SDS 238: Data Science in Banking

Welcome back to the SuperDataScience Podcast ladies and gentlemen, super excited to have you back here on the show. And today we’re continuing our series of episodes within the FiveMinuteFriday podcasts where we talk about data science in different industries. And today we are covering of data science in banking.

Banking is a very interesting sector. And what’s also quite peculiar to note that in essence, financial bank institutions are companies or corporations which provide intermediary services in the financial markets. And they’re broadly separated into depository, contractual and investment institutions. Now a few numbers. In 2016, the revenue of the banking sector was 2.5 trillion US dollars. That’s 2.5 trillion with a T. Massive, massive revenue. 

And in developed economies, the percentage of the revenue is about 20% of the GDP. So as you can imagine, banking is a massive sector and therefore there are lots and lots of opportunities for disruptive technologies such as data science, AI, machine learning to be applied. So let’s have a look at a couple of examples of how data science can be applied in banking.

Example number one, fraud detection. Very, very interesting space. As you can see that we’ve got 2.5 trillion dollars in revenue, of course, it’s very lucrative, a very lucrative area for people to commit fraud and take a percentage of that revenue in terms of fraud.

And make money that way. So in fact, about 0.1% of all credit card transactions are fraudulent and even though the percentage is so small as 0.1% because the industry is so huge, 2.5 trillion US dollars in 2016 alone, that makes for a large amounts and experts predict that online credit card fraud will reach a 32 billion US dollars by 2020. That’s 32 billion US dollars in fraud by 2020. And so you’ve got quite a lot of fraud happening. But on the other hand, we’ve got the problem of too many false positives when legitimate purchases are blocked as if they’re fraudulent. And that doesn’t come as a good customer experience. So I think we’ve all been in situations like that where your card is blocked or a purchase is blocked because you were actually trying to make the purchase, but it didn’t go through and was flagged as fraudulent.

And sometimes that can be a bit frustrating. I’ve definitely had that happen to me over the, even just this year I’ve had it happen several times at least. And that’s something where machine learning and AI can help, right? Using data science, machine learning, AI reduce the amount of fraudulent transactions, but also reduce the amount of false positives. An example here is that Zensed is a self-learning AI engine that claims to reduce undiscovered fraud by up to 80% by scoring of various aspects of transactions in real time. And of course, fraud detection in this sector doesn’t end with just credit cards. There’s lots of lots of other ways.

For instance, detecting fraud in financial statements during auditing and things like that. One of my favorite tools in this space is Benford’s law. So If you haven’t heard of Benford’s law and you’re thinking of getting into this space, I really highly recommend checking it out. It’s a fantastic tool as you can use in order to detect fraud in a very interesting way. And if you’re looking for podcasts on fraud detection in this whole space, we have podcast number five, with Dmitry Korneev where he’s actually an expert. So he worked in Deloitte for I think it was 10 years as a fraud detection expert. I worked side by side with him. Very, very interesting person to listen to. So if you want to check it out, it’s podcast number five.

Alright. Example number two of application of data science in the banking sector, customer service automation. So what we’re seeing in banking more and more is that they are becoming automated, but when you go into a brunch, a lot of the functions can be performed without interacting with a client manager.

And that’s the way of the future. In fact, some countries they are already banks, they’ve already opened up banks where or bank branches where you can walk in and there’s absolutely no people at all. Just fully automated. You’re talking to artificial intelligence. I think that was in Korea. I might be mistaking where exactly, but I saw a video about this and that’s pretty insane. We walk into a branch and there is no people there. Full Stop. In fact, Autonomous NEXT recently predicted that replacing human customer service personnel could save the banking and lending industry 450 billion US dollars by 2030. Another example here is that you can already ask Amazon Alexa for financial information and it will give you that information. This is provided by the Swiss bank UBS. It is estimated that there’ll be over 14 million users by 2020 off such queries and that is showing that financial information is becoming more democratized and is also becoming more popular because it’s more accessible.

Example number three of data science in banking, financial risk modeling. A very popular example. This is the probably most famous example of data science, the bread and butter of data science. Something that actuaries also work with very closely is predicting what kind of risk a certain person or a company holds when taking a loan. So whether it’s a mortgage or a loan for a car or another loan, how do we know what the risk in this situation is? And that’s because a lot of factors and basically the end result is should we approve the loan or not? And what’s interest rate can we possibly give to the person? We’ve all heard the example of Enron, which was a massive collapse. Enron lost in investors or its shareholders, a total of $74 billion. This potentially could have been prevented with better artificial intelligence, machine learning, data science driven financial risk modeling.

That’s just one of the examples. Of course, these loans are happening every single day, per personal level, on an organizational level and in fact, a 2018 report from McKinsey shows that machine learning may reduce credit losses up to 10%. It doesn’t sound like a lot, but still, any advantage that you can provide as a data scientist to a bank, if you can tell them that you can reduce the credit losses by 10%, they are going to be very, very interested in you. And we’ve got a podcast related to this. If you’re interested to learn more about financial risk modeling, podcast number 14 of the SuperDataScience Podcast series with Greg Poppe. A very interesting chat we had quite a long time ago about this space. So if you’re interested in that podcast number 14.

Example number four of data science in banking, process optimization.

So banks are generally organizations that have a large number of high volume, low-value processes. So this is quite important so I’m going to repeat that. High volume, low-value processes. So that means replying to emails, checking financial statements, adding up numbers, running reconciliations. So lots and lots of tasks, none of which are super complex. It’s not like a self-driving car where you have very complex processes and you need very advanced artificial intelligence to run them properly and make sure everything is running smoothly. In banks, it’s actually quantity. That is the main issue, main problem. So, therefore, whatever way machine learning, data science, AI can help with those, that quantity, that volume of processes that will be very helpful. And we’ve got a couple of examples here.

So the first one is J.P. Morgan. J.P. Morgan began using AI to process internal IT requests, which include operations as simple as employee’s attempts to reset their work passwords. In 2017 alone, the AI handled around 1.5 million requests and replaced 40 full-time employees. Can you imagine that? 1.5 million requests and replacing 40 full-time employees.

Example number two is again from J.P. Morgan. In 2017 the bank introduced a machine learning algorithm that enabled it to automate the process of interpreting commercial loan agreements. And this move saved the bank 360,000 hours of legal practitioners work per year. How crazy is that? The algorithm could do the work of a human in seconds. So as you can see, J.P. Morgan is already on top of things by the looks of it.

And the example number three in this space of process optimization is RPA, robotics process automation. It is taking over the world and really automating a lot of mundane, repetitive white collar work that humans don’t really need to be doing. Humans can be doing more fun, creative things. If you want to learn more about RPA or robotic process automation, we’ve got a great podcast with Leigh and Daniel Pullen from Melbourne, Australia, episode number 173.

Last but not least, application of data science in banking, number five, security.

And here we’re talking about things such as how do users authenticate themselves. So what we’re seeing is that image recognition is one of the fastest developing areas in artificial intelligence. And that can be a way that people authenticate themselves. Already with iPhone, most recent iPhone, you can authenticate yourself with face recognition, right? So well, Google intelligence forecasts that 1.9 billion bank customers we’ll be using some form of biometric ID by 2021. And UK Bank Halifax even experimented with Bluetooth wristbands that identified a client’s unique heartbeat to create an authentic login. Another example is typingdna.com can authenticate you based on your typing patterns. So rather than getting a two-factor authentication, SMS or some other form of two-factor authentication, you can just type and they will authenticate you. How crazy is that? All that is machine learning, artificial intelligence in the space of security in banking.

So, of course, those are just five examples of applications of data science in banking. There are plenty more. We are only looking at five in this podcast, but something to think about. This is a very interesting industry, a massive industry as we saw 2.5 trillion US dollars in revenue just in 2016 alone.

So if you’re already working in the space or you’re considering working in this space, or just maybe you have some ideas, hopefully, this podcast was helpful to get you to spark those ideas, to get you thinking in that direction. And on that note, thank you so much for being here. I look forward to seeing you back here next time. Until then, happy analyzing.