# NLP in Banking, in the Back-End and Front-End - With Lee Smallwood of Citi

- **Show:** AI in Financial Services
- **Published:** 6 Jan 2020
- **Source:** https://aiandbanking.libsyn.com/nlp-in-banking-in-the-back-end-and-front-end-with-lee-smallwood-of-citi
- **Transcript:** https://podcasts.opennash.com/shows/ai-in-financial-services/nlp-in-banking-in-the-back-end-and-front-end-with-lee-smallwood-of-citi-c8ef7a/
- **Provider:** OpenNash Podcasts (machine-generated)

## In brief

Natural language processing (NLP), software that reads and interprets human text, will enter banking first through regulatory risk reduction, not revenue growth. Lee Smallwood of Citi argues that contracts and trader chat messages hold huge amounts of text, and that compliance failures now carry steep penalties. Banks therefore have strong reasons to adopt NLP for document management and communications monitoring before they pursue other uses.

The central tension is between the stick and the carrot. Dan Faggella frames compliance as the only proven driver of adoption, while Smallwood agrees but calls it a wedge that builds comfort with new technology and startups. He expects revenue-focused uses within two to five years. He also points to voice transcription and customer privacy as the next frontier and the next obstacle.

## Key insights

1. **Compliance fear drives bank NLP adoption** Both speakers say vendors win by promising to reduce regulatory downside, not by promising efficiency. Smallwood says the main worry inside big banks is regulatory penalty, so tools that make compliance easier and more accurate get attention first.
2. **LIBOR replacement creates a huge contract-editing burden** Millions of derivative contracts reference LIBOR, a benchmark interest rate being replaced by SOFR, and were never written for that change. Smallwood sees this as a strong case for NLP to find and update contract language, and expects more such regulatory changes.
3. **Bloomberg chat holds valuable unstructured trading data** Most sales and trading communication runs through Bloomberg Instant Messenger. Smallwood says Bloomberg is building NLP software for products like municipal bonds to show how many clients requested a bond and what prices they received.
4. **NLP can shrink compliance chat review** Compliance teams already filter millions of chat lines through many layers. Smallwood and Faggella say NLP could flag a smaller curated set for human experts, even before it can identify specific misconduct.
5. **Voice transcription is the next frontier** Smallwood is excited about turning speech into text and then into action, such as speaking a trade booking aloud. He cautions that spoken language differs from written language, so transcription must improve and NLP models must adapt.

## Questions this episode answers

### Where does natural language processing fit best in banking right now?

Smallwood points to two areas: the huge volume of client contracts, especially in sales and trading, and instant-messaging chats such as Bloomberg. Both hold large amounts of text and carry compliance risk.

### Why are banks so focused on regulation and compliance?

Smallwood says that after the 2008 financial crisis, regulators moved from understanding the industry to governing it, with steep fines for non-compliance. He says the downside of poor behavior now matters greatly to the bottom line.

### What is the LIBOR to SOFR change and why does it matter for NLP?

LIBOR is an interest rate benchmark that many derivative contracts rely on, and it is being replaced by SOFR. Smallwood says millions of contracts must be reworked, which makes it a strong use case for software that can read and update contract text.

### Will banks use AI chatbots on trading floors?

Smallwood says that is yet to be done and he is unsure it is the direction things will take, because trading chats are between people with deep relationships. He sees more promise in extracting useful data from existing chats.

### When will NLP help banks make money, not just avoid penalties?

Smallwood says compliance work is the wedge that gets big banks comfortable with new technology and startups. He estimates revenue-focused applications could appear in roughly two to five years rather than ten.

## Facts

- **Daniel Faggella** [0:00:03]: "Previously, he worked in the startup world where his company was acquired."
- **Lee Smallwood** [0:01:56]: "there's contracts that govern the way that banks can interact and transact with their clients."
- **Daniel Faggella** [0:03:45]: "I think after the financial crisis in 2008, there was a period of basically oversight and understanding what on earth is happening in this industry."
- **Lee Smallwood** [0:04:21]: "whereas maybe rewind 10, 15 years ago, a lot of the focus would be on massive revenue generation opportunities, and that's still the case."
- **Lee Smallwood** [0:06:17]: "there's a contract called an ISDA that governs the interaction, particularly of derivative contracts."
- **Lee Smallwood** [0:07:01]: "And think about the millions of contracts across the industry that are governed based on derivatives that have the price contingent on LIBOR."
- **Lee Smallwood** [0:07:01]: "There's a lot of working groups to figure out how the industry will transition and sort of a business and regulatory perspective."
- **Lee Smallwood** [0:08:27]: "things like, without getting too technical, margin requirements for non-centrally clear derivatives. That was a big change in the way that we go in the interaction of certain trades with certain clients"
- **Lee Smallwood** [0:10:43]: "most of the markets and sales and trading industry communication happens via Bloomberg Instant Messenger."
- **Lee Smallwood** [0:10:43]: "any trading floor you step on, you'll see people with Bloomberg terminals. When you see the black screens with the orange text"
- **Lee Smallwood** [0:12:17]: "I believe municipal bonds is a good example, where Bloomberg is starting to build NLP-based software that can tell you that this many clients have requested this type of bond"
- **Lee Smallwood** [0:13:49]: "Compliance departments in banks are generally huge. And the level of layers and number of funnels and filters that they go through in terms of analyzing chat is quite tremendous."
- **Daniel Faggella** [0:17:11]: "And yes, you and I actually did speak about this on stage at PayThink."
- **Lee Smallwood** [0:17:01]: "maybe more like two to five years than 10 years until you start to see some really interesting applications really focused on moving the top line."
- **Lee Smallwood** [0:18:19]: "the more you realize the way that we speak is not the same way that we write necessarily."
- **Lee Smallwood** [0:19:09]: "trade booking, if you could speak that into a microphone, I bought or sold this, and all the booking is done, and it goes through the clearing, then that saves a ton of efficiency"
- **Lee Smallwood** [0:20:25]: "big institutions take customer data and privacy very seriously. And so it's how much can you extract from one side of the conversation?"
- **Lee Smallwood** [0:20:25]: "can you look at the communications between a banker and a set of clients where you don't necessarily know who the client is"
- **Daniel Faggella** [0:21:39]: "Next Monday, we're going to be sticking with our theme of NLP and we're going to be talking about searching for documents and searching for data within the back end of banks."

## Transcript

**[0:00:03] Daniel Faggella:** This is Dan Faggella and you're listening to AI in Banking. This is our first episode in the year 2020 and we're kicking it off with the theme for the entire month of January, which is natural language processing applications in banking. Our first guest is Lee Smallwood. Lee is the COO of Markets and Securities at Citibank. Previously, he worked in the startup world where his company was acquired. So he has sort of big company perspective and small company perspective.

**[0:00:31] Daniel Faggella:** Lee was one of our trusted advisors in our most robust AI and banking research, which many of you who've listened to the podcast have already downloaded the executive summary. Some of you are customers of that record. And so we're grateful to Lee for that. And I wanted to make sure to bring him back in for the new year and talk about this new theme of NLP. Lee talks about NLP use cases and applications in the front and the back end of banking.

**[0:00:55] Daniel Faggella:** I think he paints a nice overall picture that will tee up the other three interviews that we're going to be going through for the rest of this month. If you haven't already downloaded our NLP use case guide, be sure to do so. If you're interested in where NLP can fit into your business, this is probably the most succinct resource for discovering exactly where NLP could have a fit within your company. It's a nice overview and a primer of NLP in business in general.

**[0:01:19] Daniel Faggella:** You can download that at emerj.com slash NLP1. That's E-M -E-R -J dot com slash N-L -P-1. So go ahead and download that PDF. Without further ado, we're going to fly into this first episode with Lee Smallwood here on AI and Banking. All right, so Lee, where we'll kick off first is just sort of your intuition around where natural language processing plays a role in banking. Are there parts of sort of the banking operations that you think are more low-hanging fruit, if you will, for natural language?

**[0:01:56] Lee Smallwood:** Yeah, for sure. So when we think about natural language processing, obviously, we're thinking about where the large swaths of text that exist within this kind of institution. And a couple of those really jumped to mind. One is in the sheer number of contracts that a bank has with its clients. And particularly on the sales and trading side, there's contracts that govern the way that banks can interact and transact with their clients. that's a ton of information and financial markets are constantly changing.

**[0:02:24] Lee Smallwood:** So editing, updating, keeping a track of those contracts, being able to change them on the fly, or even change them with regulation is incredibly important, especially given the amount of regulatory focus in the industry right now. So that's one. And then two is on the chat side, obviously, instant messaging and Bloomberg chats, things like that are a huge part of the way that the industry still communicates with each other. So using that for things, either trade identification, for data capture, or again, for compliance and making sure that people are saying, well, they should be saying on those chat mechanisms.

**[0:03:00] Lee Smallwood:** So they're really the two big areas where I see there's a lot of opportunity for NLP. Happy to dive into either of those.

**[0:03:05] Daniel Faggella:** Yeah, yeah. Well, there's a couple of things we can kind of fly into. So you mentioned something about, obviously, examining documents and making sure we're compliant. When I look from the outside about how NLP is making its way into banking, kind of the fear of being non-compliant or the fear of a slap on the wrist or some threat to your banking license, that's the big boogeyman to some degree. And you had just emphasized, given the emphasis on regulatory matters in banking today, just for context for some of the folks who might be outside of the banking world or might not exactly know why things are so heated and regulatory right now.

**[0:03:45] Daniel Faggella:** Is this more than usual? Is this just after the 2008 issues? Why is it that this is, so many people have mentioned that this is kind of a regulatory focused time? Yeah, exactly what you hit on there. I think after

**[0:03:57] Lee Smallwood:** the financial crisis in 2008, there was a period of basically oversight and understanding what on earth is happening in this industry. And I think once the regulators got their head around that, then it's been more of a case of, okay, now we're going to make sure we govern this in a way that is good for all financial market participants. And there are really steep fines and penalties in appropriate action. And again, this is not just on the bank side, this is on the customer side, on the client side, on the asset management side.

**[0:04:21] Lee Smallwood:** But the industry in general now, there are high prices to pay for a lack of adherence. And so I think, whereas maybe rewind 10, 15 years ago, a lot of the focus would be on massive revenue generation opportunities, and that's still the case. But now the downside of behavior that is not complicit with the regulators is so, so meaningful to the bottom line. So I think, yeah, it's been a combination of the change in the environment and the penalty on the downside that's really made staying on side and compliant become such a major focus and where a lot of the natural applications of AI and specifically NLP

**[0:04:56] Daniel Faggella:** make sense in terms of launch. Yeah. I mean, we just see it as part of the sales pitch for so many of the successful companies. They don't go in there talking about massive upside or even necessarily efficiencies. They go in there talking about reducing risk. That's it. And it's a really, really interesting angle. Obviously, 2008 kind of made a big difference there. So, okay. So, that sort of affects documents, document search. We'll talk about this one first, and then we can talk a bit about the chat side of things.

**[0:05:24] Daniel Faggella:** You'd mentioned maybe the more of those that a bank sort of has to manage, the more data it has to train these systems, maybe the more applicable NLP would be. When people say documents in banking, it's so horrendously open-ended, you know, what kinds of documents we're talking about. We're talking about quarterly reports, we're talking about legal contracts, we're talking about loan applications. When you think about the kinds of documents where your gut would tell you or your experience would tell you NLP is being used, What are those?

**[0:05:54] Lee Smallwood:** Yeah, and what I would say is it's very hard to get visibility on where it's actually being used today. So I'll probably talk more about where I see the opportunity for you. Yeah, yeah, yeah. Let's do that. And also where I've heard that people are trying to do different things. But the big things that jump out in terms of contracts are things and documents. The things like sort of anti-money laundering regulation, AML, things like KYC, know your client, that comes up a ton.

**[0:06:17] Lee Smallwood:** And then as it again relates to sales and trading, there's a contract called an ISDA that governs the interaction, particularly of derivative contracts. And so these are lengthy documents that determine how a bank and a client can trade with each other. And so there's a ton of information in there. They take a long time to negotiate and really facilitate the interaction of trade flow. And so I think there's massive opportunity for NLP to both speed that up and then also address changes.

**[0:06:44] Lee Smallwood:** There's some big changes happening in finance right now. I don't know if you've covered on the podcast before, but the transition from

**[0:06:49] Daniel Faggella:** LIBOR to SOFR. Oh, yeah. This exact series that you and I are chatting in will have a number of references to exactly that. So, yeah, big deal, right? I mean, regulatorily.

**[0:07:01] Lee Smallwood:** That's massive. And think about the millions of contracts across the industry that are governed based on derivatives that have the price contingent on LIBOR. And that underlying rate is changing now. That was never intended when these contracts were put together. That's a lot of work to unwind and that, you know, the industry as a whole is still figuring out. There's a lot of working groups to figure out how the industry will transition and sort of a business and regulatory perspective.

**[0:07:26] Lee Smallwood:** But obviously, there's documentation that governs us. There's a lot of change and a lot of work that needs to be done on that side. Seems to me to be a really good opportunity to actually execute on NLP and save a lot of people a lot of time.

**[0:07:39] Daniel Faggella:** Yeah. And, you know, I think it seems like the LIBOR is a really hot topic right now, this big change in sort of what so many documents are referring to. So many different kinds of documents need to be edited slightly or edited altogether or completely scrapped sometimes. And it seems as though maybe to some degree as a selling point for NLP, although that's not my job, but it might be something that the vendors use as kind of a weapon on their side here, is that this, you know, we can estimate this is not going to be the last big regulatory change of its kind that makes us have to look at every darn document we've got again, right?

**[0:08:13] Daniel Faggella:** I mean, LIBOR is maybe a representative of something that could happen three or four years from now, where we got to do the same darn process. Do you think about it that way? Do you see it more as kind of like a singular, once a decade kind of thing? What's your take? No, I think you're absolutely right. It's happening

**[0:08:27] Lee Smallwood:** more and more. I think LIBOR is the big one. I like to use that as an example, because I think more people are aware of that. But we've had stuff going on for several years in the industry, things like, without getting too technical, margin requirements for non-centrally clear derivatives. That was a big change in the way that we go in the interaction of certain trades with certain clients across the business and across the industry. And so, yeah, I think there'll be more of this.

**[0:08:50] Lee Smallwood:** There'll be more oversight. There'll be changes. I think regulators want to be able to change more on the fly in the way that the industry is governed as a whole. And so I think that ultimately is documented somewhere and it ultimately is reflected in contracts. And so, So with that in mind, there's a lot of paperwork that ultimately will need to be done on the back end.

**[0:09:10] Daniel Faggella:** Yeah, so maybe this is part of the improving agility of the space where in the future, a regulation is going to want to be more agile to deal with new technology issues, trends, risks, whatever. And banks will consequently have to be able to roll with those punches, edit, update, and adjust things in time. And obviously, there's extremes in that where it would be unreasonable. But it seems like if things continue as they are, it's only going to be more and more important to have that kind of search and discovery capability.

**[0:09:43] Lee Smallwood:** For sure. And again, to be clear, this is on both sides. You know, on the buy side, people have these contracts in place with all the banks. On the banks have these contracts in place with all of their buy side. Obviously, that seems like a larger amount of work. But in both cases, this work will have to be done. The regulators are determining where the industry is going. I think what's interesting to me is it seems like the use of NLP is something that will really give a massive advantage to those firms that employ it because it's going to save a lot of time, decrease the error rate, improve the accuracy, and provide a really good use case for what

**[0:10:13] Lee Smallwood:** is possible with AI in this business.

**[0:10:15] Daniel Faggella:** Yeah, it hasn't been super easy to integrate it. There's so many challenges there. And it's really hard to measure the ROI, right? It's like, oh, how many man hours did we reduce? How many slaps on the wrist for regulators did we avoid? It's like, oh, we really don't know. But as you had said, it's just kind of this part of this bigger trend where it seems somewhat inevitable that it'll improve. You also mentioned chat. And actually, you brought up something really quickly that I just haven't heard on the podcast yet about Bloomberg's chat or something like that.

**[0:10:41] Daniel Faggella:** What is that just for context?

**[0:10:43] Lee Smallwood:** Yeah. So just to set the scene here, most of the markets and sales and trading industry communication happens via Bloomberg Instant Messenger. Bloomberg, obviously, the big financial data company. any trading floor you step on, you'll see people with Bloomberg terminals. When you see the black screens with the orange text, the charts, prices flashing up on traders, salespeople, asset managers, screens. So that's already embedded on most platforms. And a lot of the conversation takes

**[0:11:09] Daniel Faggella:** place on the chat within that software. Got it. Okay. Yeah. So it's just a, it's a, it's a messaging, it's a means of communication. Now, have you seen any attempts at automated chat? Obviously, Facebook is also a platform. It's for the masses. Bloomberg is very much not for the masses. It's pretty isolated to your world. But in Facebook Messenger, there's these big attempts to be able to leverage some degree of conversation, if you will, in their chat.

**[0:11:34] Daniel Faggella:** Amazon has their consumer facing interfaces with Alexa, for example, where they're taking swings at conversation, although it's a little bit more like, hey, get me some AAA batteries. Are there any of those applications being built kind of on the trade floor side? Like, is there any semblance of conversational work being done through those interfaces yet? Or is that yet to be done?

**[0:11:54] Lee Smallwood:** I think that's yet to be done. And I'm not sure that's the direction that it will take either. It's really between people with deep relationships where this typically takes place. So I don't think it's that same application that there is demand for. I think where there is opportunity, and Bloomberg as a company are actually trying to solve this from what I understand, is extracting a lot of the valuable information that can exist on those Bloomberg chats.

**[0:12:17] Lee Smallwood:** So I think in some products where there are very clear identifiers, I believe municipal bonds is a good example, where Bloomberg is starting to build NLP-based software that can tell you that this many clients have requested this type of bond, or a client can see from their chat history what they've been asking for and what prices they've received from the banks. So actually extracting useful data from these unstructured chats is something that is trying to be done within the industry.

**[0:12:44] Lee Smallwood:** And again, I think Bloomberg themselves recognize they have this data. It's a great add-on service that they can provide. And ultimately, I'm sure that they would like to

**[0:12:51] Daniel Faggella:** charge for in the future. Yeah. We've already heard of use cases where on the compliance side, again, the analysis of phone calls for, let's say, I think stock brokers to some degree, or I think monitoring compliance issues such as things that sound like insider trading or other kinds of line crossing, just kind of keeping tabs on that and seeing what the frequencies are, seeing kind of who's doing it, seeing when and where it's happening. It sounds like through chat, similar things could happen.

**[0:13:18] Daniel Faggella:** We could see if people are having their feelings or thoughts about certain topics at scale, you know, across the company, or we could find individual instances where, you know, we may be paying a regulatory price if people talk about those kinds of things in that kind of way. Do you think that it's that same, you know, spike of the downside, that same, regulatory threat that will be a bigger driver of adoption of that technology in terms of internal communication than will be, say, broader insights around what are people talking about, when are they talking about them?

**[0:13:47] Daniel Faggella:** What are your thoughts on that?

**[0:13:49] Lee Smallwood:** Yeah, I think that's definitely right. I think that's the other angle. Compliance departments in banks are generally huge. And the level of layers and number of funnels and filters that they go through in terms of analyzing chat is quite tremendous. So it's a big lift because the downside is so great. And so yet again, it does seem like a really good use case. Even, you know, not saying we'll get to a point where specific misconduct can be identified, although I don't think that is necessarily off the table, but even just taking down the millions and millions of lines of chat to stuff that people who have expertise in the area have a little bit more of a curated set of

**[0:14:26] Lee Smallwood:** conversations to review from a compliance angle. Yeah, yeah. Even just

**[0:14:30] Daniel Faggella:** cutting down on the raw bulk, right? I mean, it's the same kind of thing in cybersecurity. Instead of looking at all of our instances, maybe we can at least have them be flagged where we might find some meat. And it's the same thing here. That's kind of the interim step as opposed to detecting every instance. Can we at least have less false positives, less false negatives, and have our human experts looking at the stuff that matters?

**[0:14:51] Daniel Faggella:** Seems like that's almost an unquestionable yes. And my guess is the technology is at the level to do that. It's just going to be tough to integrate it and whatnot. So curious to see. The way I think about it, Lee, and you can let me know if there's something to correct here, is that it's less about giving banks a new carrot that maybe they could try to understand like, oh, here's these new opportunities you have for insight.

**[0:15:13] Daniel Faggella:** It's just telling them about the known stick. Here's a way to avoid this thing that hurts you and you think about it a lot. It just seems like that's the needle mover here. I wish I could find some great examples that break from that in NLP, but you let me know if you see the

**[0:15:29] Lee Smallwood:** same. No, I think you're absolutely right. I think if you step into any of the big banks and can get under the hood, I think what you're likely to see is a lot of people, the main concern, the thing that keeps them up at night is the regulatory downside. They know they've got massive challenges ahead of them, huge slates of work to try and prevent things that could be damaging for the institution, either reputationally, financially, or otherwise.

**[0:15:52] Lee Smallwood:** And so if you can come in as a company that offers the ability to make that process easier, better, more accurate, I think that's something that's going to get a lot of people's attention. And so I think that's why that's having arguably more success than some of the other things you mentioned, increasing

**[0:16:08] Daniel Faggella:** efficiency, potentially increasing revenue. Yeah, it's wild. It's wild. Well, I mean, you were with us in the creation of the full report, kind of this first phase really does seem to be risk oriented. And it might be years until, you know, at least our take from the high level, it might be years until the shift in banking goes to even looking at efficiencies. Right now, it's just decreasing those big spikes of downside. So NLP the same.

**[0:16:30] Daniel Faggella:** It sounds like that's your take on that. That's true. But if I just jump in there to say I agree, but I also

**[0:16:35] Lee Smallwood:** think that's the wedge because there's still a cultural change, something we've discussed before, adoption to either working with startups or working with new technologies. And so, yeah, this is the big pain point to get big institutions comfortable with leveraging this kind of technology, working with these kinds of people. But I think once they're in, that really opens up the door to demonstrate the talent and the other use case. And particularly when you talk about revenue upside, that's where I think you could actually quickly have a change.

**[0:17:01] Lee Smallwood:** You know, again, the timing is tough, but maybe more like two to five years than 10 years until you start to see some really interesting applications really focused on moving the top line.

**[0:17:11] Daniel Faggella:** Yeah, yeah. Eventually, we'll get there. And yes, you and I actually did speak about this on stage at PayThink. There might be like two listeners who are there at the event. But for everybody else, yeah, I brought Lee up on stage and we were presenting the banking research that he was part of and talked a bit about the pros and cons of compliance being the real driver of adoption. And as Lee is bringing up now, it might be a great wedge to start to bring AI into financial services.

**[0:17:37] Daniel Faggella:** And it sounds like, Lee, from your experience, NLP is sort of not all that different, kind of follows some of the same patterns. So just being aware of where we are on time, I'll swivel us into looking at the future. So when we think about NLP, there are some low-hanging fruit examples. We talked about picking up on those internal communications issues. We talked about some somewhat reasonable low-hanging fruit around documents and what might be possible there.

**[0:18:01] Daniel Faggella:** When you think about where NLP might bring us further ahead, where we're really not today, but where we hope to arrive to, as someone who works within a big bank, where do you really hope NLP takes us? What are the kind of big zeitgeist shifts that would really enable a lot of great change in banking that you're most looking forward to?

**[0:18:19] Lee Smallwood:** Sure. I think the thing that gets me interested and excited is thinking about voice conversations. And you spoke about this a little bit earlier, but it's really interesting. I've definitely thought before about, oh, if you could transcribe the voice and then run NLP on this, then you could have some really interesting applications across the institution. Of course, the further you go down that channel, the more you realize the way that we speak is not the same way that we write necessarily.

**[0:18:42] Lee Smallwood:** And so there's definitely a couple more steps to get to a point where, one, we get really, really high quality, low error rate transcription. And then two, our NLP models can adjust for the fact that this is a transcription of a voice conversation and not more natural text. again, natural language processing. So somewhat unnatural almost. But I think that creates a lot of interesting opportunities, especially internally. And you spoke before about the compliance example for phone call monitoring.

**[0:19:09] Lee Smallwood:** But I think other things as well that, you know, classic examples might be, you know, CEO speech transcriptions and using that for guidance on how companies!! But I think where it gets really interesting is things like, again, on the sales trading side, trade booking, if you could speak that into a microphone, I bought or sold this, and all the booking is done, and it goes through the clearing, then that saves a ton of efficiency and will make people's lives a lot easier.

**[0:19:35] Lee Smallwood:** So I think that step from voice to text to NLP, and then some direct action is really interesting to me in the future.

**[0:19:42] Daniel Faggella:** Yeah. Okay. So where can speech immediately kind of trigger action reliably in the future across banking functions. Obviously, you mentioned trading as one. I can imagine, obviously, interacting with customers would be pretty important there as well, right? People eventually are going to be talking to Amazon whenever they want something or, you know, I forget the percentage of Google searches that are done by voice, but it's obviously astronomically more than it was three years ago, probably astronomically more than it was 18 months ago.

**[0:20:09] Daniel Faggella:** You know, we might imagine that a similar expectation will be there with banks. We've got, you know, making trades? Where else, I guess, do you see potential transformation there? Because that would be a big shift. Where else might that open up some opportunities? Yeah, I think that the

**[0:20:25] Lee Smallwood:** next step on the challenge is obviously big institutions take customer data and privacy very seriously. And so it's how much can you extract from one side of the conversation? So for example, maybe moving to the investment banking side, can you look at the communications between a banker and a set of clients where you don't necessarily know who the client is, and you don't hear their side of the conversation, but you can start to look at how effective certain salespeople are or certain people in the banking side are in terms of the communication that they use, what's leading to actual deal flow, strategic investments, mergers, acquisitions, things like that.

**[0:21:01] Lee Smallwood:** So can you adhere to the customer privacy side, look just at the bank side and then also extract interesting information? I think that's where it gets really interesting to me as well. Cool.

**[0:21:12] Daniel Faggella:** Well, Lee, I know that that's all we had for time, but I sincerely appreciate you joining us here back on AI in Banking. It's been great. Thanks for having me. So that's all for this episode of AI in Banking. If you've enjoyed this episode with Lee Smallwood of Citibank, be sure to leave us a review on iTunes. It's pretty easy to find us, AI in Banking, at iTunes. If you haven't already tuned into our other podcast, it's called AI in Industry.

**[0:21:39] Daniel Faggella:** covers more broad topics beyond financial services, you may want to check out that podcast as well. Next Monday, we're going to be sticking with our theme of NLP and we're going to be talking about searching for documents and searching for data within the back end of banks. So we're going to dive full blown into search and discovery in this next episode and look forward to having you with us. I'll catch you next Monday. Thank you.
