What you'll learn
- Strategies to create seamless and personalized borrower journeys.
- Leveraging customer data to improve lending experiences.
- Harnessing AI and Machine Learning in Digital Lending
- Payments Integration in Digital Lending Platforms and Navigating Regulatory and Compliance Challenges in Digital Lending
Miriam Cross (00:11):
Hi, welcome to Digital Banking 2024. My name is Miriam Cross and I'm a Technology Reporter at American Banker. The topic of today's panel is the Evolution of Digital Lending, specifically how banks are moving from traditional approaches to digital first approaches to lending. We have three panelists who can talk about this from all different angles, so I'll start by having each of them introduce themselves. Can you give everyone your name, your role, and how digital lending plays into your day-to-Day. Rich, I'll start with you at the end.
Rich Longo (00:42):
Yeah. Rich Longo from McKinsey Company, Senior Advisor beside New York. And the question around digital lending,
Miriam Cross (00:52):
How does it play into your day-to-day?
Rich Longo (00:54):
Oh, well, it's a top of mind topic with a lot of our clients and we're seeing a lot more investment in the role, especially as they contemplate taking cost out.
Tiffany Patrick (01:06):
Hi everyone. My name is Tiffany Patrick. I'm with Citibank. I'm the SVP of AML Payments and Innovations. So I support the services business and for digital lending AI opportunities within Citi. There's actually multiple use cases that we are looking at, not just to solve the consumer needs, but also to install our internal needs as well.
Barath Narayanan (01:28):
Thank You, Tiffany. I'm Barat Narayanan responsible for banking and financial services and insurance business unit that persistent systems for us. Digital lending is almost at the heart of transformation that we see in the banking industry and persistence systems being the genesis of the firm has been product development and product engineering. We are one of the leading players when it comes to supporting fintechs and digital banks in developing their products. With that, the emphasis on digital lending has been enormous over the last three to four years where we saw maximum growth. We continue to invest in the space and we'll discuss more, but that's me.
Miriam Cross (02:14):
Great. Thank you very much. Rich, can you tell us an overview of what you're seeing among your bank clients? In our prep call, you mentioned that you're seeing, especially among tier one and tier two banks, that commercial small business is a big focus when it comes to digital lending.
Rich Longo (02:31):
So there are several, I'll encapsulate it in the top 25 banks in North America, a consistent focus on turning around at a faster rate term sheets. So where commercial lending usually took two to three weeks to underwrite the target state is a few hours and they're looking at multiple sources of information. Like a few years ago, you couldn't get access to tax transcripts from the IRSS. Now you can, there's multiple data sources from counties on property and other information that wasn't digitized and now it is. And so they're trying to figure out how they can, as they upgrade their systems, move systems to the cloud, how they can tie in more applications and databases that are externalized and they can tap into it for decision making process. In addition, we're seeing them very focused on using tools. Not like we hear a lot about generative ai, but it's more of the basics like robotic process automation. And then a bit of what we talked about is within lending because things are becoming digitized and because it's becoming easier to borrow on a commercial side, there's an increase in fraud and risk. And so they're also thinking about as they make it easier, how the bank can protect itself.
Miriam Cross (03:57):
And why is speed so important? I think one thing you mentioned last week was commercial clients may be applying to a few different banks and they'll go with whichever decision comes back first rather than shopping around for rates.
Rich Longo (04:13):
So our research shows that typically they could be at let's say a 50 75 basis point higher rate, but some of these businesses, especially the smaller businesses and mid-cap businesses are willing to, if they get the funding, they like the conditions, they're not going to wait another week, they're going to accept the offer and they're going to proceed and get the loan funded. And we're seeing that over 70% of the time, it's not as much about rate shopping because they expect the bank to be within a certain parameter and they'd like the speed and because they need to operate their business, they have business needs that they have to get done and they don't want to wait three, four weeks.
Miriam Cross (04:56):
And Tiffany, how is Citi using technology to take the manual aspect out of the lending process?
Tiffany Patrick (05:04):
Sure. So we're looking at options that benefit both the customer but also our internal processes. So by that I mean automating the ability for a customer to produce the documentation and everything that Rich was talking about so that they can submit the application. But then also you still have the manual piece on the inside where there's a person who's now going to read through all of this. You have to automate and use the technological support on both sides of the process. So not only automating the submission, but can we automate those maker functions within the bank and then utilize our skilled resources to validate what's being submitted. Okay. And where would you say, how far along is Citi in this? We are starting the journey. Citi is a big shift to turn, but making sure that we're identifying very effective use cases and experimenting as much as we can.
(05:59):
Citi just had their digital money symposium last week, and what the focus really is, is making sure that we're solving broad common use cases within the firm through all the reorgs and everything that's happened within Citi. And the focus is to make sure that all five business units of Citi work seamlessly together. If you continue to have manual processes, that's not going to be the option that you're going to come out with.
Miriam Cross (06:26):
So what are some examples of use cases?
Tiffany Patrick (06:28):
So for one of our products, not to jump ahead, but for one of our products for the digital natively digital payment, so Citi Payments Express, which has been in the news recently, that is a digital native product. So we're not uplifting any old products. We're starting from a digitally native space that we're able to then pull out analytical data from not just for our consumers and our clients to understand what's going on, but also from the control perspective as well, so we can use the data of the transactions to train our controls and our monitoring at the same time.
Miriam Cross (07:04):
And I want to get back to that concept of Citi being a big shift to turn soon, but Barath, what are you seeing banks investing? What technologies are you seeing banks investing in terms of lending?
Barath Narayanan (07:16):
Sure. Just follow through to what Rich and Tiffany said. We see institutions in two different segments. One who are the digital banks fintechs. For them, it's all about scaling their entire customer base. They want to go from, they're at a $2 billion under assets, they want to increase it to three to 5 billion. For them, it's all about scale. Starting from loan origination to client onboarding. That is absolute importance for the clients in that segment. But when you look at the commercial lending side, the wholesale large institutions, Tiffany spoke about Citi or even Wells Fargo or many other clients, it is all about process digitization. Now you have a huge number of people taking care of the underwriting process, taking care of various different client onboarding process, the exceptions, the false positives that comes out. That's a huge manual effort for the large institutions. It's largely around the process digitization improving the entire digital value chain. And that's how we are seeing helping our customers using the engineering discipline. Of course, with the AI and the whole evolution of machine learning, it only continues to improve.
Miriam Cross (08:37):
Rich, one thing you said earlier that I thought was interesting, you said people talk a lot about AI, but you've seen a lot of return to basics like RPA.
Rich Longo (08:45):
Yeah, so I think Penny kicked it off very well in the very beginning and said the Google example and something about glue being nutritious and good for you, it's about the data. Think about these large institutions or many of the institutions, you run into problems where only 70% of a particular system is data mapped. The data is not normalized. So for example, in one system, I might be rich long, another system Longo Rich Junior. And so it can't normalize that data and say there's one consistent identity for that individual. And you could have tons of other use cases. So without cleaning and normalizing the data, without it all being connected for you to actually say you can leverage AI in a process that's very complex, like commercial lending is very hard. You have to start in small areas. You have to really prioritize systems. And we're talking about banks too, that not all their systems are in the cloud.
(09:53):
Not all of them have API connectivity. And so you have to address the basics. So if you think about the tools that are within machine learning or ai, one that has been around is common tools like common processes, robotic process automation is a lift. It takes out cost, it takes out friction. And so you build from there, but you're still working on the data simplification strategy and the data connectivity strategy. And eventually we'll get there, but it's like the blocking and tackling first. And quite honestly what we're seeing from the AI space is there are a bunch of POCs, it's POC heaven, but they can't ever get to the execution. And quite honestly too, we're hearing from the regulators, they have a lot of questions as well around it because the consistency, they don't want, for example, a system to go to the product and pricing engine. You might have three product and pricing engines. It looked at the wrong pricing data and now just gave the customer an understanding of what their rate band would be. And it was completely off, right? There's a whole bunch of attributes here. And so we have to be careful.
Miriam Cross (11:03):
And Tiffany, how do you feel about this?
Tiffany Patrick (11:05):
A hundred percent agree with all those points, but just to touch on a few of them. And just going back to the back to basics and the data strategy. So this is only beneficial if we get the data, and by that I mean the data that a potential customer or a current customer submits. And then what we're doing with the data once it's inside the bank, if it is not normalized, if it is not in a way that we can leverage it and really have data as an asset, the accuracy of that process is diminished. And that defeats the purpose of using gen AI. And for the customers, it defeats the purpose of the process because they're looking for something that is easier, streamlined, quick. Everything should be done on my phone like we talked about this morning.
(11:47):
But if we take accuracy out or we leave accuracy behind that convenience is completely destroyed and now we're back to this constant calling the customer, there's friction again in the system. So I think that's number one. Number two, just touching on the regulatory point, we have to be able to understand exactly what the AI solutions are doing from someone who consistently has to speak to regulators during exams. I cannot go and say, well, we just send it to that engine and it does its thing. It's not going to be a good response. And it has to be a consistent performance of that engine as well. So again, POC Heaven, absolutely, because we only have so many data sets that are ready with that accuracy to deploy this solution, and then we should expand it. And as we look to expand it, and I think we might get to this in a minute, but then you have to talk about the bias of the dataset that you're using and making sure that you're not launching that before you're ready.
Miriam Cross (12:47):
Yeah, that's another part I want to get to, but when you talked about Citi being a big shift to term, what did you mean by that?
Tiffany Patrick (12:56):
So number one is the data and the infrastructure and the tech stacks that we have. So large institutions that have acquired other business units have acquired fintechs or other small startups that the tech stacks are very different. They're not necessarily meant to talk to one another, they're very disparate. And then you have to look at what is your data normalization strategy. Instead of just trying to build pipes to every single different data system, that's not going to help you. You need to see if you can decommission those data sets first or data systems first, excuse me, and then bring that into a centralized hub and make sure that you're also cloud ready. Because as you're doing this, any institution has to be able to scale to the amount of payment traffic that is coming your way. Things are becoming as easy as a couple of clicks on your phone. The product I talked about earlier is going to be able to do 5,000 transactions per second, and we're looking to expand that into 30 countries in 30 months. That volume's exponential, so you better be able to handle that data. So again, that's one use case, that's one payment product in one line of business. So that's the big ship to turn. We've got to start to do that across the board.
Miriam Cross (14:09):
And Barath, how familiar does this sound to you and your work?
Tiffany Patrick (14:13):
A lot. A lot I would say. But as a firm specializing in engineering, product engineering, I'm going to slightly defer by adding a dimension of what are the future possibilities. I totally take where we are today, no question on the data frame. Data needs to be standardized. Data quality needs to improve data transparency and availability should be there for decision making. No second doubt. But the possibilities we are seeing with our customers with respect to digital lending space especially is phenomenal. For example, we have, rich, you're aware of this, we have worked with one of the clients in digitizing their entire loan origination process and decision making. And there we have used machine learning and LP to process the enormous amount volumes of documents that is existing in the past. And the effort is reduced by more than 60% inability to make decisions that are logical for that particular transaction.
(15:21):
Now, what Rich called out, and Tiffany also alluded to it, you can't just give it to the system and assume, oh, that's it, it's a black box, whatever it takes. And this is where Tiffany and I were discussing earlier, the maker and checker process comes into play. We should digitize more than 90% what we're seeing with our clients on the maker process, the checker process still needs to have the right level of manual intervention before things are taking a final decision to a client. And that can also be automated. But that is how I would bring in an element of how you use the technology and engineering to the benefit of scalability, speed and protecting the data, but not going into the journey of adopting them. And that's where I would say the possibilities are enormous. And yes, we did 90 POCs last year in last 12 months. Of course it's all POCs, but with few customers, we have already started rolling out into production for document processing, certain level of origination process. We are seeing things going in live.
Rich Longo (16:35):
I'd like to just add something to that. In some cases we've seen more in production automation, but it's generally on short-term lending. It is not a migration to a new system. So they're still servicing, let's say one year buy now, pay later, or something like to that effect. And they run on that system, the existing accounts, but new accounts being onboarded in service are now in a separate system. They built a digital bridge with its own separate ecosystem, so there's no conversion, no migration, and they're able to build it, but it still sets outside of the enterprise. And that's how it's almost an extended POC in some ways because it's not fully in production. They're still servicing on one system and starting to service on the other system. Eventually it'll fully run as those accounts get paid off. And there's one in production today and another institution.
(17:36):
These are again in the top 25 that's in that journey of doing that right now. And they're taking, and they're externalizing. So they're building a digital bridge to various systems. In this case, it was a product and pricing engine. It was a payments engine and it was a servicing engine that provided a lot of automation behind it and tied it with a process orchestration underneath it. And a client could go in there in the application process. And it was almost mixture of bots and generative ai, but mostly generative AI where they could ask dynamic questions and it could respond in terms of understanding the product a little bit better or the processes. So we are seeing that, but not in mass production.
Miriam Cross (18:23):
So then in terms of AI, Tiffany, how is Citi using it?
Tiffany Patrick (18:28):
So right now we're looking for more of the very basic automation of the processes or learning large language models, small language models to help condense. We have a lot of documentation in Citi, a lot large clients have a lot of documentation. It builds up over the years and those PDFs or attachments, unless you have a person sitting there and reading it, you're going to spend hours. So I think we even saw a use case of that earlier this morning where you're taking hours of manual processes and condensing it down and then really looking to leverage that in one business line and then expand to the next business line. Because we do have customers that span multiple business lines. We shouldn't have someone from private bank calling the same customer because they're going to look for a line of credit in their commercial business.
(19:18):
It should all be centralized in one spot. And then the business line leading that gen AI capability should be the one leading the way.
Miriam Cross (19:27):
And then to return to the topic of explainability, what does responsible deployment of AI look like to you?
Tiffany Patrick (19:35):
So we need to check a few boxes from a control perspective, we already touched on them, but data accuracy, data standardization, and then also making sure we are not introducing bias into the dataset or into the technology itself. There's a lot of regulation out there. And then to add on another layer of complexity for institutions or fintechs that are supporting cross-border payments, you need to make sure that you've got those very nuanced regulatory requirements taken care of. So if your product that you're servicing is already going to be ISO compliant in that data set, you're off to a better start. But if it's a legacy system that's still being uplifted or a legacy product to the ISO standards, that will take longer
Miriam Cross (20:22):
And Barath how big of a problem is bias in underwriting?
Barath Narayanan (20:27):
When we work with multiple clients, what we see is this is an evolving topic which has not been solved yet. So I want to be very honest in saying that. Has this been sorted out? No, it's been watched very carefully. But what we are seeing is what I called out earlier, the maker and checker process. And that's the element where you adopt the evolution of technology to significantly automate or eliminate the checker process. But on the maker side, there is series of validations still to be made on. Are there any biased outcomes being seen? And to what people spoke about the glue in the pi are any stupid decisions also being made? But that validation on the checker side has to be kept, and that's what we are educating to our clients when we are doing the whole engineering process is what I would say.
Tiffany Patrick (21:20):
Can I just add to that real quick as far as explaining to regulators and making sure we understand, we're all here at this conference and we are excited about it and we want to understand the technology, but as you implement this, the rest of your team needs to be able to understand the output, understand the very basic mechanics of what's being done, because it can't just be one person in one department explaining this to a regulator. The whole team needs to understand because the regulators are also on a knowledge curve right now as well. So when you bring this up and you say, we're going to move to machine learning transaction monitoring, that's a conversation I've had. Or we're going to use Gen AI for KYC, they're going to say, okay, well how is that the same as what you already have? You're going to have to translate what you're implementing into terms that they understand so that you won't get hit with a million follow-up requests and really targeted exams. But that's a really key point is whatever you're learning here needs to be shared across your teams and your senior leaders so you don't have regulatory pitfalls.
Rich Longo (22:27):
No, I was just going to add on that this is unproven to the regulators right now still. So if you make a mistake that'll quickly for the next few years and any of your automation projects, end it and on the IT examinations, you know exactly where they're going to go first. And don't expect, be proactive as you're explaining it. We're thinking about this, if you have product managers in there, disclosure, BRDs, right? Your TRDs, let them see it. Explain the business issue you're solving, who you're partnering with. That's the other thing. They don't trust that you have the talent internally that can do all this. So a colleague of mine earlier, Vic talked about rewired and all these other things, but yeah, will also take additional new talent. And generally that talent will come from outside of banking because there's been more advancements in other industries, not as regulated as us, but that's when they'll have to partner very closely with the regulatory partners within the bank. But be very careful because you could automate and you could feel do random tests and say, yes, we've caught through our sanction screening and we've auto adjudicated these files better and it's popping through. It only takes one or two big ones, and then you're in a lot of trouble.
Miriam Cross (23:52):
And Tiffany, how are you educating your employees?
Tiffany Patrick (23:56):
So we do knowledge sharing discussions with any vendors that we're talking to, and we also take the initial ideas straight to our technology partners. And then our product team has also taken their ideas straight to us. So part of my role in payments and innovation, yes, I'm the control person, but we all have, the ultimate goal is to reduce friction across the entire system. Not just the customer product relationship, but the control relationship. Because if something goes wrong, you miss one, and then now there's a stop sell on your product from a regulatory perspective, then we've all lost. So making sure that you have a representative from your regulatory engagement, your product team, your technology team, and your senior leadership from the get go is how we're educating.
Rich Longo (24:46):
And most of these banks, we're not seeing successes with business, fully understanding what technology's bringing to the table and what they're delivering or delivering in smaller chunks. So they're looking at it as a strategic project, but they're kind of boiling the ocean a little bit and making it too wide. And then business is waiting for something within six to 10 months, let's say, and now it's going to take two years. And then business loses attention and says, I still have these pressing problems solve this. And so even though technology in a lot of these banks are brought in as we're talking to the business side, they think it's just like this Rubik's cube of complexity that is mythical and it's like a square peg in a round hole sort of situation.
Tiffany Patrick (25:35):
What we have observed, what we have observed in multiple clients is what Satish called out in the morning. While the technology is so keen to innovate, we don't address solving the fear that the business has. And that I believe was one of the important takeaways for me with SAT session, is how the technology organization can equally focus on addressing and onboarding the business into the whole journey upfront and clarify the fear part rather than, oh, we allow this tool, there's one more product we want to roll out. I think it needs a lot more attention is what I would say.
Rich Longo (26:15):
Absolutely. And it comes down like the basics, right? Banks are struggling for deposits right now. Hey, if we want more deposits, we need to lower our overall costs so that way we can increase our interest rates. So there's a give and take here. So you have to be very clear from a business case perspective how this is going to impact business and address current concerns on survivability for some of these banks.
Miriam Cross (26:41):
So we have two minutes left. Does anybody have any questions? Maybe we could fit one in no questions. We educated everyone. Okay. I'll just add one more rich. Something you said in our prep call, you talked about quick wins as a way some of these projects can take so long to come to fruition. So what is a quick win and how can a bank do that?
Rich Longo (27:06):
The project I just mentioned on the short-term lending side, that project from start to finish was 10 months to get into production. They brought in three vendors. They worked with another vendor to build the digital bridge. And sorry, I don't have permission to disclose the client, but they were able to from start to finish, get it done in 10 months, and it literally reduced their CAC by 80%.
Miriam Cross (27:37):
All right. That covers it. Thank you all so much for being here. Thank you. Thank you. Thank you.