Transcript:
Transcripts are generated using a combination of speech recognition software and human transcribers, and may contain errors. Please check the corresponding audio for the authoritative record.
Penny Crosman (00:18):
Welcome everyone to this event. I want to introduce two special guests that we have tonight. At my far left here we have Luis Valdich, who is managing director of the Venture Capital group at Citi Ventures in New York and London. He's kind of social distancing himself because he's got a little bit of a cold coming on. He and his group do a lot of investing in fintechs. Some of the companies he's invested in include Socure, HighRadius, iCapital, Feedzai, SmartAsset, and Capitolis. Before this, he had a similar role at JPMorgan Chase, where he was head of strategic fintech investments. His fintech investments there included Block, Markit and Tradeweb. And we have Alex Sion, who is managing director of Motive Partners, which is the largest private equity firm investing in fintechs. And he originally was one of the founders of Moven, which was one of the original neobanks. And then JPMorgan Chase hired him to be its general manager of all mobile banking when mobile banking was in its sort of heyday. And he also worked at Citi Ventures for about a year and a half, and he and Luis overlapped. So again, welcome to everybody. So let's start with you, Alex. What do you see as the biggest opportunities for AI in banking and in fintech?
Alex Sion (02:09):
Thank you Penny. I think AI unequivocally is another inflection point technology. I think mobile was a big one that really transformed the way banking worked. And then there were several [transformations] within payments over the past 10 years. And then obviously blockchain and crypto was another one. I think a lot of people are looking at AI as another one and perhaps as profound as the internet itself when it comes down to what does it do for the world. Within the banking sector, it's a little trickier because obviously there could be profound implications in many industries with AI, but banking fundamentally is going to struggle with AI just dead stop because of just regulatory risk, compliance data, all the things that are just really tough for banks to get their arms around, just so happen to be the things that are necessary to really kind of unlock the power of AI. So even though there's a massive set of opportunities and early use cases coming up, most of them focused on internal efficiencies and sort of customer service, maybe in a very light way, it's going to be a while for before the industry is able to practically action things that are at scale.
Penny Crosman (03:42):
And just quickly, do you think banks are being too cautious or are they moving too quickly or something in between?
Alex Sion (03:49):
I think in certain spaces, the banking services that rely heavily on human interaction and require massive scale to drive personalized experiences should probably think of AI as significant risk and then should be investing in AI, at least within venture land or at least researching it because it's kind of more existential for that kind of interaction.
Penny Crosman (04:23):
And so Luis, just starting with more traditional forms of AI, we'll talk about generative AI in a few minutes, but when you think about traditional AI, machine learning, deep learning, neural networks, the things that have been around for a while, can you say a little bit about how Citi has been using some of that already?
Luis Valdich (04:44):
Sure. So we've been engaging within Citi with AI, and I was going to talk a little bit about some gen AI related use case, but I'll put them for later based on the context of your question, but in multiple ways. So since probably the beginning of Citi Ventures, which is more than 10 years now, we have been actively investing in companies that either have AI as foundational from an infrastructure standpoint or companies that are applying machine learning and other forms of it, including several of the examples that you gave about companies in my portfolio, the Socures, the HighRadiuses, the Feedzais for sure would be in that bucket.
(05:40):
Additionally, Citi has innovation labs that have been very actively working on AI also for many years. Within the innovation labs, there is an AI center of excellence, which has a dedicated group of folks that are very actively working on it and in terms of the multiple businesses that use models and quantitative research, and there's many different flavors of that that are being used across the banking industry. And there's very rigorous controls, risk management procedures, and the sort to ensure that you look for all types of potential unintended consequences to ensure that it's understandable what the model's doing, that it's really working as intended and from a safety and sound standpoint, but multiple uses within Citi when you define it so broadly, a lot of employees are quants, data scientists, and they've been working on different various forms of AI for a long time.
Penny Crosman (07:15):
I'm sure. So Janet, you just completed some research and one of the questions you asked bankers how they think AI will help their business. What were some of the top things that they said?
Janet King, vice president of research, Arizent (07:26):
Yeah, it's interesting because this research is literally hot off the press. We just wrapped up data collection across all seven of our communities here at Arizent. Last week we had close to a thousand people take the survey and 130 in the banking community alone. And a third of those folks were the biggest banks, and six and 10 were senior leaders or executive leaders. So I was excited to get the perspective of those people, but to answer your specific question about what were they expecting AI to do, and again, we asked that broadly like AI, in all of its iterations, how do you think it will benefit the business? And most bankers think that AI is going to have some benefit for their business, at least everyone could pick something out of it. But this early sentiment is really around kind of what you spoke to Alex, it really will help them operate more efficiently with fewer resources, that it will help them accelerate automation and that it will be something that they can use to protect against fraud. So some of the other things that they talked about are things like improved employee productivity, lowering their cost and supporting risk assessment and compliance monitoring. So there's a lot of things that they're looking to AI potentially to help them with.
Penny Crosman (08:50):
Alex brought up some of the challenges and risks involved with AI, especially advanced AI. Luis, when you think about the challenges and risks, what are the top things that come to mind?
Luis Valdich (09:04):
Yeah, one is around the ethical implications, biases, for example, inadvertent discrimination associated with it. There's lots of questions with respect to the accuracy, the very well-reported instances of hallucinations. There's issues that have to do with, and this comes with lots of new technologies, which is how will the bad guys take advantage of that? So cyber attacks, the opportunity to open a new set of doors around vulnerability. There's probably issues around the accuracy of information. We're talking earlier about fake news out there, or issues with infringing on copyright and the sort. So I think there are a lot of significant issues with the technology
Penny Crosman (10:22):
For sure. And Janet, you also asked bankers how AI might hurt their business and what did they say?
Janet King (10:29):
We saw a lot of those same things, right? So we asked about it with AI holistically, and then more specifically to gen AI and AI at large. It was things like introduction of new ethical concerns or biases. The top one actually was loss of personal touch with their customers, which I think is really interesting because customer experience was also an area where at least half of those people that we talked to said that they think AI will be beneficial for their business, but there's this corresponding concern that it could cause them to lose touch with their customers. They're also concerned about a reduction in critical thinking or analytical skills across their workforce. So skills degradation was a top one, and there were some folks that were expressing concern about job losses. I know we're going to talk about that, but it was about half and 44% who think that they're going to potentially have some loss of customer trust, and that's just AI at a high level. If you look at gen AI specifically, it's that misinformation, right? The hallucinations, the nonsensical kind of information. And then concerns about being able to support the algorithms or explain the algorithms for auditors.
Penny Crosman (11:39):
That point about skills degradation is interesting. Are tools like ChatGPT making us dumber?
Luis Valdich (11:48):
I would take the opposite side of that argument, and maybe it's the optimist in me, but I think that there is an opportunity to leverage these tools to address more mundane aspects such as summarizing significant information, distilling important insights from it, and then use critical judgment and higher level thinking to take that distill information into second order implications that in order to get to those points, you would need to do a lot of the harnessing of all the data and summarizing it. I cannot think of it similarly to other software data analytics tools that do great jobs at charting information report and allowing you to see it. Of course, there's an embedded assumption, which is the information is accurate, validated, transparent, but if the tool is working as intended, I actually think that it would require upskilling and higher level skills than the reverse. But maybe I'm an optimist.
Alex Sion (13:13):
No, I'd build on that, and I 100% agree with Luis. To me, one of the first practical use cases, which even Motive is exploring, is the creation of investment memos in the private equity world because it's a powerful tool to be able to assemble, collate, and create a narrative around something that then someone with experience and pattern recognition can quickly pick up and do something with. And it takes all of all the things that Luis said, the mundane tasks of just kind of surging through everything and can put it in a coherent, like I said, a humanistic form. So to me, that becomes extremely powerful. But the implication is I don't know what our analysts going to do now. It'll favor folks like Luis and I with a bit more gray hair because it's very powerful, but it needs to be mobilized by then other people who know we could do something with it.
Luis Valdich (14:19):
By the way, think about the comment and saying that I have gray hair implies I have hair. So that is in itself a compliment.
Penny Crosman (14:30):
Well, it's interesting, these use cases where generative AI is creating the first draft. It's looking at everything. It's looking at contracts and pulling out the most important points. It's taking meeting notes and coming up with your takeaways. But do you think that there's a danger in people relying too much in things like that and things kind of slipping in there that nobody caught because it sounded reasonable, it seemed plausible. And if you're not really thinking about it carefully, it could be like the lawyer who submitted fake case references in a brief to a court. Do you think it's going to take much more of a compliance mindset or detail-oriented mindset to kind of cope with this output?
Luis Valdich (15:24):
Yeah, I think fair pushback. I think creating an awareness to be vigilant about all these topics would force everyone to up their game. And I would make the analogy that if a junior member of my team goes to a meeting, reports back, we meet with lots of startups, some I'll have the opportunity to meet with, and then comes back with summaries and talks about it, we need to apply judgment and does that make sense? And by poking holes, you are in fact getting to those higher order types of things, and it's part of the apprenticeship, but it's also part of the quality control. So I think that so long as individuals are very much aware about the dangers of just sitting back and assuming that all that gets produced is sufficient, then I do think that human judgment and ingenuity will continue to progress and will help us get to even higher levels of insight.
Penny Crosman (17:00):
So there could be a chief hole poker title out there.
Luis Valdich (17:04):
Well, we all should play those roles, right?
Alex Sion (17:07):
Absolutely. It does bring up the integrity of the data becomes absolutely critical because I feel like the rest of it, the sophistication of how it can assemble information, the narrative, even what it sort of deems as important, not important, those are things that I feel like are inevitably going to be figured out at a very rapid pace just based upon how the technology itself is evolving and just the fact that it's learning constantly. But it's garbage in, garbage out. So if you don't have the right foundation to begin with and those trusted sources, then it doesn't matter how well the narrative can be assembled on the top, if the primary sources are bad, then the problem gets exacerbated. So I feel like there's got to be a lot more investment in that, the underlying trust or the quality of the data.
Penny Crosman (18:03):
That's a very good point, and we're going to have a series of articles on getting the data right for just that reason. So just to finish out the conversation about how is AI and advanced AI changing jobs, Jamie Dimon said generative AI might bring about a three and a half day workweek for younger workers. And I would say some of us take that a little skeptically, but what do you guys think about that? And are there any other ways you see jobs changing throughout this industry, in fintech and banking due to generative AI?
Alex Sion (18:48):
I think there's already in the service center world and a lot of the research analyst world, and it's already having an impact, if not on current staffing, but certainly future projected staffing, especially as Outlook starts to get some of these tools get embedded into the core software, you think naturally about a leaner model for anything that's more operational processing, assembly oriented. So I'm almost certain that's in people's HR plans. A good example, every portfolio company we invest in, we're undergoing a risk analysis of just basic, what's your thinking on AI? How are you thinking about deploying it? Because we just know it's going to be a significant thing. So I feel like some of that is already baked in, but where it goes, I think is up to the data and the technology itself, but some of the more mundane operational tasks, I'm sure, even software development and that how a staff software development teams, I know GitHub is kind of all over this, and if they're all over this, then it just cascades.
Penny Crosman (20:10):
Now you guys are experiencing that with your programmers using GitHub.
Luis Valdich (20:16):
Yeah. No, indeed. I'll talk about that. But also I want to come back with another point on the broader topic, and indeed when you qualified your previous questions to what Citi is doing, but you can talk about gen AI. You stole a little bit of my thunder there, but that's okay. One of the things that we have been doing has been experimenting with run experiments first just using with a small set of people, just standard ChatGPT with public data to see how it works, what are the implications, opportunities, and the sort and learn from that. And then there was another set of experiments with pilots using LLMs, including Citi data to think about Citi specific use cases. And exactly as you said, Penny, very well informed. As a result of those successful pilots, there's this plan to deploy some of those co-pilot tools across 40,000 developers within Citi.
(21:24):
So it's happening now, coming back to the previous comment about what happens with some types of jobs and the like. I think, and we were talking about that earlier, right across human history, whenever there were new inventions there were always worries about what happened to jobs, and the analogy I was talking about earlier was washing clothes, there were armies of people that were doing that manually. And then when the washing machine was invented, lots of jobs were lost. And yet when you look at employment over time, certainly folks found other types of activities probably more fun than washing clothes all day long for a living. And then similarly, I think that the use of AI in financial services, in banking will help power a very important trend that we've been watching and we've been investing across, which is the democratization of the highest order products that exist out there that requires significant human involvement, and which are really difficult to bring to not the largest 21, but the largest institutions or the ultra high net worth individuals and the like with the aid of automation technologies of AI and generative AI and many of these other tools, you could see that marginal cost to serve to come down, and therefore you could see opportunities to then bring these terrific products to the next tier and so forth.
(23:24):
So I am, as you can tell, a bit more glass half full on this.
Penny Crosman (23:33):
That's why we invited you. So Janet, you asked our readers how they think AI will affect jobs in the industry. What were some of the things they said?
Janet King (23:45):
Yeah, it's interesting. So I think I mentioned earlier that close to half when you asked them about concerns, put it on their laundry list of concerns, but when you ask them more directly, is it going to affect the nature of jobs or the number of jobs? Most bankers feel like it's going to change the nature of jobs, but won't have any major impact on the size of the workforce. I think something like 40% said it wouldn't have a major downsizing, but another almost equal amount, another 36% said that it would substantially create new jobs so that it would lead to job creation. So that was especially true at larger banks. One really interesting finding from the research was that we asked about hiring AI proficient employees and whether banks were willing to pay a premium for that kind of talent. And what we saw was something like, I think it was around 28% of banks said that they would be, and a lot were still undecided, but when you compare that to fintechs or insurance carriers, closer to 50% of those respondents from those sectors are paying more, paying a premium for AI fluent employees.
(24:58):
So that's just I think something to think about, particularly as the larger banks are trying to compete with those organizations for talent. I thought that was interesting,
Penny Crossman (25:08):
Janet, you also asked what are the top use cases for ai? What were some of those responses?
Janet King (25:15):
I mean, it was very consistent with what you guys have been talking about. It's mostly improving marketing communications and emails, especially when you're talking about generative AI. So it's about content creation, it's about marketing, it's about general office productivity, as you said, taking notes and that kind of thing. They're also looking at it to be used in the contact center to answer routine questions or to generate those kinds of routine informational reports. But we also are seeing a fair number who are helping using it to help developers generate code. So a lot of what ground you've already covered, frankly.
Penny Crosman (25:53):
And some of the companies here are finyech startups. Can both of you say a little bit about what you look for when you think about a company you might want to invest in or partner with? We'll start with you.
Luis Valdich (26:11):
Yeah, part it relates to AI. I think that we've talked about many of the challenges with AI. So think about ways in which you are mitigating those risks and protecting from all those risks. So the more you can address all those issues from a transparency, scalability, data quality, et cetera, issues. And then the other topic, which is very important is that also you think about, and this applies not only to AI focused startups, but that if your ultimate goal is to sell to large financial institutions as opposed to if your goal is selling to other fintechs, that having very robust processes being highly scalable, scoring very highly in terms of safety, soundness will also become very important just because the stakes are different.
Alex Sion (27:30):
I'll just build on that by saying fundamentally, you have a lot of generic and not generic specialist technologists in the AI space that are looking for financial services as an opportunity in the market. It's very big. But the big question I'd say is we try to look for, do they know what they're getting into? Do they actually understand financial services or are they just looking at it as a market opportunity? Do they understand the risk compliance data challenges? Are they just kind of naively thinking that it's such a big market and we can go into it? I think the go-to-market strategy for AI firms, it's got to be very carefully thought out because you could go lots of hype, but not much traction that could be meaningful at the scale that will be meaningful for venture investors. So you could be setting yourself up for a three year long sales mission into any given bank because they can't action anything. So do they know, do they understand the world that they're entering into or is it just too much hype.
Penny Crosman (28:48):
You need a lot of patience for a three year sales cycle. And Luis, you mentioned that some projects you do, you've got your own innovation lab, you ideate and curate ideas internally, and then you often invest and partner with fintechs. How do you make those decisions about which things you're going to do in in-house and which things you're going to find a partner for? Are there certain kinds of problems where you look for an external help or there certain types of projects that lend themselves more to in-house versus outside?
Luis Valdich (29:25):
Yeah, it's a good question. And frankly, the partnership decisions are really driven by the businesses within Citi and it will depend on a variety of questions, including what is the nature of their environment, what type of tech book of work budget they have to bring someone new, what are the objectives, et cetera. Within our side, within Citi Ventures specifically, we're mainly looking for startups that are driving exciting innovation, which we think have the potential to be category leaders in different spaces that are relevant to Citi, maybe in the near term, maybe in the long term, but we're trying to with that, bring some of the outside in and help also some of those startups navigate Citi and find eventually potential partner sponsors, sometimes in multi-year processes, as Alex is mentioning. But yeah, I think it really depends on the situation.
Penny Crosman (30:43):
So let's just close with a question about using AI in customer interactions, because I think you both mentioned that as a pretty strong use case, but then I think Janet, you also mentioned that that's a concern. Are we going to lose contact? Are we going to lose that relationship by automating too much? What do you guys think about that balance between automating the things you need to, but not just casting people to the wind, making sure you keep the customer relationship strong? Sure. Go ahead.
Luis Valdich (31:20):
No, no. I was asking Alex.
Alex Sion (31:21):
Yeah, no, I can start. I'd say I agree that you need that personal touch but scale. I think if we admit to ourselves that financial services is not personal, it's very commoditized and inboxes because scale is difficult and it gets to Luis's point, the democratization, it's not that democratized. If you're at the top, top end, you get incredibly personalized experiences and then it just falls way off a cliff after that. I think the opportunity is to scale that personalization through AI and just that combination of the, take out the remote, the rote tasks, assemble the right information, and then allow somebody to give it that personal touch at the end is the thing that will help everybody. It'll help customers, it'll help the firms, it'll engage people more in their money, and I think you can just see that scale potential happening with AI.
Luis Valdich (32:27):
To add to that, I would just posit the example that I'm sure many of us have experienced when dealing with call centers and being handed off multiple times in order to solve some more complex problem and then having to explain again to a human because somehow it's being lost in translation and then despite leaving comments and notes calling again because the problem hasn't been solved, and then having to do that over again. The frustration that I think we've all experienced on that is pretty significant. And the promise of having tools that can enable the first agent to connect the dots and help you solve that the first time and to get it done, it feels to me as such a powerful and positive dynamic, and not only for the user, but also for the customer service agent. Sure, some of them get yelled at by frustrated people that are calling the third time about the same thing. So I think that all these things, there are both potential pros and cons and those that really harness that balance, most astutely will be the ones to win.
Penny Crosman (33:55):
That's a great point. Well, I feel like I learned a lot tonight, so thank you very much all of you for participating.