Artificial Intelligence – The New Digital Divide (Part 2) [Podcast]

Artificial intelligence and machine learning has been a topic that’s been on our minds quite a bit lately at PrecisionLender.

In this two-part series, Carl Ryden and Jim Young discuss some of our favorite articles about AI and how it relates to the banking industry.

Podcast Transcript

Jim Young: Hi and welcome to The Purposeful Bank 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. Last week we gave you part one of our two part conversation with PrecisionLender CEO Carl Ryden about why we’re so excited about the impact artificial intelligence and machine learning is going to have in banking. In part one, we talked about the article “Artificial Intelligence is the New Digital Divide”. We left off just as we were transitioning into another recent article that had caught our eye. That’s where we’ll pick up in this episode. Enjoy.

Carl, the next article I want to talk about is called “The Simple Economics of Machine Intelligence”. It’s a Harvard Business review article by AJ Agrawal who’s a professor at Rotman School of Management, University of Toronto, and it put things in really simple but elegant terms to me. It talked about as AI and machine learning become more prevalent then prediction as a resource or as a value will become much more of a commodity. The barrier of entry, in other words, to get to those predictive powers, but in the process judgment will become much more valuable. I wanted to get your feel for how does that play itself out in a bank. As prediction becomes more of a commodity, judgment becomes more valuable. What does that look like at a bank?

Carl Ryden: A lot of what banks do are prediction … predicting if this person’s going to repay particularly in the utilization line of credit, predicting what the prepayment rate is on a loan account.

Jim Young: Right.

Carl Ryden: And overall predicting the future profitability of that relationship, right, when you’re making a decision about pricing the next component of that relationship or not, whether or not you participate. In the past I think banks have leaned on humans, with a limited number of samples, to kind of stick the wet finger in the wind and make a guess. In fact, humans were the best means of doing that. I think we’re getting to the point where a lot of that predictive piece … computers are getting better.

Jim Young: Right.

Carl Ryden: Systems are getting better at that. It’s not to say, “What does that leave lenders and relationship managers and those folks?” Instead of guessing at things now we can kind of have good data around these things and actually know that. Again, that amplifies the human in the process to where, “Okay. I think this is where you’re heading. Let’s help you manage that more effectively.” And judgment becomes very important.

In the article it talked about the economic idea of there’s certain inputs- calculation, prediction, judgment- and what happens is as the other inputs become more and more of a commodity and when they move into computing and into cloud it gets more of the same thing that we talked about happening with storage is going to happen with- Prediction is going to become very … almost free.

Jim Young: Yeah.

Carl Ryden: Right? And when you have that, the other complimentary things, the compliments to those inputs become really valuable. Right? And so judgment’s going to become quite valuable. The system’s I see is most systems you build, instead of automating a human workflow, it shouldn’t just automate a human workflow it should learn from that human workflow- what makes it work, what makes it not work- and then offer suggestions on how to improve that workflow. Or maybe even, is this a workflow that is core to what we do and is core the experience we’re trying create or is it standing in the way of that experience we’re trying to create?

Folks who can make good judgments really what’s important to that customer …

Jim Young: Right.

Carl Ryden: … socially, emotionally, functionally, contextually. Having a AI system kind of as wing man in that interaction I think is where we see a lot of it going.

Jim Young: Yeah, and you mentioned about the social and emotional. One of the examples we always use a lot in training people is that whole getting a guarantee from your father-in-law sort of thing. The machine might tell you that this is the right thing to do because that is going to change the numbers around, but the relationship manager would be able to tease out- A good one would be able to tease out, “You know what? I have that. I don’t really want it. I’d really like to go a different direction with this.” That’s something that … I like to hold onto those things because when we had an intern over the summer give us a presentation, I jokingly raised my hand and said, “How soon before the machines come for my job?” I think that is a natural reaction when you see a lot of this stuff doing things that humans have done, but it’s one of the reasons I like this article is it gives a really good explanation as for why the human part of it is going to be not just there, but immensely valuable going forward.

Final article here, “Why Artificial Intelligence and Machine Learning Need to be Part of Your Digital Transformation Plans”. This is a ZDNet article. One of the quotes in here is from Bryan Solis said it. “Machine learning is going to allow companies to see things they wouldn’t otherwise because of the cognitive bias that exists in the relationship between humans and the data they collect.” Let’s put it in banking terms. What are some of the cognitive biases you would see in banks in that situation?

Carl Ryden: To numerate all of the cognitive biases that …

Jim Young: But a couple.

Carl Ryden: What makes humans good at that limited data set, humans are really good with limited and ambiguous data and making a gut instinctive prediction based on that. Our brains have been built to do just those things. The problem with that is, the flaw in that is that it is very prone to cognitive bias. Now there’s a bunch of those- cost bias, biases created by compensation systems is one we see all the time where you actually do things that are detrimental because, in the long run, because of short term …

I think one of the more interesting ones is … I call it the “we’ve always done it this way” bias. You ask a lot of bankers, “Well why do you do this?” And they really can’t tell you. A lot of times as you kind of dive in and we get in the math of profitability, we find things and you go, “Why do we do that? That doesn’t make sense.” One of those, for example, is if you ask a lot of bankers, “Why do you amortize loans?” Some can tell you why it is. Some just say, “We always amortize loans. It’s what we do.” It turns out when you look at why you amortize loans, you amortize a loan because it reduces the credit risk. It reduces the LTV over time and that reduces your risk of adjusted return, and all of those things make sense. But then the question is, What’s the appropriate am term for different types of deals? Why wouldn’t you amortize a loan? It turns out if I loaned you $100,000 secured by a $100,000 CD held at my bank, then the appropriate am term is infinity which means it’s an interest only loan. Right? Because my credit risk is zero if you don’t pay me back. I’ve got collateral that’s worth exactly what you owe me. My loss given default is zero. So the am term on that should be infinite.

One of the things I push banks on is I hear a lot of banks say, “Our problem right now is we’re running off a lot of our portfolio. It’s paying off and we got to replace it. We got a hole we’ve got to dig ourselves out of every month.” I say, “Are the people coming in and prepaying their stuff and just paying off their account, or is it the natural amortization?” They go, “Most of it’s natural amortization late in the am cycle.” And you say, “Your problem is that people are repaying you when you don’t want them to repay you, but they’re repaying you because you’re forcing them to repay you. Stop doing that.” Right?

Jim Young: Right.

Carl Ryden: It turns out most of these people are late in the am cycle where most of their payment is principle. Their LTV on their loan is 30% so that the best credit risk deals you have, they’re probably almost fully collateralized if you look at the recovery of a collateral, and they’re essentially the same as the $100,000 CD security thing. You say, “Well why don’t you do this? Why don’t you go out and do a database pull and go back out to those folks and tell them, ‘Put you’re money in your business, not into you bank,’ and switch them to interest only and that will stop that problem.” You say this to folks and … I’ve said this to quite a few banks and they go, “Oh my god. We never thought of it that way.” I’ll say it to some banks and they’ll say, “Oh my god. We never thought of it that way, but we can never do that because folks just believe that you have to am things.” These sorts of things are things that you can’t let the machine run the face, but having a machine that will challenge you and say, “Well why are you doing this?”, having data that shows you why are you doing this … Once you kind of go through and understand the math, which machines are quite good at understanding the math, then offering suggestions of “Wait a minute. Maybe we should do this.”

We’re big fans of, particularly in banking because it’s people’s money, it’s so emotional, having a human in the loop. They call it HITL, having a machine as the wing man to a human and a human with an open mind and good judgment who’s willing to except that gets to a really good place both for the bank and for the customer.

Jim Young: One thing I know … Banks a lot of times when you talk to them, they’re worried about their data and they think they’re data is messy. It’s not quite good enough. By comparison, I guess when we’re talking about data sets and the importance of them for machine learning and AI … The bank industry’s data, how would that compare say to some of the other industries that are diving into AI and machine learning?

Carl Ryden: A lot of the bank’s data was not built for the purpose of distilling it into intelligence. Right? Most of the banks … A lot of the banks- I would say most is probably a fair statement. Most bank data is the data that’s meant to process transactions. It was built for a tactical means. I’ve got to debit this account and credit this account and make sure no pennies get lost. That is the core of what we do. It’s also the most commodity thing you do. That’s the table stakes now.

Jim Young: Right.

Carl Ryden: But most of your systems were built merely to handle table stakes. Over time they’ve been modified to where they can satisfy regulators and pull reports and other things, but they weren’t build to deliver intelligence back into the system. And again, it’s a criticism, but there’s reasons why. There’s always reasons why. Most of these systems were designed and built 20 years ago when storage was $500,000 a gigabyte. You have specific two digit codes for certain things and there’s only a few of them. Every bit was precious. Well today you say, “Well people are giving me fifteen gigabytes for free. Store everything that might be useful.” Then you’ve got the processing power that’s cheap to kind of sift through the wheat from the chaff later. That’s an entirely different design principle at work. Most of a bank’s data is built to process transactions and do the transaction processing bit of it.

The other things that banks … They only care about the outcome. What was the rate? What is the payment that I’m due? Because that’s all that’s needed for that transactional bit. But the how you got there and the decisions that were made along the way by humans are really important because you want to train the machine to be able to understand those decisions and be able to supplement those and augment those and offer suggestions. It can say, “Okay. I see banker A over here actually performs much better.” You want to get to that causal mechanism of, “What does he do differently?”

Jim Young: Right.

Carl Ryden: But most of the systems, they don’t keep track of what do we do differently. It only keeps track of the output, the end state. I think you’re starting to see a lot of systems start not just the what, what happened, but how did we get there. And then learning from that. That’s something we integrate deeply into PrecisionLender is we want to understand not just where you ended up but how you got there and what choices were made along the way and why. And then when we see a similar circumstance, we can offer a suggestion.

And by the way, the huge opportunity for banks is most of the things … A lot of the banks you’ll talk to, “our situation is unique.” And there is some uniqueness, but I think it’s overplayed a bit because a lot of the things … We see immense value in instead of taking things that have already been done in e-commerce, in Amazon, and Google, and other folks, take those and apply them into banking. Right? And in fact, a lot of the problems in banking are much easier because they are by nature data.

Jim Young: Right.

Carl Ryden: It’s numbers. It’s balances. It’s finance. I’m not trying to do voice recognition or face recognition.

Jim Young: Not trying to understand sarcasm.

Carl Ryden: I don’t need natural language processing. I’m looking at accounts and numbers, and data, and things that are naturally fit to a computer. So you can actually take the hardest problems that are being solved in the consumer sphere, a sub-set of that is a really easy problem in AI and machine learning, and apply that into banking. We see a lot of that.
For example, in PrecisionLender when you’re using our tool and as you’re pricing a loan or pricing an opportunity, in the context of a relationship it’s no different than when you’re going through Amazon and searching for big screen televisions and whatever. They know the transaction you’re in. They know which things you’ve looked, which things you’ve considers. And they know which things you might consider next. They also know the relationship they have with you. What are the things you’ve bought over time? What are the things that you’ve rated? What are the things that you’ve said was really and said was not so good?

When you’re talking about pricing an opportunity to a commercial customer in the context of that relationship, it’s not just the current state of what they are, it’s the entire history we have of how we got to this point that we can bring to bare. And that’s something that a computer can bring to bare a lot easier than a human because lenders change over, and they move, and you have new relationship managers and whatever. Being able for a bank to build a brand and maintain that relationship with that customer, you have to have AI machine learning.

You will have to have it. If you don’t you will be at a severe disadvantage.

I think, again, the idea of amplifying humanity, understanding the greater and richer context, being able to see across lenders, across the bank, across time, and looks at huge sums of data … Not just about what happened, but how it got there. Then feed that and put that in the hands of a lender as a coach, a digital coach, to the next lender saying, “Hey, you might ought to know this. Hey, you might ought to consider this.” Those things will be incredibly powerful and will be table stakes.

Jim Young: Carl, I know you can keep going on this subject for a while, but we’ll stop there for this episode. I know we’ll be returning to the subject of AI and machine learning frequently in the coming weeks and months. Remember you can always find more information about today’s episode at precisionlender.com/podcast. If you like what you’ve been hearing, make sure to subscribe to the feed in iTunes, SoundCloud, or Stitcher, and we love to get ratings or feedback on any of those platforms.

Thanks for listening. Until next time, this has been Jim Young and Carl Ryden and you’ve been listening to The Purposeful Bank.

The post Podcast: Artificial Intelligence – The New Digital Divide (Part 2) appeared first on PrecisionLender.