunity.

17:44We want our employees kind of spending less time on repetitive work, more time on judgment, more time on customer problems and on decisions that really matter. And so one of the hardest parts of AI transformation is changing how people work because that way they can experience it for themselves. That means putting AI into tools that people already use. We are doing a lot of things around leveraging Microsoft, its various tools with Teams and Outlook and all of the Copilot 365 that essentially broaden access across to all of the employees. so they can experiment and they can actually use it and kind of try things that work on their own.

18:27Plus, we are redesigning processes around what AI can do well and training people to use it effectively. And the important thing is being able to identify the metrics for you to be able to measure outcomes. And so because we believe the best AI system in the world doesn't create much value if nobody changes their workflow. And one aspect maybe just to kind of pick up on is software engineering, you know, where, you know, just across the world that's been, it has had the most impact. AI is fundamentally changing software engineering. You know, we're certainly moving from developers writing every line of code to developers increasingly directing, reviewing, and orchestrating AI-generated code.

19:13The opportunity isn't just productivity, it's changing the speed at which we can turn an idea into a product and improving quality and the reliability, which is kind of a critical aspect to our customers. And so part of the change management is encouraging kind of our teams to see the customer impact and kind of appreciate kind of what they're able to do, but they were not previously able to do with these new tools.

19:42Peter High: Yeah, fascinating. And Dilip, you've been in role for eight years now. So at the outset of your tenure, very different circumstance than what you've just described. The role of a software engineer was the traditional role, not the evolving one that you've so nicely articulated. And I wonder, you know, how do you think about the complexion of your team evolving as a result of some of these roles that, you know, some of the most sought after positions, in fact, having such consequential changes to them? How do you see that in terms of the size, shape, scale of the organization that you will continue to lead?

20:19I think technology is, as you can appreciate, going through a significant change. The interesting part was previously the discourse was about what does it mean for that number of jobs? And our organization is going to get materially smaller. It is true that AI increases productivity, efficiency directly. It is also kind of true, what AI has done is focus people's time less on the repetitive, but more on the judgment on the more critical things that people need to do. So what that has done is in some respects, as much as AI is creating efficiencies, businesses are eager to leverage AI to create business value.

21:05So the what of what needs to get done is actually expanded. So what I expect to happen is a few things. And, you know, primarily in the immediate few years, I don't see a significant across the industry reduction in numbers as much as a change in what people are spending time on. Yes, that does change the, you know, relative needs for kind of, if you're looking at people with, you know, greater ability to judge, to make decisions, It has implications for a few things. It has implications for what kind of people are going to be more effective. It has implications for how, as organizations, how do we move people to get that level of expertise over time to be able to better judge?

21:52How do we keep the apprentice model going in almost every function? There is a notion of an apprentice model. You come in when you're a junior with somewhat limited, you understand the tools well, but you don't have kind of experience. That's what you bring over time. that experience of what helps you judge the outputs of these models, of these new tools. And so we need to keep that pipeline going. We need to keep kind of the constant battle. On the other side, you know, banks have struggled over time with areas, for example, where we had declining skill bases. For example, a lot of banking software, you probably have heard about this.

22:31We have these, all banks have systems called core systems. These are the fundamental systems of record that manage, for example, the account balances and so on. Those were traditionally mainframe-based applications. Many of them were written maybe 30, 40 years ago. Many of them were written in COBOL. And the reason they don't change as much is there's a huge cost to change. These are the ones that underpin the entire network, and those continue to be the case. Previously, all of us as an industry were concerned that the number of people who, for example, knew COBOL was a declining skill base and that we as an industry would be struggling with what we call the retirement challenge.

23:11And how do we keep our systems kind of continuing to hum as well as they do right now? what we have realized is AI is actually a solution to that because what it does, among other things, is enables these new tools, enable an easy translation of what was written maybe 30, 40 years ago and make it easier for someone who does not understand Copal to essentially understand and rewrite something. So in some respects, there are new opportunities emerging from this, new challenges that are being solved by AI. So I think being able to articulate all of this collectively and being able to make sure it is true for the teams is the important part.

23:52Peter High: You've rightfully spoken about the necessity to focus on outcomes, Dilip, as a way to orient and prioritize where organizations focus their attention. And you've highlighted some of the outcomes already that you're seeing in terms of greater efficiency and productivity and so on. How do you see the outcomes derived from AI evolving in the foreseeable future into perhaps other or even grander benefits for an organization like U.S. Bank? Sure. I would say one of the first questions that people kind of spend time thinking about is AI. Is AI primarily about cost reduction? You know, cost is certainly a part of the opportunity.

24:41The bigger opportunity is to make organizations faster. For example, one of the immediate effects we believe is in our ability to build products faster and help make decisions faster. For example, make people's credit decisions much faster, make dispute decisions much faster, and give not just our employees more leverage, but help the customers achieve their financial goals in ways that they could not previously do it because of a lot of what we do. So I believe those would be, even time to market is a big deal. Then you go from there to enabling more personalization of capabilities, of products, of services.

25:28So there's a range of areas. And I think the exciting thing about AI at a large bank is that small improvements in a variety of areas, when multiplied across the millions of customers and thousands of employees, can truly become transformational. And so the objective that we have is not simply to have the most advanced AI, but to build an organization that can essentially turn these AI capabilities into trusted, measurable, impacted scale.

25:59Peter High: And talk a little bit further also, Dilip, about how the ecosystem that you have built to bring this to life, not only within your four walls, figuratively speaking, but also through the partnerships that you have assembled to ensure that you are advancing. If we stay with artificial intelligence, for example, advancing with great partners, be they scaled ones or startups, as the case may be. Talk a bit about the ecosystem you're building and curating. The ecosystem, by the way, in AI has dramatically also matured and shaped itself over the past few years. So the first and foremost is clarifying kind of what it is.

26:38Now we've realized that we need, you know, there are a few core things, just like we used to have the notion of operating systems in a prior world. We look at LLM's large language model kind of players as a base, if you will, foundation layer of AI. So we do, first and foremost, we believe, so we spend a lot of energy partnering with them, working closely with them, understanding kind of the capabilities of each model, kind of the next ones. And, you know, how do you work with them most effectively and, you know, both for defense as well as for offense. The next layer is translating these, in the traditional world, we used to call it middleware.

27:14But it's how you connect these kind of 16-cylinder engines, if that's the right analogy, to the actual kind of the crankshafts. And this is where you get a variety of players across each layer, you know, players that, you know, their MCP servers, their players that, you know, provide certain aspects of the scaffolding that let you, you know, harnesses, as they call them, right, that enable you to kind of use these models and connect them to your data capabilities, to our applications and so on. And here is where I've found over time, this is where we've had the most change. What we're finding is the frontier model companies have actually stepped up quite a bit.

28:04They're evolving their messaging from being just core LLM providers to, I would say, more like software solution vendors. I found that a fascinating change, one of more recency where they're talking the language of what does it take to deploy as opposed to talking the language of here's the LLM and its capabilities. It's a fundamental shift. Second, the middleware players are getting a lot more clarified. To engage with them, we've spent a lot more time than we've traditionally done with early stage players. The speed to maturity is much quicker now. Previously, we would wait for companies to scale up, to maybe have scaled deployments before we take them seriously.

28:46but now the gap between someone who's just launched a product and they're being used at maybe five banks was literally six months so we need to be there earlier kind of understanding being uh you know with our with investors vcs pe firms and so on learning kind of what they're seeing and being out in the market so we spend a lot of uh our team's uh kind of energy just kind of taking in um kind of um earlier stage players understanding what they have and just letting it soak in. So we do a lot of that. We do a lot more pilots with some of these companies to understand, to test the limits of what they can and cannot do.

29:22So the engagement model has been a lot more kind of fleshed out in those areas than it was previously. So I think it's a combination of those two. And then the traditional players, you could look at all our SaaS players that we use for various for HR, for finance, they are also evolving, you know, CRM. All of them are getting good at applying AI to their respective problem sets. And so we need to kind of absorb those as well and connect those with the rest of our capabilities because those could be faster in us in achieving some outcomes in specific areas.

30:04Peter High: Dilip, are there other trends that you're tracking now that particularly excite you that you believe will be particularly impactful to your company and industry more generally speaking? I think the softwareization of AI is actually a big deal. We're seeing early signs of that now. And that basically means it is more usable and more usable by enterprises by a larger number of people than it was maybe six months ago. And I give a lot of credit to the frontier model companies also to many of the players in the ecosystem that are quite advanced in their capabilities.

30:50Second, as much as we have our own capability, our talent, and we spend a lot of our time educating with learning and development and change management, most banks work with a large number of systems integrators and a variety of third parties. We have seen maturing of the capability there as well. Previously, everyone was learning, and now we are seeing all of them get more and more effective in a specific solution. So long way of saying, I think the industry is getting much more capable of building. The key decision that companies need to make, enterprises need to make, is to revisit how they think about the AI tech stack, in my opinion.

31:34Now, previously, there was a lot more build as opposed to buy. I think now there's opportunity to actually leverage more and more industry scaffolding to leapfrog to get to outcomes faster. If you look at our objective, the single most important element for us and for probably every other bank and enterprise is time to value. How can you get to outcomes, to business outcomes much faster? And so now what was true for doing that a year ago is not true now. And so being able to adapt and change that internally is really the most important, I think, management set of actions.

32:19Peter High: Can I also ask Dilip, what do you read or listen to to stay current during times of such great dynamism? I'd be interested in what you'd recommend to others that makes up some of your intellectual diet. Sure. I would recommend a few sources. I was just kind of thinking about them, and I actually go through them regularly. So the AI Daily Brief, Nathaniel, is certainly an important one.

32:48Dworkish's Podcasts, I find it interesting because they're longer-form interviews. Certainly the A16Zs, kind of, you know, various, you know, podcasts are interesting. No priors, latent space. And there is a podcast called Straticherry. I don't know if you've come across it. It's interesting as well. So a variety of ones, and of course, your own kind of columns as well. So, and they all come with different kind of flavors and perspectives, which is kind of quite interesting.

33:24Peter High: Excellent. Well, thank you for that, Dilip. And thank you for the fascinating conversation, more generally speaking. Wonderful to hear about the advances that you and your team are making on behalf of U.S. Bank and your company's customers, of course, and your own thought process of what's to come, a very exciting future that you foresee. And I'm so pleased that we had a chance to speak about it today. Thank you very much. My pleasure, Peter. Thank you.

34:03The가지고 dancing. Thank you.