Latest / CXOTalk / Enterprise AI at Scale: How U.S. Bank's Chief AI Officer Deploys AI Across 70,000 Employees | CXOTalk #906
Transcript
- 0:00U.S. bank is the 5th largest commercial bank in America with
- 0:0470,000 employees, nearly $700 billion in assets and over 2000
- 0:11branches. Deploying AI at that scale isn't
- 0:15a technology problem, it's a business challenge.
- 0:18Prashant Naratra is the bank's chief AI officer.
- 0:23Prashant, where do ideas for AI initiatives come from?
- 0:27How do they get decided? How do they get funded?
- 0:31We work across various different lines of business, both in a
- 0:35bottoms up and a top down approach, ensuring that these
- 0:39ideas are aligned to our business priorities, serve our
- 0:44customers in and and are executed in a fair and
- 0:48responsible manner. We evaluate these ideas not just
- 0:54for the value they bring, but also for things like technical
- 0:58feasibility. They're managing the risk,
- 1:02aligning with the risk posture of the bank and ensuring that we
- 1:08are doing these things in a transparent and commutative
- 1:11manner. And we have hundreds of ideas
- 1:14that we review and we are intending to deploy.
- 1:19Now let's take a moment to learn about Emeritus, which is making
- 1:23CXO talk possible. If you're a business leader
- 1:27navigating change or driving growth, explore executive
- 1:31education programs with Emeritus, a global leader.
- 1:36They offer programs developed in collaboration with top
- 1:39universities, designed for decision makers like you.
- 1:43There's a program tailored to your goals, whether that's AI
- 1:47and digital transformation or strategy and leadership.
- 1:51Find your program at www.emeritus dot org.
- 1:57How do you go through that evaluation or filtering process?
- 2:05Because I'm sure as you said, you have hundreds idea of ideas,
- 2:09but you're also under a very large set of constraints.
- 2:15How does an idea actually make it through that tunnel?
- 2:20And also, what's your role as chief AI officer in that?
- 2:24I'll start with the last point first, because I think that that
- 2:27is that is a key question. We see, we see more and more of
- 2:33Chief AI officer roles being, you know, executed and hired
- 2:38across not just banking, but across, you know, various
- 2:42business sectors. I look at my role as the Chief
- 2:47AI officer in in four ES. It's not just a new technology.
- 2:55I look at it as a transformative force.
- 2:58So I look at when I think about four ES, I think about
- 3:02evangelizing, educating our workforce, enabling various
- 3:10different builders across the US bank in building these AI
- 3:15capabilities. And then there are certain
- 3:18strategic priorities or key initiatives that we want to
- 3:22execute on that require absolutely new, you know,
- 3:27emerging technologies capabilities.
- 3:30I'm going to walk you through a few little details of how we
- 3:34actually work across in the bank is we first of all we need to
- 3:41make sure that these ideas are aligned with our overall
- 3:44strategic objectives and business priorities.
- 3:47So for that we work with our lines of business and our chief
- 3:52product officers to ensure that we are these ideas are something
- 3:56that they want to pursue and align.
- 3:59Then we once that has happened, we look at the feasibilities,
- 4:06the cost, sometimes these, you know because of how fast these
- 4:11technology is evolving, we want to make sure that these ideas
- 4:17can be executed using the capabilities that are already
- 4:21available. Sometimes, a lot of times they
- 4:25are, but sometimes they are not. So those ideas we will actually
- 4:29you know, add it to the backlog and we'll revisit that
- 4:33periodically. Once we have done that triage,
- 4:36we will evaluate these ideas on the change management, the
- 4:40readiness, but also the value both in terms of help to our
- 4:46customers, revenue growth or savings.
- 4:50And then we will try these ideas.
- 4:54We'll first pilot them, we'll put it in hands of a few of our
- 5:01users or clients, and then we'll scale.
- 5:05Throughout this journey, we are always looking to make sure that
- 5:09these ideas are feasible, they're delivering the right
- 5:12value, and they are aligned with our overall, you know, you know,
- 5:19values and risk posture. What you're describing sounds
- 5:24like a a a fairly typical process for technology in
- 5:29general. Are there unique aspects that
- 5:34come into play because because we're talking about AI here or
- 5:39or is it all part of the the general bucket of technology and
- 5:44therefore all handled simply the same way?
- 5:46One of the things that we are doing here when it comes to AI
- 5:51is making sure that intelligence that we are introducing in the
- 5:57process, we are not just making some process more efficient.
- 6:01We are not using just a better technology.
- 6:04There is a cultural shift here happening on asking the question
- 6:09about how does this intelligence, and I would call
- 6:13it an intelligence layer, not just a capability or a
- 6:16technology, How does that help us transform that interaction or
- 6:22transform the process to make it entirely seamless for our
- 6:27customers? One of the questions when it
- 6:29comes to AI that we are asking is, is this process necessary or
- 6:36required in its current iteration or current form?
- 6:43There's a lot of talk about how many of these ideas or pilots or
- 6:51prototypes depending on, you know, what phrase you are using
- 6:55are making to production. And what I can tell you is you
- 6:59can make, you can use AI to make something slightly better, or
- 7:04you can use AI to completely transform how you interact with
- 7:10our clients. Something that we, we just
- 7:14launched and our Chief consumer digital officer just talked
- 7:18about is we have, you know, we, we provide APIs and you know, we
- 7:26are have our developer portal, you know, that, that we make
- 7:29available to forms that we interact with.
- 7:32Not only, you know, we, we introduce generative AI based at
- 7:38the developer assistant. It actually helps them not just
- 7:42find the right APIs and capabilities for them to use,
- 7:47but it also helps work side by side with them as a wingman to
- 7:51help troubleshoot and identify those opportunities.
- 7:55So this is not just doing something slightly better,
- 7:59helping you find or creating a FAC or providing an assistant,
- 8:04but working with you as a wingman on how we can do these
- 8:08things entirely differently. If you're a business leader
- 8:13navigating change or driving growth, explore executive
- 8:18education programs with Emeritus, a global leader.
- 8:22They offer programs developed in collaboration with top
- 8:26universities, designed for decision makers like you.
- 8:30There's a program tailored to your goals, whether that's AI
- 8:33and digital transformation or strategy and leadership.
- 8:37Find your program at www.emeritus dot org.
- 8:43When it comes to AI, you are also evaluating the we could say
- 8:50the transformative impact of the technology as opposed to many
- 8:57other technologies. I don't know CRM system for
- 9:00example which is primarily driving greater.
- 9:04Efficiency, that's true. We are looking certain
- 9:07efficiencies, but it's completely changing the game.
- 9:12It's not just maintaining a Rolodex of a customer in a
- 9:19better system kind of thing. It's how do we contact the
- 9:25customer, how do we actually whether it's internal workforce
- 9:31or external clients, how do we interact with them better.
- 9:35Another example I'll give you is how we get the information while
- 9:41the customer is on in moment and is on the call with us.
- 9:49That's another example where we have seen our customer
- 9:53satisfaction jump significantly because now instead of taking
- 9:58minutes and a lot of awkward pause, we can provide them the
- 10:03answer quickly. And these are the things that we
- 10:06are significantly changing the game on.
- 10:08Another example I'll give you, and this one is really excited
- 10:13is not only we are changing, you know, doing something more
- 10:17efficiently, but changing how we are creating these assets.
- 10:22We are introducing AI for our developer workforce, for
- 10:27engineering workforce. And what we are seeing is that
- 10:32that time I use, you know, internally is Wingman.
- 10:36It's not just creating a, you know, software or features
- 10:40faster. It is creating the more
- 10:43intelligently being able to test them, being able to actually
- 10:47deploy these things faster. And we are looking at the value
- 10:51chain of end to end software development, deployment quality
- 10:55engineering. Can we do these things all at
- 10:59once? We will always like to, but we
- 11:03are going, you know, we are going about it in a methodical
- 11:06fashion and, you know, doing these things much better and in
- 11:12a much different way than you ever used to.
- 11:15We have a question from Arsalan Khan on Twitter and he's asking
- 11:23about metrics when, when you're generating the ideas of what
- 11:29could AI do for U.S. bank? Are there are there metrics?
- 11:35How do you how do you make those evaluations?
- 11:39And I know you spoke about this, but if you can again kind of
- 11:42focus on any metrics that you use?
- 11:44I always believe that you need to measure what's important, and
- 11:48what you cannot measure, you can't manage.
- 11:53We look at metrics on multiple different dimensions and, and
- 11:57different scales. So I don't know if, if I can
- 12:01actually talk about, you know, a more holistic view, but I'll
- 12:05give you 2 examples. You had me talk about, you know,
- 12:10creating better software and doing software faster.
- 12:15One of the things we did was automated, you know, using AI,
- 12:21both generative and classic AI to be able to review.
- 12:27Do code reviews provide a higher coverage and do those things
- 12:32faster? And when we started, we had a
- 12:35fair sense of like we think it will, you know, we might be, you
- 12:40know, 30 or 40% better than before.
- 12:44And as we developed the capabilities, we deployed it to
- 12:48a large to a small pilot group, we saw, we measured how what it
- 12:56used to be and what it is doing now.
- 12:58And then we saw that the actual results were significantly
- 13:06overshooting our initial expectations.
- 13:11And when we moved out of the pilot, we were much higher than
- 13:16a 50% faster rate of being able to analyze and review our code.
- 13:25And what that really means is, you know, how can it tells us
- 13:33what the benefits are, but also resetting those expectations and
- 13:38working with our business, with our development community.
- 13:42The other one, I'll give you an example.
- 13:44So that's an internal. The second one example I'll give
- 13:47you is when we rolled out being able to provide the nudge and
- 13:54the information to our contact center workforce on user Gen.
- 14:00AI, we saw what used to take us minutes, multiple minutes in
- 14:07providing the answer and a solution to our clients.
- 14:12We are able to bring it down into 10s of seconds and that's a
- 14:17significant improvement. Our clients are busy.
- 14:22They want to you know, get the information, act on it and go
- 14:26about their way. And now we can actually do it
- 14:30order of magnitude better than you know that they were before.
- 14:34So my advice and my ask of my team and direction is, has
- 14:40always been, is let's make sure that we set an initial
- 14:44expectation, we review it during the process and make sure we are
- 14:50only scaling up the most impactful of use cases and
- 14:59capabilities both internally and externally.
- 15:03Two of the reference points that I've picked up are #1 how does
- 15:10this benefit our customers? And then #2 Prashant, what what
- 15:18you just said, which is are we making the largest impact that
- 15:25we can make with the investment that we're proposing on this
- 15:30particular project or issue? I'm positive that all the users
- 15:34who are all the audience who are tuning in today, they have all
- 15:39heard about the MIT study that came out talking about 95% of
- 15:43the pilot's sphere. If you don't have the right
- 15:48metrics and if you don't have the right focus, which for us
- 15:52has always been our clients, As long as we keep those two things
- 16:00in focus, we can make sure that not only we have a very high
- 16:05degree of impact, but we are also being able to scale these
- 16:11things. There was one thing as you were
- 16:14asking me the question that I remember that I should say is
- 16:22it's really important for us to measure as we are rolling out
- 16:25these capabilities throughout the life cycle from pilot to,
- 16:28you know, a general availability scale version.
- 16:33But what's really important is what do you compare it to?
- 16:36So you need to also have a baseline of how you are doing
- 16:40those things today, how long it takes, what's the quality around
- 16:44it. Because you can say we are
- 16:46reviewing code faster, but do we know that we are making a
- 16:49measurable impact on the quality of the code or the quality of
- 16:53answers we are providing to our clients.
- 16:55So those are the things that are really important all the way
- 16:58from baseline in your existing to your new transformative
- 17:03interaction and capabilities. We do have a question from James
- 17:08Hook on LinkedIn who says how does the AI strategy cooperate
- 17:14with the sustainability goals of U.S.
- 17:17Bank? We publish data and report and
- 17:20information on it on an annual basis.
- 17:22We work with our partners who provide these capabilities.
- 17:25As we are bringing these capabilities and enrolling them
- 17:28out in a thoughtful fashion, they are not just aligned with
- 17:34our sustainability, sustainability goal, but we are,
- 17:39you know, constantly working throughout and sharing
- 17:43transparently these, you know, these capabilities and, and
- 17:48their usage externally. And for, For more information on
- 17:54this one, I'll, you know, I'll refer you to our, our corporate
- 17:58governance information online. On the topic of governance, we
- 18:03have a question from David Lowe's who on LinkedIn, who says
- 18:08what's currently the biggest constraint that you run into
- 18:14when trying to achieve outcomes at scale?
- 18:17Is it data governance, operating model or adoption?
- 18:23Where are the challenges? I've been very lucky to have the
- 18:27risk partners and the governance partners and the compliance
- 18:30partners I have at the back. We have treated this not as
- 18:38something that the AI Center of Excellence or the TPI officer
- 18:43does and then passes or works with our data team or works with
- 18:47our technology team and then passes it to the rest team.
- 18:50It's never been like that. It's been a highly collaborative
- 18:54effort. We meet regularly.
- 18:56We talk about not just individual use cases, but how
- 18:59are we accelerating AI adoption and building the capabilities
- 19:04and introducing the capabilities to the back and, and we look at
- 19:10the entire life cycle of of these capabilities on how we are
- 19:16able to do that. Working with our with my
- 19:20business line Chief Risk Officer, I was I was really
- 19:26encouraged on on his approach. Working closely, I can tell you
- 19:32that we have cut down. We have not just increased the
- 19:36time that it takes to roll out something this new and this
- 19:40transformative, but we have actually been able to cut down
- 19:44in the last six months the time it takes for us to go through
- 19:49the governance and approval process.
- 19:51Now, that doesn't mean that we are adding risk, but what that
- 19:55means is that we have made this a repeatable process.
- 20:00We are engaging my risk partners very early on in the process.
- 20:07We give them a sense of what is coming down, providing them the
- 20:10transparency and making sure that whatever we can actually
- 20:15learn during one new capability or use case introduction that we
- 20:21build it in the platform and we roll it out repeatedly and
- 20:26repeatedly. I should say, we are seeing in
- 20:31some cases the approval and the risk approval times being cut
- 20:36down in half so that the next one and the next one are going
- 20:41faster and better in terms of risk.
- 20:44I know you asked a multifaceted question.
- 20:48Risk governance data AI is not just another new technology.
- 20:55It's not just a slightly better tool to do what we were doing
- 21:00before as is. So that kind of transformative
- 21:04technologies, we have to look at everything that is involved to
- 21:08bring these things, this capability and putting to life
- 21:13and putting them responsibly in the hands of our users.
- 21:17So I work very closely with our Chief Technology Officer, with
- 21:22our Chief Data Product Officer and we are we are improving each
- 21:27one of these, in some cases incrementally and in some cases
- 21:32taking significant step changes and we are seeing good progress.
- 21:38AI by itself is in the panacea that we want it to be.
- 21:41We need to make sure that we are getting all the right partners
- 21:46across an enterprise and I should say internally and
- 21:49externally to make that improvement.
- 21:52Let's jump to another question. I love taking questions from the
- 21:58audience. You guys in the audience are
- 21:59really smart and your questions are wonderful.
- 22:03I should also mention that now would be an excellent time to
- 22:07subscribe to the CXO Talk newsletter so you can join our
- 22:11community and we'll notify you about upcoming live CXO talk
- 22:16show. So go to cxotalk.com and just
- 22:18subscribe to our newsletter. Do that now.
- 22:21So here's a question from Rohit Nakpal who says he's asking
- 22:27about that MIT study where it said 95% of pilots fail.
- 22:35And let me just, and he's asking where to get it.
- 22:37And just Rohit, just search on MIT pilot failure study.
- 22:43You'll you'll find it. But let's jump back a little bit
- 22:49and talk about pilots in general.
- 22:53Many organizations do struggle to move beyond pilots.
- 22:57So what has your experience been and what are some of the lessons
- 23:01that you've learned in moving from pilots to production?
- 23:06When it comes to AI or any other new transformative capability,
- 23:11there's always a lot of excitement.
- 23:12You want to be able to harness that excitement, those ideas, so
- 23:18we always end up with a lot of pilots, but if all we are doing
- 23:23is jumping from that excitement to running some of the pilots
- 23:26without a strategy and a goal of scaling, it really never works
- 23:32out. When it comes to AI
- 23:35specifically, scaling is the key.
- 23:38You want to make sure that you take that initial excitement and
- 23:43don't turn it into a list of pilots or a whole bunch of that.
- 23:48We talk about the strategy of having 1000, you know, flowers
- 23:54bloom. What we need to do what we have
- 24:01done is we create these pilots and we bring those groups
- 24:07together in, in terms of pilots and make sure that we have, you
- 24:12know, we have a very quick and transparent process to take it
- 24:17to scale again. I'll, I'll go back to this
- 24:21example of, you know, product development or AI assisted
- 24:29product development. We we had three or four
- 24:35different pilots on how to do code reviews.
- 24:38We ran it, we brought the group together and we took the right
- 24:43ideas and the right approaches from various different pilot,
- 24:47decided on one final product and then we quickly scaled it.
- 24:55But we did it with all the groups having a seat at the
- 24:58table. And right now it's in the hands
- 25:02of thousands of our code developers.
- 25:07If, and we need to make sure and, and I, I know you asked
- 25:12this in the very beginning of the conversation, is the, you
- 25:16know, the, the funding side of it.
- 25:19We need to make sure these pilots, we are able to find not
- 25:24just the pilot, but also the scaling of it.
- 25:26And scaling requires tackling enterprise, you know, size
- 25:33constraints and you know, things like, you know, disaster
- 25:38recovery being change management.
- 25:41So we have different groups engaged in the change management
- 25:45side of it to make sure that these pilots are taken to a
- 25:51successful evaluation stage. This is from Sonia M, who asks
- 25:57about how do you align the AI strategy with the bank's overall
- 26:03business objectives and risk management frameworks.
- 26:08And specifically, can you talk about the governance structures
- 26:13you've established to oversee AI development and deployment?
- 26:17When I joined and I started to set up the AI strategy, me and
- 26:23the team started with the business objectives and the
- 26:26business strategy because AI by itself or AI strategy by itself
- 26:32doesn't occur in a vacuum. Now in some cases, these
- 26:38transformative capabilities or AI can actually certainly
- 26:42influence us, but we always are not Star is our client or our
- 26:48customers and it's the business strategy that needs to be
- 26:53enabled by AI strategy. So we have never, I don't, you
- 26:59know, get in, get into the, you know, situation where AI
- 27:04strategy is, you know, not a lie.
- 27:07So I don't, you know, that that's one thing I want to, you
- 27:11know, call out. Secondly, in the governance
- 27:14structure, very early on in the process, U.S.
- 27:17Bank I would say is you know, when it comes to risk and
- 27:22governance, we have a very good posture and a very robust
- 27:27posture when it comes to risk and governance both.
- 27:31And what we do here is we make sure that we are creating both
- 27:36something that's transparent up to the, you know the leadership,
- 27:41but also down below. So we have a governance, you
- 27:47know, in terms of regularly reporting to our board, to our
- 27:51management committee, but also to my day-to-day partners who
- 27:57are, you know, not just showing them the capabilities or the,
- 28:02you know, their use cases that we are tackling, but also the
- 28:05platform we are building. So that ultimately for a well
- 28:10managed and well run, you know, company using AI, this
- 28:16capability has to be in the hands of every single builder,
- 28:22creator, developer and user. And for that we need to make
- 28:27sure that they are educated, there's a good platform and the
- 28:33governance is a feature of the platform rather than an
- 28:38appendage. You're raising a very important
- 28:41set of issues around AI culture and what that means.
- 28:47And we actually have two questions on LinkedIn that
- 28:52relate to this. So let me combine these.
- 28:54And this is from Michelle Allman and Maya Cunningham.
- 29:00And Michelle asks, what are the minimum skills every leader and
- 29:03risk partner should have for AI enabled work in 2026?
- 29:08And Maya asks, what does AI education and enablement look
- 29:14like for employees at U.S. bank? And let me roll both of these up
- 29:20into this broader question of the culture transformation
- 29:25that's required in order to achieve the kind of broad AI
- 29:30literacy you're describing. The advancements that are
- 29:34happening, the baseline keeps on improving and and going up and
- 29:39it's upon all of us to make sure that we are finding the right
- 29:44resources and working with the teams across the board to
- 29:50educate ourselves. One of the things that we are
- 29:53doing at the bank is we are making sure that we are rolling
- 29:59out learning programs for our entire workforce.
- 30:05It's not just, you know, the the leaders themselves, it's the all
- 30:10the way from the top, you know, to, to the frontline associates.
- 30:16And we are doing it based on a persona and the, and the roles
- 30:20they are playing in leveraging, building, strategizing and, you
- 30:28know, influencing the direction of the company.
- 30:31So all of these resources have to make sure that they are
- 30:35thinking of of you know they are they are well educated and aware
- 30:40of AI capabilities for us to be an AI first back.
- 30:45And I have great partnership, I would say with our learning and
- 30:52development and human resources and rest teams in making sure
- 30:57that these these resources are not just simply rolled out once,
- 31:04but they are continuously updated.
- 31:07And we connect these resources to the tools that we are rolling
- 31:12out and to their roles that they are playing in part of the
- 31:19company. So no matter where you are, we
- 31:22have something to help you adopt AI, or build using AI, or
- 31:29strategize using AI better. It sounds like enabling this I I
- 31:35call it AI culture shift is a very important part of your role
- 31:41as the Chief AI Officer at U.S. bank.
- 31:45I'm not trying to put words in your mouth.
- 31:47It absolutely is. The way I actually would say is
- 31:51it's not the job of 1 chief AI officer or one small group to be
- 31:56able to, you know, roll this out.
- 31:57It's all of us are in this together.
- 32:00It's a transformative capabilities, a capability and
- 32:04we need to make sure that we are all in it, well educated, well
- 32:08prepared for the future to help our clients.
- 32:12For those of us who have been around for a while and have
- 32:16observed things like ERP roll outs, the language in some ways
- 32:23is similar. If you're rolling out an ERP,
- 32:26really it's a transformational change to the organization.
- 32:32But AI is different. So can you, can you talk about
- 32:36the unique aspects or or unique issues that AI drives?
- 32:42There are two things that I would say #1 is for the first
- 32:47time we are able to use the quality, the quantity, the scale
- 32:55and the speed of data and make that decision.
- 32:58We are able to push that decision to the front line
- 33:02faster than ever before #2 the problems we are facing are
- 33:08coming are compounding or being able to meet that at the scale
- 33:14unlike anything we have done before.
- 33:16When we think about an ERP is a great example.
- 33:20When we think about using ARP, you are using it in pockets,
- 33:26different groups are using it slightly differently and not
- 33:31everyone is using it right. When you think about AI, and I
- 33:35know we use the term AI very broadly, but no matter what your
- 33:41pane of glass is, no matter what data you are using, you are able
- 33:47to use, you are able to bring that intelligence or augment
- 33:52your human intelligence with this, you know, with AI in
- 33:58general capability better and being able to do at a scale that
- 34:03has never been done before. And we are, I know it seems
- 34:10cliche to say we are in the beginning stages of it.
- 34:13We talk a lot about agent AKI. We'll talk a lot about
- 34:17autonomous decision making. We have a, you know, we are able
- 34:22to do those things. And if we don't, if we don't
- 34:27actually get our workforce educated and trained and ready
- 34:31for it now, we'll be very much behind, you know, later on.
- 34:35And I don't mean to say we as just the US Bank, but we as you
- 34:40know everyday developer and users of of these technologies
- 34:47and systems. It's a really interesting point
- 34:50that now is the time to prepare the ground for the technologies
- 34:56that we know will be here in the next, you know, year or two.
- 35:01It's a really important point that you're making, seems to me.
- 35:04Absolutely. This is a really interesting
- 35:07question from Connor Whalen. And he says, as you scale AI
- 35:13driven outreach, how do you make sure that automated customer
- 35:19contact remains aligned with evolving consumer contact
- 35:25regulations? There are two portions to it.
- 35:28As we actually scale the contact, as we scale the use of
- 35:33AI both internally and externally, we have guardrails
- 35:37in place, We have human in the loop in the place, which I would
- 35:43say is our driving principle right now.
- 35:49And we are making sure that we are evaluating and reviewing
- 35:54those contacts. So that's one.
- 35:56Number two, we work with our both our regulatory partners,
- 36:00our internal risk partners and our vendors to make sure that
- 36:05these systems and these capabilities, we are always at
- 36:11the forefront of bringing in the latest and greatest not just
- 36:15technology, but also adhering to the compliance and regulations
- 36:23to and that it is in line with our risk posture and with our
- 36:31clients expectations of how we interact with that.
- 36:37Final point I'll bring on this one is even before we roll out
- 36:42any capabilities, we want to make sure that it is done in
- 36:46such a fashion that we are mindful of our customers
- 36:53expectations and of their perception of us.
- 36:56Your reference point again continually seems to be how does
- 37:03what we're doing with AI affect our customers?
- 37:07Is that an accurate perception on my part?
- 37:10Understanding. Absolutely.
- 37:12Our clients are the center of everything we do.
- 37:17We have another question from this is now from Lisa Pratico
- 37:22says, what does the leadership organization structure look
- 37:27like? And is there a steering
- 37:30committee that includes you, your Chief Sustainability
- 37:33officer, if that role exists, and your CIO on strategy
- 37:39assessments and execution of, I'm assuming, AI projects or
- 37:44proposed projects? We have a core group between the
- 37:48Chief I already mentioned as Chief Officer, our Chief Data
- 37:52Product Officer and the Chief Technology Officer.
- 37:56We also work very closely. I also work very closely with
- 38:00our managing committee and different teams across the bank.
- 38:09We have regular touch bases and they are they're not just
- 38:16learning about, you know, the three of us.
- 38:20We are not just learning about, you know what we are doing, but
- 38:24we are strategizing. We are creating these plans and
- 38:28we are understanding the impact of the decisions we are making
- 38:33and working and building those in the plans together.
- 38:38AI by itself doesn't exist outside as I said.
- 38:42And Michael, thanks again for pointing it out is we talked
- 38:47about putting clients at the middle, you know, at the center
- 38:50of it. It's our stakeholders, it's our
- 38:53shareholder, it's the community that we serve and then that we
- 38:58live in. We need to make sure that we are
- 39:02bringing all of these things together and our strategy and
- 39:05our vision serves that. Let's talk about data.
- 39:10We have a question from Arsalan Khan, who is asking about the
- 39:17value of data. Do you see data as an asset and
- 39:21how do you think about the value of that data?
- 39:25And wouldn't different senior leaders place a different value
- 39:32on data? And so how do you manage that
- 39:36type of thinking? One of the unique things that I
- 39:39feel like U.S. bank has done is we have put our AI group
- 39:46together with data, our Chief Data Product Officer and our
- 39:52digital team. So our group is called digital
- 39:55data and AI and in this role. And the second point, and I'll
- 40:01elaborate it in a minute, is. AI without data, I know we heard
- 40:09a lot about hallucination in generative AI, but AI without
- 40:14data itself is a hallucination. So you really need to have a
- 40:18very good and trustworthy data foundation for us to be able to
- 40:25leverage AI. So it's essential.
- 40:29I cannot emphasize how foundational that is.
- 40:35And the other thing is, as I was mentioning about digital data
- 40:38and AI is from building a very good foundation of data and
- 40:45being able to and interacting with our customers and clients
- 40:50and employees in a digital layer.
- 40:53AI is that intelligence layer that bounce it together and
- 40:57makes that interactions not just more efficient, it makes it
- 41:02really fast, it makes it much more holistic.
- 41:07And bringing these three groups together under 1 leader has
- 41:14really helped us accelerate rolling out these capabilities,
- 41:25reducing increasing the number of pilots to production and
- 41:30scale, but also making sure that we are we are coming across to
- 41:36our customers, our clients and our employees as one entity.
- 41:41I keep saying this and and struck by is the reference point
- 41:47for you on pretty much everything you're doing is what
- 41:51will the impact be on the your clients?
- 41:55That's always the North Star, it sounds like.
- 41:59Absolutely. Client is are not starred.
- 42:01They are the center of everything.
- 42:03And you know, I, I might have mentioned this before is and we
- 42:07also want to make sure that our, you know, various different
- 42:10stakeholders like community, you know, are, are part of that
- 42:15consideration. But clients are are not starred
- 42:18in the center of everything we do.
- 42:20This is from Grace Pappas, who says a simple quick question.
- 42:25Can you talk about the technologies or platforms that
- 42:28you use to drive these initiatives?
- 42:31We already have a lot of, you know, you know, technology and
- 42:35vendor partnership. We work with different hyper
- 42:38scalers, but when it comes to AI, we are bringing in new
- 42:42capabilities and we are working with a lot of, you know,
- 42:47startups and entrepreneurs who are thinking of making available
- 42:54or solving some niche problems that come together with AI.
- 43:02Guardrails is one example that we are partnering with vendor
- 43:10partner of ours. I unfortunately won't be able to
- 43:13name names for us, but there are some unique ones and we are also
- 43:19leveraging our preexisting partnership across the board.
- 43:24This is from Sean Tumenov on LinkedIn who asks, as we shift
- 43:30from a build from scratch error to fine tuning leveraging open
- 43:35source foundation models on proprietary data, how must a
- 43:39bank's organizational strategy evolve?
- 43:43Specifically, structural changes required to transition from
- 43:47model development to high stakes model integration.
- 43:52Build versus buy has always been a interesting conversation when
- 43:56it comes to any new technology and there is not a single answer
- 44:03when it comes to build versus buy.
- 44:05It can be various, different lines of business, your
- 44:08different, you know, use cases, your different, you know,
- 44:11internal customers and employees who are using it.
- 44:16So there's a lot of art to it in addition to, you know, the
- 44:20science about it. What's really unique about, I
- 44:27would say generative AI is for the first time that we are able
- 44:32to leverage external models at scale without modifying them.
- 44:39And what we do is provide a unique perspective in terms of
- 44:44the user, the client, the prompts and the data that we
- 44:48provided. And over the past a year and a
- 44:53half and at the bank and really before that is I have worked
- 44:58with various different risk teams, various different audit
- 45:02and compliance teams in making sure that we are evolving our
- 45:06process in evaluating and managing that risk.
- 45:11Because if you look back at I would say November 2022 and
- 45:19before, 90% of the models that you have that any organization
- 45:26used to use was really homegrown or inbuilt and had a high level
- 45:32of customization and control. When models like transcription
- 45:40and image analysis came by, those were the first time that
- 45:44we started to use external model set scale without modification.
- 45:48But the the balance is not tilting in favor of models for
- 45:54hire more than Model C Bill. Here's a question from Austin
- 45:59Dirksen on LinkedIn, who is a an MBA student, a finance student,
- 46:07and he says this, he has a good question from how how should we
- 46:11think about AI as a strategic investment rather than just a
- 46:16technology cost? What signals tell you that an AI
- 46:21initiative is actually creating durable value for a bank like
- 46:25U.S. bank? It's a great question.
- 46:27We are measuring the value at every step of the way.
- 46:30We are understanding the applicability and the reach of
- 46:34the AI capabilities that we are building, adopting or rolling
- 46:38out. And we ensure that it aligns
- 46:42with overall with our risk posture.
- 46:44And it's something that we want to add, we want to add to our,
- 46:50you know, to the capabilities that we offer to our clients.
- 46:54This is from Eric Vosch, who says when you're scaling AI use
- 47:00cases, how do you prioritize use cases to start testing that to
- 47:07ensure that they align with your business objectives, feasibility
- 47:10and readiness requirements? So how do you prioritize when
- 47:13it's time to test? In terms of prioritization, you
- 47:16know, we, we, we work and depend a lot on our business partners
- 47:20and business leaders on what makes, you know what is
- 47:23important for them and their customers.
- 47:26In terms of when we are ready to test, I, I think it's more and
- 47:31maybe I'm just not getting it the question right.
- 47:34It's more of a question of what capabilities that we can you
- 47:38know, build or roll out or adopt fast and when is our user
- 47:43community who's able to use that is ready because change then it
- 47:49comes to AI, an educated workforce and a change ready
- 47:56workforce is really important. U.S.
- 47:58Bank has significant insight into client behavior.
- 48:04How do you translate that insight into AI experiences that
- 48:09customers will find valuable, but without feeling invasive or
- 48:16crossing that line into creepiness?
- 48:19Many of us have actually, you know, bought homes, cars or you
- 48:22know, any other capability, any other asset over a period of
- 48:26time. How would it feel like if if
- 48:29you're a bank who you borrow for your mortgage, sends you a
- 48:36marketing material or an offer three months after you have
- 48:40closed, borrowed from them and tries to sell you a better
- 48:48mortgage? And it's like, you know me, I've
- 48:52banked my banks for decades and decades.
- 48:57Why are you not offering me a more intelligent service?
- 49:01Our expectations as a consumer have evolved over time, where we
- 49:07almost expect with these institutions that we interact
- 49:13with, to know us better and interact with us in a smart,
- 49:18intelligent and knowledgeable way.
- 49:21And so one of the things we tried to do is make things more
- 49:25contextual, take into account the right information that we
- 49:30are aware of and have of our customers and make sure whatever
- 49:34we are doing with them feels relevant to them.
- 49:38It's a very fine line between using data to make the give the
- 49:45customer an intuitive experience versus making them uncomfortable
- 49:50because you know too much. I would agree with that.
- 49:53And that's why we also have a responsible AI principles, you
- 49:57know, that we established at the outset of the program.
- 50:00And as I said before, let's make sure that those interactions are
- 50:05relevant, timely, and you know, we need to contextualize that
- 50:11based on a customer's expectation of us.
- 50:14We are out of time. Prashant Mehrotra, US Bank's
- 50:19Chief AI Officer. Thank you so much for taking
- 50:23your time to be with us today. I'm very grateful to you,
- 50:25Prashant. Thank you.
- 50:27And everybody, thank you for watching and thank you for your
- 50:31awesome questions. Before you go, check out
- 50:34cxotalk.com and subscribe to our newsletter so we can keep you up
- 50:39to date. And you can join us for upcoming
- 50:42episodes, which we have great ones.
- 50:44Thanks so much, everybody. We'll see you soon.
- 50:47Have a great day. Bye bye.