Machine Learning in Banking [Podcast]

 

According to St. Meyer & Hubbard, 57% of the loans that go to committee get rejected. How will machine learning shape the way banks do commercial lending in the future?

In this episode, you’ll hear from the CEO of PrecisionLender, Carl Ryden and managing partner at Georgian Partners, Justin Lafayette about ways that banks can use artificial intelligence to cut costs, save time and improve efficiency.

   

 

Podcast Transcript

Jim Young: Welcome to another episode of The Purposeful Banker podcast. The podcast brought to you by PrecisionLender, where we discuss the big topics on the minds of today’s best bankers. I’m your host, Jim Young, and thanks for joining us.

Today’s a special podcast because we actually have two guests on today. One is Carl Ryden, which most of you know as the CEO of PrecisionLender, and the other is Justin LaFayette, managing partner at Georgian Partners. We’re going to be talking today about artificial intelligence and machine learning and what that means in the world of banking. Before we dive into the topics, Carl, can you fill in for the listeners about our, for lack of a better term, our partnership with Georgian Partners?

Carl Ryden: Sure, sure. Again, it’s worth telling a little bit of the origin story behind this. As we’ve grown at PrecisionLender, we’ve developed a nice relationship with a company down in Austin, Texas, that we’ve come to know. The CEO there is a guy named Chris Hester. We came to know and like Chris. Last year at BankOnPurpose we had a chance to sit down and spend some time with Chris in person. During that meeting we told Chris a lot about the things you were doing with data and advanced analytics here at PrecisionLender. Chris has always been a little bit of a mentor to us, telling us how to think about our business and grow in our business and helping our customers. He said, “The guy you really need to talk to is Justin at Georgian.” He raved about Justin as a board member, as a partner, as how much they’d helped him grow their data business and find new ways to serve their customers with advanced analytics.

He said, “When you talk to Justin, he’s going to offer to do a workshop for you and have his team do that.” He said, “Whatever you do, say yes to that, and it will be worth it.” That’s really how we came to know him. Chris and Tyson came down and spent a day with our team about how to take the data we have and the knowledge we have and find new ways to deliver creative value to our customers. That was kind of what spurred the whole partnership. Ultimately, we just wanted to go deeper with that. Justin was joining our board, Georgian had invested in our company, and we couldn’t be more excited about that.

Jim Young: I know you guys can go way deep into these topics, but first let’s start off with the highest level here. Carl, just give us a general explanation of what we’re talking about when we say artificial intelligence and machine learning.

Carl Ryden: Well I actually kind of hate the term artificial intelligence. It is what is commonly … folks are referring to now as artificial intelligence or machine learning. The real idea is how do we take the vast amounts of data we have about how loans are priced, opportunities priced, how relationships evolve, and take that and deliver back actionable, valuable information, to our customers in ways that help them better serve their customers. I really like the term automated intelligence or augmented intelligence or other things like that. I think our customers experience that with the user experience within PrecisionLender right now.

What we’re going to do is take that to the next level and find ways to deliver more actionable intelligence right to the moment where it can have the biggest impact. One of the things I like to say is, “No one really wants artificial intelligence, they just want more intelligence, so how do we actually deliver more intelligence at the point where it can actually have the biggest impact in the organization?” I think that’s what we’re trying to do.

Justin LaFayette: Yeah, this is Justin, and let me say first, how excited everyone at Georgian is to be involved in the story now. We’re a bit of an unusual investment firm in that we look for companies that are taking advantage of certain … we call our investment pieces, but certain trends that are affecting business software. Artificial intelligence is one of three main things we’re pursuing now. Finding companies that are committed to leveraging these new technologies in applied ways on differentiated, meaningful data sets, is what we look for. This is a perfect fit. We have a team that helps our companies take advantage of these trends and adopt them in not just technical ways, but how you build market, how you price it, how you sell it, etc. Even, primarily, where you focus on, what are the first things you could focus on to get started.

It’s quite fascinating how varied the definitions of artificial intelligence and machine learning are. That’s an hour long conversation unto itself. One of the ways we like to think about it, is there’s a spectrum of sophistication. When you really start to get into interesting parts of it is when, based on very large data sets across a domain, like in this case in banking, algorithms can start to learn from new data on their own and anticipate action. That’s one of the things that we’re really interested in, the ability for an environment to, as you use the term, augmented capabilities for people to anticipate valuable action and inject that into business.

Sometimes, people use artificial intelligence to describe things where it seems intelligent, alerts are coming, people are being notified of things, but the really interesting things start to happen when all that data can be turned into action, that no person was going to take on their own, and then automated action in large transaction volumes is where really interesting value can get created.

Jim Young: For either one of you guys, Carl or Justin, who wants to take a stab at this one first. Banking often has a reputation, and sometimes undeserved, of being a little slow to adopt new technologies. Do you have a sense of what the attitude of the banking industry is towards artificial intelligence and machine learning?

Carl Ryden: I’ll take that, and then I’m sure Justin’s probably done a wider research. Maybe ours is biased. The idea of machine learning, predictive analytics. There’s a bunch of flavors of this, if you want to get into taxonomy, but really predictive analytics piece of it. Being able to predict what’s on the outcomes and what’s going to happen based on the data that’s in front of you and that you trade on. A lot of the best banks have been doing this for a long time. It’s really a part of the credit process in a lot of them. Just saw a talk from a guy named Dave LeGassa, from Capital One bank, about the foundational principles of using that stuff. They built their business around that, their card business.

I think you’re starting to see that come into the bank from the card business, from the consumer business, and leaders like Capital One in that area have done a lot of work. You’re starting to see it move into different parts of the organization, and I think one of the things we’re excited about is taking a lot of these same things and making them operate on a different data set within the banks, particularly around commercial lending and commercial relationship management and commercial pricing as well.

Justin LaFayette: Yeah, this is a fascinating time in banking. I used to run a software company that sold primarily to the banking industry for ten years back in the 90s. It’s a particularly interesting market to me, and I’ve never seen banks as an industry so simultaneously excited about he potential of a technology, but also threatened by it. I think it’s part of the larger FinTech movement. You see very interesting behavior with banks in the last couple years. Things like, for instance, taking investment positions in funds, just so that they can get more access to what’s happening from a software disruption standpoint. You see banks trying to get connected to the local software and development communities in the cities that they’re based in, participating in incubators. In some cases, I’ve seen banks moving parts of their IT staff into technical hotbed kind of centers, so they can be immersed in the culture of startups.

It’s kind of might be happening for two reasons. On one hand you’ve got companies like PrecisionLender who are arming them with software that can accelerate their business and create efficiencies. They’re realizing that some of those companies, because of the nature of SaaS business models, are aggregating data sets around specific domains, like commercial lending, bigger than any one bank has on their own. That’s a new phenomenon.

There were a few data businesses, historically, that worked very hard to get people to share data, but now you’ve got lots of businesses who sell them to an interested banking, who, by the nature of how SaaS business model works, just end up, if they get permission, and do it appropriately and with trust and permission, can see an aggregated set of data across a huge swathe of the industry. Suddenly, banks are saying, “There are companies now who can help me do things based on data sets, with artificial intelligence, machine learning, predictive analytics, etc.,” on data sets bigger than they have themselves. That’s quite amazing.

On the other end, there are companies who are using those technologies to compete directly with banks more aggressively than ever before, and the entrance of everything from peer-to-peer lending to … well, pretty much every part of the lending spectrum. You’ve got startups coming and saying, based primarily on the capabilities of data, and ability to underwrite risk better, that they’re competing directly with banks. That’s certainly causing banks a lot of concern. They’re simultaneously getting all these amazing new tools brought to them, who gave them abilities on broader data sets than they have before, but they’re having new competitors entering their markets.

Jim Young: You guys have talked about there’s clearly this need out there, and there’s this potential out there. What are some of the … Carl, you tackle this one first. What are some of the biggest obstacles, then, standing in the way of making even greater use of AI machine learning?

Carl Ryden: I think one of the things that we encounter a lot, big banks all the way down to small banks, different flavors of the data they have is a mess. It wasn’t built for the purpose of extracting insight. It was built for transactional systems. It was built to debit this account and credit this account and make sure you never lose any pennies, which is great, that’s absolutely necessarily, but these days it’s not efficient for building insights. It keeps track of the state of where things are today.

I think something that helps us as a SaaS company is interfacing those other systems, the LOS systems, the CRM systems, the core transaction system, and bringing that data together in a way that transforms it into something that is conducive to drawing insights from, in a more rationalized state, and launching a state that allows us to actually look at trends over time, to actually look at cross-sell opportunity, upsell opportunity. Have us understand what do we think the prepayment rate is for this loan, and really bringing together these different perspective on the data, together into one spot, so that we can use them all to make better decisions.

I think the biggest obstacle in banking, from a technical standpoint, is that the data is a mess and wasn’t built for the purpose of generating insights. The second problem is some perceived, some real, is the regulatory issues with it and convincing the fear of what regulators might say. I think that’s starting to erode a bit as banks start to make a case and start showing that this actually produces a better and even a fairer deal for customers as well.

Justin LaFayette: Yeah, I couldn’t agree more on both those points on the data end and the regulatory issues. I would add that another thing that we’ve certainly seen banks struggling with, with this concept of artificial intelligence and automation, is the further you go into automation of decisions and action taking … The industry, historically, has had more checks and balances than most, for good reason. It’s a pretty interesting ethical question, privacy question, security question, lots of dimensions to it, but all related to the issue of if a system start making more intelligent decisions in automated ways, who’s responsible, and what’s the comfort level, both with banks and the people that work there and also their customers?

There’s a very interesting paradigm playing in other industries that are a little further ahead on this, of kind of resetting expectations and norms and people’s comfort level with the amount of automation and intelligence that’s being injected into processes. You see in commerce, a high degree of comfort now with the level of sophistication that companies like Amazon inject into the issue of transactions and processing, in order to get speed and costs down, efficiencies, and all the gains.

Banking is a lot earlier into thinking through some of those … what will be the norm on comfort and responsibility, as you automate more actions and more decisions?

Jim Young: Along those lines, then. Carl, you talk about when you implement this sort of stuff, when you’re at that lender customer conversation, the relationship manager. Where relationships end there, and it’s very much a human element to it. How do you see AI and machine learning integrating into that, changing that conversation, and is there a concern, as Justin said, about the level of automation in that?

Carl Ryden: As Justin was talking, I’m thinking about the evolution of a lot of the … If you take things like automated driving, autonomous vehicles, it starts out with there’s always a human in the loop, and then there’s systems, cruise control, or other things, that help the driver do what they do better, and take out the routine and mundane tasks, so they can focus on other things. But now, it’s come to where the machine must always protect the human, so with lane assist, and blind spot tracking, and those things.

I think you see kind of the same thing, where the data, the information we’ll share with the banks or with the relationship manager, will help them take away the mundane task, take away the simple task, uncover opportunities that would have taken too much time and too much energy to actually uncover on their own. You’ll this is where it actually augments their process of dealing with that customer and the relationship allows them to focus on the highest value things that humans do. One of the things we often say is we try and do everything that a computer can do well, so that humans can do the things that humans uniquely do well, and really create that augmented experience.

I think you’ll also see, on the consumer side, particularly in the thick of things in the banking world today, where human systems often run amok. I think you’ll start to see machines kind of protecting the guard rails of, wait a minute, maybe these accounts that are being created here are not real, right? Keep that from a recent case, as well. Ultimately, I think you’ll find that running a large financial services organization, you’ll have to have a certain amount of machine learning intelligence helping your folks help your customers.

Then you also might have things detecting when the humans have gone off the rails a bit. Think about lane protection or blind spot correction. You know, wait a minute, we’re heading off the road here, we need to take a course correction. I think you’ll start to see a lot more of that. It will become almost a requirement of business, to do it safely.

Justin LaFayette: Yeah, I think it’s also an element of the way that some of this capability manifests itself, that will make it more comfortable, and easier to take advantage of. For decades, banking automation through technology and software was about screens and trained people sitting down in front of screens that conduct transactions and business. Then the web came along, that got pushed out, and eventually to mobile devices to end customers. It’s entering data, reading things on a screen, choosing options from menus, and deciding on transactions.

One of the interesting things about AI is that it’s overlapped with, actually, another one of our investment pieces around messaging for business, which really means a conversational style interaction with software. Whether that be through the big text messaging platforms, or even human speech directly. When some of these capabilities, and I like the way … your, Carl, the analogy to automated driving and the safety for people. When that interaction moves more and more into a conversational style, and software becomes almost personified and human-like in a way you can converse with it or it can understand human spoken language and written language.

It’s a lot easier to get comfortable with than to take advantage of the capability. I think some of these parallel things that are happening as major trends are going to make it actually easier for people who both work at a bank and are customers of banks to take advantage of this kind of intelligence in a comfortable, conversational, style. You see this happening in the consumer world very, very rapidly. There’s incredible advances in the last six months with the opening up of APIs to all the popular messaging platforms, and the sophistication. Google’s latest things that they’ve released around the generation and understanding of human language is phenomenal capabilities now being exposed and put in the hands of software developers everywhere. We think it’s going to be not only a change in the capability but the style and the way that people interact with that capability.

Jim Young: Carl, you already touched on this a bit, about anomaly detection. Are there other ways you can see that being used? Maybe not necessarily to keep the bank on the guard rails, but in a way to … other ways that banks can use that to save money? To be maybe more efficient.

Carl Ryden: I think there’s a bunch of ways. Anomaly detection is one recipe, and it’s actually a really, pretty simple one to put in place. One of those anomalies you might detect is, is this a deal that ultimately we think is going to get done? From an efficiency standpoint, more than anything else is, in a commercial setting, it’s understanding, is this a deal that’s going to work? Can I qualify this as a deal that’s going to be simple for both the bank and the customer, before we go deep into the underlying process?

When you go deep in the underlying process, the expenses ramp up for the bank, the pain ramps up for the customer, and the damage you do to that relationship with the customer and the brand, if you openly don’t get to a conclusion that’s from the loan, escalates. Being able to, earlier in the process, detect the sort of deal that’s maybe just not going to be a good fit for the bank, from a credit policy standpoint, from a relationship trajectory standpoint, from a pricing standpoint, to all sorts of things. The sooner you can take those out …

One of those statistics that I shared with you before, and I think it’s a pretty good one, that came from St. Meyer and Hubbard, that for a lot of banks, particularly smaller banks, something like 57% of deals that make it into the origination system, back into underwriting, never make it on the books. Because, at the beginning, they were either never acceptable to the customer or to the bank. Finding better ways to qualify that this is a deal that actually can get done, is a really powerful thing for the bank and for the customer.

I think something that is really important, on that relationship side, from a commercial standpoint, is really focusing on that customer experience you’re creating, because your brand on the commercial is the sum total of those customer experiences you create and being able to manage that effectively and create better and more tailored and more mass-customized solutions efficiently, for those best relationships, those more profitable relationships, is really cheap.

Jim Young: There’s a belief that the majority of this, I think with a lot of stuff we’ve been talking about, there’s a lot of numbers, but that there’s a lot of usable business information that originates in unstructured form, i.e. text. Talking about natural language processing and the improvements that’s shown as of late. How can banks use natural language processing to turn all that text into some of these insights that you’ve been describing?

Justin LaFayette: The number of inquiries that the people have and simple instructions and things that they like to get done that involve talking to people is very high. If you break down the percentages of all that cost and effort, to provide that service, from a customer call center, or in person situation. Natural language processing unlocks this capability of being able to divert a huge number of high percentage, 80% and upwards, of that activity into an automated response. As I was mentioning earlier, if it can be done in a way that’s truly natural, literally like talking to a person, and it’s indistinguishable, there’s tremendous cost savings in that. The accuracy in the automated situation can be higher, even, than people.

Then that frees those people to answer things that really, really do involve and require a person. Where there’s emotion and customers are upset or happy. It takes human intuition to resolve it. The interesting thing that we’ve seen examples of, is companies start with some level of automation on the natural language processing front. Often, in text, but not full, spoken situations. First places that this takes hold.

If they do it right, what they use is the results of the human in the loop, the system seamlessly switches over to people jumping in on the situations that require people. If you follow those cases right through, you actually get a closed loop learning kind of situation, where the intelligent environment learns more and more of the edge cases, of the ones that require people. Become problems that the system can solve, by actually being exposed to what people did to resolve them, which frees more and more time to deal with the high value relationship driven type activities that people are best at and should do.

From an overall efficiency standpoint, not necessarily reduction in roles for people but focusing people on much higher value things. It is quite extraordinary how fast natural language processing married to artificial intelligence. How fast the capabilities have improved. Our view is, it’s gone from 0-100 in the last year. It’s just an astonishing pace. It’s just beginning to be adopted into real business situations. The reasons are not because new technology was invented, it’s just that data sets that allow for the training have been … are incredibly large now compared to what they used to be, and some of the major, major technology platforms like Google, Amazon, Microsoft, Facebook, who have made massive investments in this, are exposing the capabilities those platforms have for everybody to take advantage of. There’s a lot in it for them, and there’s some risks to businesses because of that capability being driven by them, but there’s no question it’s accelerating the adoption at an incredible pace.

Carl Ryden: For our customers, they’re typically commercial customers, they’re high net worth, they’re in the private bank or other things. One of the private bank folks use our products. One of their biggest fears is that the rest of the bank won’t really know how important this relationship is. One of their biggest fears is that a relationship, multi-million dollar relationship, generating several hundred thousand dollars of net income to the bank, calls into the cost of their overall boat loan, or car, or some small thing, and have a horrible experience. Because of that we lose the whole thing.

The national language on the inbound processing allows folks to detect synonym analysis and other things to say, “Is this person upset or frustrated or other things?” But then it also, and hopefully before they get that, they get in a higher level of services Justin talked about, so they’re not there. Then when they do, varying that with, wait a minute, this is somebody who’s really important, and I’m learning their relationship manager, to have a human connection ump on it and make that right, and provide them the level of service we want to provide the best relationships at our bank. I think it’s going to be a really important thing. Having that knowledge of how valuable is this relationship to our bank is critical to that, and I think that’s what folks use PrecisionLender for a lot.

Jim Young: Finally, the talent needed to develop all of these capabilities. The AI/ML capabilities within banking. How much of that is a challenge attracting that? Is there, to be frank, an obstacle in attracting developers who might say, “It’s banking, I’d rather do something maybe that seems a little sexier, a little more cutting edge?”

Justin LaFayette: I can give you a point of view on that that’s recently played out in history. One of our original thesis that we pursued from an investment standpoint was applied analytics, kind of on the path heading towards artificial intelligence. Applied analytics was taking all the big data technologies that were being open sourced and released and applying the capabilities to specific business problems. The skill sets required, new tech stacks, new programming languages. A data science capability is different than programming, and there’s a real shortage of data science people. That’s played out now, and it’s more ubiquitous, but the interesting part of that trend, when it played out, was just where the people went. Where they were found and where they’re trained and where they pulled and where they wanted to be.

It was definitely big. The big technology platform, social companies, like Google and Amazon and Apple and Facebook, who attracted a lot of those people. Then it was very cutting edge software companies. A lot of traditional businesses, banking in particular, struggled to keep people. In fact, a lot of people trained in the skill sets you need to do that kind of work were in banks and got lured out.

Artificial intelligence is a whole other level of rarity, in terms of skill set. It is incredibly few number of people out there compared to the interest level in artificial intelligence. A lot of them were in academia, where artificial intelligence has been around for a long time, and lacked the ability to get applied. That has changed quickly. Some of the eminent leaders, thought leaders, have come out of universities and been lured and a lot of them went to contracts at Google and Facebook and places like that.

Google, several years ago, paid $400 million to buy a London, UK based company called DeepMind. That’s in the media a lot these days, because a lot of things they’ve been working on for the last two years are being released now. They had no revenue and 48 artificial intelligence people. They were one of the largest groupings of that skill set, and that’s what Google was willing to pay to get them.

I think it’s like applied analytics. The trend is playing out. There’s even fewer people to draw from. You’re seeing universities respond, you’re seeing the pockets where this data exists out there. A lot of competition to hire people. Companies like PrecisionLender are going to be at a tremendous advantage, and this is one of the things that we think is key to trends like this. The banks … The desire and the appetite to take advantage of this technology is going to be far greater than the resources that can build it. A company like PrecisionLender can go to people who have that skill set and say, “We’re doing something really innovative and interesting in the banking industry.”

What drives a lot of those people is how interesting and innovative it is, and also the size of the data sets that they get to work on. As I was mentioning earlier, PrecisionLender is going to have the ability to go to some of these people and say, “We have a data set that spans the entire industry and we’re doing something really cool. We’re a software company with value that’s growing and you get to participate in that.” That is tough for banks to compete with. I think there’s certain types of solutions like this that are far more likely to be created using some of these new technologies by companies like PrecisionLender than banks trying to build it themselves, with more limited data sets, limited to their own business, and a culture that is just going to be more difficult to track that kind of skill set.

Jim Young: I think that’s going to wrap it up for this episode. Carl and Justin, thanks to both of you for coming on and making time for this episode and for educating our listeners in an area that is really, as I said, on the cutting edge, and fascinating in its long-term implications for the banking industry. Also, big thanks to everyone for listening. I will provide links to a few resources in the show notes for this episode, and you can always find those at PrecisionLender.com/podcast. If you like what you’ve been hearing, make sure to subscribe to the feed in iTunes, SoundCloud, or Stitcher. We’d love to get ratings and feedback on any of those platforms. Thanks for tuning in. Until next time, this has been Jim Young, and you’ve been listening to The Purposeful Banker.

The post Podcast: Machine Learning in Banking appeared first on PrecisionLender.