What It Takes to Build One of the World's Biggest Banks

24:41You sent it through snail mail. It was timestamped at the right time, but it went through snail mail. It got lost. You got penalized to fee, and you go into the branch, and you say, you know, it's really not fair. I sent this, and you're hitting me with a fee. and the branch person, what do they do? Right? There's some answer to that. What's our procedure? Well, you know, now with AI, you can just go into a large language model that is built off of all of our customer service protocols and say, what do I do? This just happened. It'll tell you. You know, credit the customer the fee and send a note to the, you know, whatever the answer is for that particular problem.

25:16But the person sitting there needs to know, like, which system do I query? What's my browser? What am I? Who do I talk to? You know, if I'm going to make an exception, do I go to the branch manager? Can we do that here? Or do I get to call somebody in some strange room in a back office in Philadelphia? All of that with us, because we automate so much of it, is now a sign of query and self-learning in the branch. And then importantly, more and more with our customers, self-curing. You know, they got locked out. You know, just be able to fix the problem yourself. One thing I've wondered for a few years, does local expertise matter anymore when it comes to actual underwriting, either in retail or commercial?

25:57Because I think most of America used to have this very romanticized notion of the banks, the sort of it's a wonderful life model where your branch manager knows you and knows your entire life and business and all of that. My sense is that doesn't exist so much anymore. But tell me if I'm wrong. You're wrong. Okay. It's harder to accomplish when you become a larger bank. But how has credit evolved through time? Consumers. Consumers now, if they're an existing client of you, are scored based on their cash flows that you see. So not a FICO score. You can actually see when did somebody pay the rent or their spending pattern is predictable.

26:39You can build models that say, you know what, this person is a really diligent good credit, even though it might not show up in a FICO. So local market business, you're going to lend to this small developer. You're going to lend to this store on the corner. Branches really matter and local bankers matter for small businesses because they see where the foot traffic is. They know the reputation of the person who's running that business. Do people, you know, are they a community influencer in a good way where they're, you know, people want to support them? All that stuff still matters. And all that stuff, the more you can give somebody voice locally in the decision, I always kind of call it putting a thumb on the scale a little bit on a yes or no that's otherwise mechanical, the better your credit underwriting is.

27:25Well, since we're talking about consumer credit, what happens to America or the industry if there's a 10 % cap on credit card rates as President Trump has proposed? So the only thing I know for sure, if that happens, is that all the credit card businesses in the country lose money if run exactly the way they are today. So we have a credit card business. I get swipe fees. I get most of those away with rewards. I earn credit when somebody is revolving a balance. I lend credit when somebody pays it at the end of the month. I'm fronting that money for you if you clear your account every month. But hopefully I'm getting some swipe fees to make that good.

28:05After all that, you charge interest rates, you have losses, and you get a margin, maybe you make 4%. And assume interest rate, I'm making up numbers here, but assume maybe the average rate is 18%, I make 4 % at the bottom. Now I'm going to take my 18, I'm going to drop it to 10. Uh-oh, I just made minus 4%. And so all that math is going to come out, which is why I don't think this goes. Like, there's almost no way this can happen. You'll just simply shut down consumer credit in the U.S. They can't force, they can say the cap is 10 % and I'll say, fine, I'm out. Like, why would I do it if I'm going to lose 4 % a year on that?

28:40Something's going to change. Fees are going to go up. Lines are going to get cut. Rewards will go away. You know, and something will emerge out of the dust. But it's not going to look at all like it does today. What do you think the genesis of the proposal or the rationale behind it actually is if, you know, everyone in the industry says this basically ends credit cards? Yeah, I think because, first of all, the emotional appeal of it. I mean, it sounds absurd that you're charging somebody 18 % for a revolving credit. I think that sounds high. That happens to be the rate that you need to charge to make money, but it sounds, which is why I never revolve a balance.

29:17I pay it off every month. So there's this emotional appeal. Jeez, you ought to be able to lend. Like, I'll lend money to you at 10%. I'll lend money to you at 10%. But I can't necessarily figure out who you are in a sea of people. There's a bunch of people I'll lend money at 4 % because they'll never de-pull. That's the issue. When you try to get to the broader population and extend credit to the consumer that drives our economy, science isn't so exact to say you get nine and you're 22. I don't know about you. Well, actually, I want to press this further because intuitively, I would think you'd be able to do that.

29:52You just mentioned the fact that you have granular access to propensity to payback that is not even captured by FICO. You have incredible amounts of data and predictive capacity on the difference between whether me or Tracy are going to evolve. So when someone says 18 % seems high, are we supposed to just like believe, oh, just trust me, it's high. This is what I need to charge in order to make money in this business. I mean, I could trust you, but like how would you? But the math is the math. Okay. At the end result, right, you say when you run your models and you look at propensity to pay you back, you basically – whether you're looking at FICO or anything, FICO is only as good as whether that person has a job next week.

30:38Right? You can look at predictability and, yes, this person has been paying their rent. They have a steady expenditure thing on food and on gas and on entertainment, and their savings account balances always stay pretty steady. That ought to be a good credit. Well, it isn't a good credit if they get fired or lose their job because the economy's in a downturn. And all of a sudden, well, I thought that was a really good credit. I should charge 5%. Turns out I just wrote the whole thing off because there's no recovery value in that credit. That's where the science fails. There's exogenous variables you can't model.

31:14And the wider you cast the net, the bigger the dispersion on outcomes. However, if we've settled there, that interest rate kind of sets a reasonable profit margin. The largest card issuer in the country, you can figure out who this is, right? Best at it, best computer models, best finding credit, best extending credit to non-prime borrowers, they trade at one times book value. If this was a genius business like this way too much, I'd drop the rate by 1 % and grow share. The problem is it's not so great, actually. I want to talk about private credit a little bit because obviously this is an area where we always hear about banks face new competition from middle market lenders and all of that.

32:07But at the same time, we also see banks teaming up with private credit now. We have also. Yeah. Okay. Talk us through that relationship. Yeah. So traditionally, standalone, high-yield credit. If all you did was walk around and just lend money with no particular expertise, but across leverage clients at a single big credit, that's a really lousy return on capital. All the science going back to when Moody started tracking this stuff in the 80s, you get kind of a 35 % recovery. You have a 10 % default rate. You go through cycles, sometimes a lot worse, sometimes a lot better. It's not a particularly good business.

32:48It becomes a better business and sometimes a fantastic business if you couple that with the wallet that goes with that company beyond just the credit. So IPOs, M &A fees, bond field, whatever the stuff is that goes along with it, all of a sudden it becomes a pretty good business. What's happened in the last couple of years, we've seen this drift into more and more credit leaving banks, going into private equity and other alternative forms. At PNC, it manifested itself where we'd have a client that we might have banked for 100 years owned by third generation family that just got sold to private equity.

33:24And our client's gone. Because I don't want to make that loan at that leverage ratio. And whoever does make the loan is going to demand the treasury management relationship and all the other stuff that goes with that client. So where we come out at it is I don't know that I necessarily want to make the loan. I will sometimes. But I want to keep the relationship. So our partnership with TCW as an example is when that happens now, we put capital in the TCW fund as did a bunch of other people. But it happens now, we'll keep the client. You know, we'll diversify the credit risk, but keep the fees.

34:04So the return on equity on that thing, you know, goes way up. The whole notion, remember, like private credit in the last three years, oh, my God, it's a new, new thing. We're great at it. We're going to make you a fortune. Why is that? Private equity kind of hit the skids. Nothing was getting sold. They need to raise new money. We haven't had defaults in 10 years. So a manager shows up and says, look at my track record. I can earn, you know, with two times leverage. I'll give you 10 % return after fees. This is fantastic. Private equity sucks. Come with me. Of course, we haven't had a credit cycle in 10 years.

34:38So those stats are all garbage. You actually go through a credit cycle and you have two times leverage on that thing and you start having defaults, which we've had a few of. All of a sudden, it's not really exciting.

35:03Hey there, OddLots listeners. As we come into 2026, we are realizing that one thing we're constantly thinking about on the show is how companies actually get built. Not just like the headline version of that story, but the messy operational reality of it. Right. We love messy operational reality of things. The never-ending quest to dive deeper, how companies make it big, what causes one company to succeed, why others fail. Well, I have good news. That is exactly what the Acquired podcast does. Ben Gilbert and David Rosenthal pick a company and then explore all the ins and outs of its trajectory.

35:44Lots of detail there, how it scaled, the ups and downs, and so much more. Yeah, and we actually had them on Odd Lots back in February last year. We talked to them about everything from TSMC, NVIDIA, Mars, Hermes, scale, capital structure, the importance of incentives, all of the different, I guess, ingredients that go into some of the success of these names that we talk about every day. Also, their show actually turned 10 years old in 2025, just like us. So we're, I guess, the same age in podcast years. Big year. Anyway, if you like OddLodge, the way we get into various market dynamics, how the economy actually works under the hood, you'll obviously appreciate and enjoy the Acquired podcast.

36:25They do similar work, similar ideas, all focused on the context of individual companies. So go check out the Acquired podcast. You can find them wherever you get your podcasts.

36:39Can I ask one macro question in this conversation, which is where do you think we are in the credit cycle at the moment? I've given up trying to predict it. Credit today is pretty good, better than pretty good. A lot of the leverage that's in this system that everybody complains about are actually with things that are very good enterprise value. So I don't worry so much about the disappearance of a company. I think about it in terms of reallocation of the ownership structure. If they're over levered, they go down. Either private equity firm bails them out or they say, no, here are the keys and now the debt holder owns them and then the debt holder restructures the debt and sells them again.

37:18Somebody lost money, but the company is still a good company. It was just over levered to go. Most of the noise you're hearing right now on credit beyond the fraud, you know, the cockroaches that Jamie said come out when things get stretched. Most of it are decent enterprises, just over levered. And then, of course, you have the fraud. You haven't had anything that's really going down today because the economy just crashed in a spot other than some isolated issues with tariffs. There is a cyclical element to fraud, right? Because in good times, people look the other way. I don't think it's cyclical.

37:54I think it's always there. I just think, you know, who's it said, you know, when the water goes out, you can figure out who's swimming nude? I think that was Buffett. It was Buffett, yeah. Yeah. Well, Galbraith, however, talked about the bezel, right? The bezel. The propensity of flim-flams and scams to exist during the good times and people just like don't – and that accumulates and accumulates. That could be – We've had like 20 years, as you said. Like we haven't had a credit – I mean it's almost 20 years since the financial crisis now. Yeah. Finance always finds a way to blow itself up. I mean, it's not an opinion.

38:34You just look at history. I think what happens is we find good things and then you write them. And in every instance where something's gone wrong, you just keep pushing on something until you take it over the edge. Finance is an industry where historically you can make a lot of money at it. You're playing with other people's money. you know pennies and nickels come off of other people's money i'm just going to keep pushing i'm going to keep pushing i keep pushing when it finally falls it falls hard we haven't had that you know what's the new new thing now it's fintech and it's crypto and it's you know banking has been kind of throttled by regulation for 10 years well speaking of finance always finding a way to blow itself up can we talk about uh silicon valley bank and 2023 because you know they managed to blow themselves up for a variety of reasons.

39:28But one of them was they just had too many bonds, which is kind of an unusual one. What was that experience like for you, first of all? Let's go back to Silicon Valley. I actually think they had a great institution. Who's been saying that many times on this podcast? The Silicon Valley Bank respecters. Okay, keep going. No, it's just it was a unique business model with a very loyal client base that they – Not that loyal in the – No. I mean, look, if you're dying, loyalty goes out the window. But they blew themselves up in something that was not at all core to their basic business, right? Interest rates are super low.

40:09They got as long by some calculations as like long-term capital was back in the 90s. I mean, they were really, really levered and underestimated the negative convexity of deposits, right? You think, you know, on average through the cycle, they're going to be there seven years. Oops, they're gone tomorrow. And so they just blew themselves up on rates that had nothing to do with their core competence or what they were as a franchise. And it's unfortunate. It was a rookie mistake. There was nothing. It's just a crazy position to be in. How scary, you know, those days afterwards. And then people started hunting for like any bank that had the word California in the name, like people were shorting it or people were selling it.

40:49Or, you know, you had headlines where, you know, the regional bank index, the regional bank index, that's why I'm a national bank. We are not a regional bank because the next time they put regional bank headlines in there, I don't want to be associated. Yeah, I get it. No, we actually benefited during that. You know, we're a safe haven and we actually gathered deposits and opened thousands of new accounts during that period of time. And then every day somebody would flash the regional bank index and we'd get blasted on our valuation, which is fine. And it always, you know, valuations are short term.

41:20It all came ripping back after that. So one of the issues that came up during the 2023 banking drama, I guess, is the discount window and banks' reluctance collectively to use it. And this is something that the Fed's been trying to change, right? And actually, do you remember in late 2022, there was a big spike in tapping usage of the discount window? And it turned out that wasn't any of the banks that ended up getting into trouble. It was Buell Bank down in Texas. It's a weird bank. Yeah, it's a very weird bank. But anyway, regulators trying to get banks to use the discount window more, feel more comfortable with it.

41:59How do you feel about it? Well, that's a great question. So before the financial crisis and before regulation and LCR and long-term debt proposals and all that other stuff, there was a very active Fed funds market. So in a given day, the excess reserve a bank would hold deposited in its Fed account was tiny. And we would trade Fed funds around with each other. If I needed$2 billion, you'd give me$2 billion. Next day, I might give you$2 billion. Like big size Fed fund positions. And that's how you balanced your books at the end of every night. And then they come out and they say, hang on, we don't want you doing that anymore.

42:42There's an LCR requirement. You have to have this much liquidity stuck aside instead of going into the Fed funds market. That gave rise to all these excess reserves. Fed fund market is basically dead. The Fed sets that rate. Nobody trades on that rate. It's as dead as LIBOR. It's gone. I don't even know why they talk about it anymore. Now you have, they basically said, you don't use the discount window and hold all this cash. You just froze the economy because you stopped the circulation of capital transformation. Home loan banks step in. Because I have to hold all of these treasuries and mortgage securities, I can pledge them to home loan and they'll lend me money at a pretty good rate.

43:24So that's going to be my outlet. I'm going to go to home loan and I'm just going to borrow there at a pretty good rate or I'm going to repo my treasuries that I hold. Silicon Valley happens and they had home loan lines. They didn't have any discount window lines. Now, the discount window will lend against treasuries and agencies pretty simply because it's basically settled electronically through effectively the Fedwire system. If you're pledging loans, it's really hard to pledge. So if all I have is securities to pledge, I should just repo them or I'll give them to the home loan bank. Discount window will do more than that.

44:05But if I want to draw against my loan balance, I need to give them physical wet signature loan docs in a vault, guarded, audited, 24 hours a day. Like there's got to be a guard and I have wet signature stacked up. So they know that when I draw against that loan, I actually have that loan because it doesn't settle Fedwire like a bond does. This is a loan, right? I lent you money. You signed the thing. I put it in the drawer somewhere. Then you say, all right, let's pre-position these loans. We spent, PNC spent millions and millions and millions of dollars. So the vast majority of our C &I loans are now at the discount window.

44:42We don't borrow under them, but they're there. If we wanted to borrow under them, I got a call. It's gotten a little better. But basically, like during COVID, you know, or post-COVID when nobody was coming into the office, you'd call. Like the way you borrow from the discount window is you call Cleveland. They're our Fed. And then they say, oh, OK, yes, you're in good shape. We'll allow Washington to allow you to use the discount window. And then they'll call Washington. And then maybe like a day or so later, you can have money. When nobody is coming into the office, nobody would answer the phone.

45:12I got to draw in the discount window. That just rings. Nobody's there. God, there's no system. It's not like I can type in and draw. It's not like I can do a Fed funds trade through a broker and draw the money. It's unusable. But now you're pre-positioned to borrow against loans if you need to. Yeah. We have enough liquidity on hand basically to cover anything that can happen because we positioned all of our collateral in the optimal spots. Are you talking about physical spots? Because you're talking about like wet signature documents. Yeah. So the only place you can borrow against CNI loans in effect is the discount window.

45:50So, OK, so they're going to go over there. That's wet signature. I can borrow against securities in the repo market, at the home loan, or the Fed. So this is kind of the confusion that happened. Where was stuff pledged with Silicon Valley? I have$50 billion of cash sitting on the Fed's excess reserve book. So that's instant liquidity. And I own a whole bunch of treasuries that I can sell if I had to. So you just – you figure out how do I optimally earn whatever I can on these assets I'm pre-positioning but get them in the right place so if I need liquidity in an emergency, I can get it. Silicon Valley did not do that and the Fed did not make it easy for them to do that.

46:32And I'm just trying to be clear. When you say get them in the right place, you mean like a physical place? Some of it's physical like the loans, the wet signature loans. I literally think we take them down to like a vault and spin a dial with the big crank and they get locked in there. But no, securities by QSIP, you would say like this batch of QSIPs is prepositioned with the Fed. So I just know where they are. I know the balances. I know I can send them right over Fedwire and draw on that. So it's that thing. One thing I always wondered with the discount window. So part of the reluctance historically to tap it on the part of banks has been they're worried it'll get out into the market, right?

47:10And there'll be rumors that they're tapping the market. How does that actually happen? I don't understand how that info leaks. So in theory, because let's imagine somebody drew on the Fed. So the discount window is regional Fed. Let's assume somebody took down$100 billion or even$20 billion from the discount window through the Cleveland Fed. It can only be PNC. So the people can figure out who it is. So why are you doing that? And then I got a story and I got to explain it. Where I think this should go. So go back to the original where I said Fed funds is dead. Nobody trades it. We ought to just trade the discount window.

47:45The discount rate ought to be I'm just lending and borrowing all day long. I can leave money excess reserves if I have money to give. And I can borrow money discount window when I need money. And there ought to be like a five basis point bid offer on that. You'd put more liquidity into the system here to grow the economy if you did that in a great way. But it's like they wiped out LIBOR. They wiped out Fed funds through all their LCR. You know, now you have the discount rate. Just bid offer on the discount rate. And then all of a sudden, discount window becomes right-way transaction. It's happening every day.

48:20Yeah, the stigma goes away. Yeah. So you have to forgive me because there's a lot going on in the world these days. So I can't pay attention to everything. I have no idea where you stand on yield-bearing stablecoins, which I understand to be a big thing of debate in the sector. And I'm, again, I've been paying attention to other things, but I see headlines about this pop up. Yeah. So Stablecoin Legislation Genius Act comes out. They left a little loophole in there. They said no interest, but it kind of got run around by offering rewards, promos. Same thing. Clarity Act is what's being debated right now.

48:59They're trying to close that loophole, they being kind of banks and people around banks, keeping the loophole open as crypto. Why not? If you can pay interest, it's going to drive volume. I want to pay interest. Banks say, I don't want you to pay interest because you're going to potentially suck deposits out of the system and you're messing around with a capital system that's worked for 250 years. You sure you want to mess with it? At the end of the day, I think the following. Stablecoin has been built, launched, marketed as a payment mechanism. It's somehow going, by the way, I'm not sure it will, but it's going to make payments easier, faster, cheaper, whatever.

49:39Not as an investment vehicle. An investment vehicle is a different animal and needs different regulation. So all of a sudden, like a stable coin that's buying T-bills that pays you interest, that looks an awful lot like a money fund to me. And that's fine. If it's a money fund, then regulate it like a money fund. Can't break the buck. You've got to have this much liquidity on hand. all the disclosure rules that go around with it, then do what you want with it. Go ahead and pay people in money funds. I just think they're mixing things. And I think my uninformed worry is we're not done blowing up in crypto yet.

50:19It's in fits and starts as you learn these things. Why do we need to start everything at once? Let's figure some stuff out. if it works and there's a logic to doing it, then go ahead and do it later. I don't know that you need to do it all at once. What's the rush? Yeah. You know, we are coming up to the Super Bowl. And when I think of the Super Bowl, I always think of the Basil Endgame proposals and that ad that they ran in 2024. It was surreal, wasn't it? It was a very effective marketing tool, at least for me. And we are expecting a re-proposal of the endgame. What's your sense of what might change between the original document or suggestions versus now?

51:04I have no inside information on that. But the original proposal that was written was in contempt of math. They basically, they had a goal of raising capital, and then they used bad math to cause that outcome, which was a lot of what the outrage was over because if you start making up rules in capital allocation, you inadvertently cause risk to emerge in places that shouldn't. So I think the math is fixed. The lawsuit on the stress test and the fact the models need to be public, so the math is going to be a lot better on what they measure. You've heard about indexing G-SIB scores to inflation and some of the boundaries.

51:48countries. I don't know what they're going to do on operating risk capital. It's been a fight. I think they're going to give some credit on credit risk ratings that Europe gives that we don't. And my guess is there'll be something in there for everybody. And then everybody will collectively be disappointed about something. They're not going to give away the shop, but they're going to try to make it directionally correct and based on real math instead of made up stuff. I want to go back to something you said about the bank integration, and you actually gave a very rare, I would say, actual 2026 application of generative AI implementation where you mentioned that the person at the local bank branch can actually ask a question about the policies and learn.

52:35So that wasn't just a hypothetical. That's a real thing that's happening right now. Yeah, that's true. Can you talk – this is a theme that's been – we've been talking about on the podcast lately, which is, you know, not just AI per se or implementing – we're going to, like, implement AI. But, like, you know, the changing relationship between a company and their software vendors and, you know, does AI change some of these relationships? Do you have leverage because there are certain, like, niche functions that you could, like, build yourself? Can you talk a little bit about how you're seeing some of these things play out within your firm?

53:07Yeah, this is fun because there's so much bad information about that. So start with the basics. In order to utilize AI, you need clean data, defined data with a single source of truth. We've spent 10 years doing that and a fortune. So, you know, silly as that sounds, think how many different places inside of a company you might have, you know, A checking account balance that's then morphed into the average checking account balance over seven days or the median or the this or the that. And so when someone's trying to pull data to run a report or build a model, they don't even know what they're pulling.

53:45So you've got to get it clean at the base source and an index. Now you have the capacity to build a model, whether it's large language or analytical. Large language models, there's open models out there you can just use, which is what we tend to do to build our document sorting. we're using it on our own data, our own documents. I don't need their data and I don't need to retrain their model per se, other than running it on my own data, which is a much smaller compute factory, right? I don't have the whole internet downloaded. I just got this thing, right? And then once we train the model, running the model doesn't take a lot of compute.

54:20Training the model takes a lot of compute. When we implement talk about AI, some of it's the fancy object thing. Oh, I'm going to implement CoPilot and now people can write emails for me. Everybody knows when CoPilot wrote my email because it's way too polite and the punctuation is correct. Does that make us any money? Probably not. But it's cool. So when you hear most people saying, oh, we're using AI, they just plugged in CoPilot and paid somebody 20 million bucks a year for licenses. Where it starts to make a difference is when it can answer questions in a hurry for our employees or our clients and where it can automate process the same way we've been on this automation journey for 20 years now, right?

55:04Starting with internet and then kind of microservices and applications. AI is just the next stage of that. You know, we've got, we over the last 10 years probably get 30 points of productivity efficiency in our operating environments. Probably the last seven years, actually. AI is going to give us another 30. And it's going to come through time. But think of it as automation. All it is is, even when people talk about AI, they'll say, oh, you know, it's an AI model. It's not an AI model. It's just a decision tree. You know, it's automating a script and an answer, moving data in a way where somebody doesn't have to swivel, you know, pretty basic stuff.

55:44What about on underwriting? Would you use AI to automate that particular process? It's actually not legally allowed today. um so there there's i mean it's truth in lending or something you know one of these laws where if i tell you no you applied for your credit card they just kept my rates at 10 so i said no i've got to give you the three reasons you know of why i said no with an ai model we look at a thousand variables and the weight of each of those variables is dependent upon their interaction in your particular profile. So the weight of rent might be very heavy for one person and not for another based on the profile.

56:26I can't give you three reasons. The model told me no. The model told me yes. So what ends up happening today in credit decisions, this is crazy, but it works, is we'll set with our normal criteria, this FICO, this debt to income, whatever else you look at, we'll say yes or no. And then if we tell you no, we use AI to try to tell you yes. Because I can still use my original excuse as no, but if I turn you to say yes, after the fact, you're not mad at me. So the point is - It's crazy. But the point is that legally speaking, if you say no, you have to be able to articulate a reason. Yeah, so what ends up happening is everybody sets this bar way too high.

57:11This is one of these silly rules that needs to be looked at. I mean, we do use, it's not so much AI. There's a lot of decision tree models on automation that help with underwriting, but it's not necessarily a machine learning application. Tracy and I, I mean, I get this idea of like AI as another form of workflow automation, very intuitive. Yes. Tracy and I used to publish transcripts of the podcast, but it was a very labor intensive process to like clean it up, et cetera. And so I thought like AI was going to be the panacea. It's like, clean this text. It just, when we were doing it, it wasn't there yet.

57:47It was like creating fake, it was creating sentences that weren't there. But I understand this intuition that AI can then be like trained on a repeatable process. Right now, and again, the things that we call AI, which these days, large language models, those workflow automations, like, are there certain workflow automations that today, workflow processes, that you could say, we've been able to automate them in a way we couldn't have five years ago. Oh, absolutely. But let me just give you kind of some scope and some numbers and what's actually happening. And we have a pretty clean tech stack, so maybe we're a little further ahead at this than some.

58:25But we focused on a billion four of addressable spend inside of our care centers and our operating environments. This is just solving problems for customers, moving loan, all that stuff. A billion four we spend. And we've come up with 171 different ways to use AI in reducing that spend. And the addressable piece of the billion four, we think is probably 40 % of it. Of the 171 use cases, we prioritize five that are live. And so what does one of those look like, for example? So imagine inside of our wealth business, we've got a bunch of trust lawyers. So my grandfather, it's actually with PNC. My grandfather created this small little trust, Skip Generations, and it's written.

59:12It's beautiful. It's handwritten on like three pages. Maybe it was my great grandfather. Some lawyer every month has to look at that thing and say, you know, this qualifies or doesn't. We're going to do this distribution. All of those things are manual. Now just read them into the document writer. And all of the things that are mechanical outcomes from those millions of pages of trust documents are right in front of you. Right? payment on this date, here are the beneficiaries. It just makes it real simple. So document reading, expression of what's in documents on the basic stuff, you know, now I don't have to just go and search and find and human error.

59:51That stuff will make a difference. I worry we have too much focus on the cool fun toy, chat GTP. I mean, this is, you know, it's a new Google search. It's cool. Give me the answer. That isn't going to make the world better if that's all it's used for. How did you get a clean tech stack? Because this is something that also goes hand in hand with bank consolidation. You always hear that banks end up with this tangled mess of technological systems, maybe some that include COBOL or like old languages and things like that. COBOL? COBOL. I can never remember. You weren't even born yet when we were programming in COBOL.

1:00:29So how did you manage to do that? Honest answer, right in the middle of the financial crisis, right? We merged with National City under duress from both firms, kind of a regulatory wedding. Neither of us had the capacity to run a larger firm. I mean, I distinctly remember, you know, talking to the combined organization. We have to rethink how we run something. You can't manage by walking around anymore. and neither tech stack could survive on its own. And so we saved less money than we said we would in the announcement, or actually we saved what we said we would save, but quietly we reinvested$2 billion a year for 10 years.

1:01:16Wow. Yeah, so literally all new, we went from single server stack, so this computer runs that application and each computer is different because whoever the application owner was went to dinner with Dell that day or HP the next day. We had 11 data centers. None of it was in a virtual environment. So building our own first cloud environment and then secondly, writing everything in cloud language in a microservice application. So it's cloud native. That's forever. But we spent a ton of dough and just have invested, invested, invested and got to a place where our tech stack is as modern as any fintech you'll find out there.

1:01:59All right. This is fun. This is a lot of fun. Thank you so much for taking the time. We had a lot of questions. Thank you for being interested. Yeah. That was great. That was great. Really appreciate you coming out on AdLaz. Good to be here. Thank you.

1:02:22Tracy, I really enjoyed that conversation. I learned a lot about banking in that conversation. We hit a lot in that. But I just wanted to keep going. There were so many interesting angles. Bill actually answered the questions. Which cannot be said for every CEO. No. I would say most of our guests are very good at answering questions. Not every CEO is great at always answering questions. But I was like, oh, I'm learning something with each one of these things. Let's keep going. Also, genuinely, he seemed to have a good handle on like the granular and practical aspects of running a bank. So I was really interested to hear how they redesigned their tech after the financial crisis.

1:02:57I guess they got a head start on everyone else. And then also the pre-positioning of collateral at the discount window. I knew that the Fed had been encouraging banks to do it. I did not realize that you had to send the actual mortgage docs if you want to borrow against those. It is crazy how much is not just done automatically electronically, right? And here was, I didn't ask it, but I was going to ask, does blockchain solve this question? But the idea that like, okay, all of these things really are electronic-ish, although there's clearly this physical component. But the idea that here is something which is theoretically monetizable, which has liquid value, is a liquid instrument, but because of the way that the system is structured cannot actually be liquidated or turned into liquid assets immediately if it were not prepositioned, et cetera, that actually does still seem a little bit wild to me that that is the case.

1:03:58I always thought that mortgage assignations were like one use case for blockchain, right? Right. Yeah. Yeah. Yeah. You could reasonably see blockchain, you know, helping out with that problem versus, I don't know, tracking the origins of lettuce and things like that always seemed kind of ridiculous. Well, it seems like one of these things which is in banking and again, is like integrated databases. So there's like a be a central bank database. There will be a home loan bank database and the regional bank and the bank database. Right. And the question is, like, why can't they just all see? Talk to each other.

1:04:32Yeah, or see the relevant parts of the database, right? And I suppose that's easier said than done. I really like hearing him talk, too, about practical applications of AI, which, again, maybe feels a little different than some of our conversations last year. It feels like now there's more articulated examples of this is a sort of labor-saving or time-saving technology. So that's another thing I think we have to pay more attention to in 2026. Also, the database issue in the context of AI is kind of interesting because we were talking about how so much of programming nowadays, so much of the tech issues are those coordination problems.

1:05:11Yeah. Right. And so even if we have something like Claude, we recorded that episode recently. Even if you have Claude writing software, it doesn't necessarily solve that coordination issue. Totally. And then, you know, bank branches, how crucial they are. Oh, yes. That was surprising. I had heard that, you know, there's – and again, it may be just like one of those sort of things people say. Like people are less likely to switch banks than they are to get a divorce. And so, therefore, it still makes sense to build out a physical branch. But like the strategy of that and thinking about where and how the fact that they're building hundreds of branches still, even if the total number of bank branches is actually declining, That's just like a super interesting, real part of the business.

1:05:57It is true that I've had the same bank since college. I don't know about you. And I hate them and I still won't leave. But since I moved to New York, I think I've had, so that's 21 years. So basically, yeah, one bank. It's kind of wild. I'm going to switch. I'm going to prove the anecdote wrong. I'm just going to switch for the hell of it. Yeah, absolutely. Okay, shall we leave it there? Yeah, let's leave it there. This has been another episode of the Odd Thoughts podcast. I'm Tracy Alloway. You can follow me at Tracy Alloway. And I'm Joe Weisenthal. You can follow me at The Stalwart. Follow our producers, Carmen Rodriguez at Carmen Armand, Dashiell Bennett at Dashpot, and Kale Brooks at Kale Brooks.

1:06:36And for more OddLots content, go to Bloomberg.com slash OddLots for the daily newsletter and all of our episodes. And you can chat about all of these topics 24-7 in our Discord, discord.gg slash OddLots. And if you enjoy Odd Lots, if you like it when we look into the business models of banks, then please leave us a positive review on your favorite podcast platform. And remember, if you are a Bloomberg subscriber, you can listen to all of our episodes absolutely ad free. All you need to do is find the Bloomberg channel on Apple Podcasts and follow the instructions there. Thanks for listening.

1:07:21Thank you.

1:07:48Held in conjunction with our official partner, Qatar's Ministry of Commerce and Industry, and premier sponsor, Qatar Investment Authority, the forum will convene world leaders and international CEOs who will provide unique perspectives on the critical issues facing the global economy. Request your invitation at katareconomicforum.com.