Can AI make Banks Brainy : Use case of AI in Banking and finance

Everyone loves talking about AI is going to change everything we do. And Banks, which have been seen as a laggard when it comes to adopting technology, this time don't seem to be shying away either.?

From conferences titled "Welcome to the future of banking, powered by Artificial Intelligence (AI)", to people talking about AI Agents taking over the mundane jobs of bank employees, we have a wide range of use cases being theorized.?

In this article, we’ll dive into the different types of AI models that are currently implemented in the industry. We’ll explore how AI is enhancing the underwriting process, making it faster and more efficient. We’ll also delve into the role of AI in anti-money laundering efforts and fraud detection, providing a safer and more secure banking experience.

But that’s not all. We’ll also look at how regulators and central banks are leveraging AI to stay ahead (or behind?) the curve. We’ll discuss how AI is simplifying compliance and regulatory documentation requirements, saving countless hours of manual work. And finally, we’ll see how AI is transforming customer service and support, leading to happier customers and higher Net Promoter Scores for banks.

So, sit back, relax, and join us on this exciting journey into the future of banking. Because the future is not just coming, it’s here.

AI In Banking : TL;DR

Now, I know some of you find it hard to go through a 15 minute read on a Monday morning.

I recorded a video podcast on this topic, so if that is more of your jam, then do give it a quick glance through

Video Podcast : Banks get Brainy by leveraging AI

BTW here is a TL;DR of it all:

But if you are interested in knowing how these banks went about their AI pilots, and what kinds of AI models were developed, and how to develop your thinking when it comes to implementing AI into your financial products, then the rest of the article is for you.

Type of Artificial intelligence models

Now, as I was researching for this article, I was overwhelmed by the sheer number of AI models which are present. But as I started delving more into each of these AI models I realised that they could be classified into 3 broad buckets:

• Artificial General Intelligence (AGI): This is the ultimate dream - machines that are as intelligent as humans, capable of independent thought, reasoning, and problem-solving across various fields. It's still in the theoretical stages, but scientists are actively working on it.
• Narrow AI (ANI): This is what powers most of the AI you see today. These are specialized tools designed to tackle specific tasks exceptionally well. For example, facial recognition software or a program playing chess at a superhuman level.

Then there is third category of AI which is imaginable today known as Artificial Super Intelligence, where the machine intelligence far surpasses that of man. It could solve problems beyond our current comprehension and potentially usher in a new era of technological advancement. However, like what every doomsday movie warns us of,? if an ASI's goals doesn't align with ours, it could potentially wipe out the entire human race from the face of the planet.?

And since its all hyperbole for now, I wouldn't worry much 'bout this.

The rise of Financial crimes due to AI

Anyway, back to financial services.?

Would it surprise you to know that the biggest place where AI is being deployed, is in trying to dupe the financial services agency into committing huge frauds ?

And the sophistication of these frauds is just being supplemented by AI tools.

Just recently in Hong Kong a corporate executive transferred $25.6 million to the accounts of fraudsters. How did he get duped into transferring this huge amount to an arbitrary account?

?

So the fraudsters arranged for a video conference with this executive and 15 others. The catch? The CFO and 14 others on the call all happened to be deep fakes. Naturally the executive not being that high up in the chain of command, got pressured into committing this crime now in technical parlance this is known as financial muling, where you make an unwitting legitimate person and accomplice in your financial crime.

Using AI to combat money laundering and financial frauds

But it seems quite intuitive to assume that if AI can be used to scam people financially it can also be used to combat it right?

?

And true enough we have several use cases of Fair AI is being used to detect these kinds of financial frauds.

There are several AI tools that use biometric authentication methods layered on top of your existing authorization / authentication protocols. In fact AWS has this entire repository of APIs known as AWS rekognition which do what is known as a ace detection and liveness detection, Using a combination of facial expressions movements dynamic words etc.

AI is also being used to fight forgeries in the financial services sector. Document forgery also happens to be one of the age old ways of committing financial crimes nowadays. especially with AI its become even more easier to forge a signature because after all it is just a bunch of pattern recognition right? Thankfully AI can be very useful in identifying and saving out these forged documents.

In fact hyperverge any AI based kyc company released a case study of how they were able to figure out if the documents were originally signed or generated using a as well. And they were able to do this with a fair amount of precision and low false positives.

Fighting Money Laundering with AI : A case study from HSBC

Money laundering happens to be the biggest pain point for banks across the world. In fact one estimate by HSBC pegs the amount of money laundered to 2 to 5% of the global GDP. That comes up to being close to about $2 trillion. Wow!

?

And so HSBC tied up with Google last year to run a pilot to see if I can do a good enough job of detecting these money laundering transactions and stop and root out organizations which are defrauding the financial systems.

Traditional AML technology happens to use rule based systems for detecting patterns and stopping money laundering transactions. Now if the fraudsters are able to figure out the pattern it becomes easy for them to avoid detection as well.

This is where Google comes into the picture with HSBC and their pilot that they ran last year the results of the pilot were simply mind blowing in retail banking the AI agent was able to identify 4 times as many fraud transactions and in corporate banking 2 times as many fraud transactions.

Now the obvious question that arises is what happens to the false positives?

False positives are nothing but legitimate transactions which were flagged off as fraud transactions due to the rules that the system with set up. The AI agent who was able to give 60% lesser alerts then the rule based system.

This meant that the people at the other side who are actually saving out and stopping these AML transactions was spending more time in sewing out the fraud players from the system. A job very well done I would say.

Also because this AI agent happened to be self learning it meant that new patterns that crime lords we're using were more effectively identified and seived out of the system.

Building Underwriting Models with AI

Yet another area of banking which is pretty much rule based happens to be loan underwriting. Loan underwriting is the process of figuring out how much of a loan can the bank lend to a customer and expect the customer to pay them back in full. In a sense your figuring out the need the customer has for a loan the ability of the customer to pay the loan back and use some software elements to figure out their willingness to repay that loan back.

As you would've guessed this another huge area for artificial intelligence in the banking sector. But there is also a pretty dark side to it.

Biases in AI Models : A case study from Goldman Sachs and Apple?

What is the rule based system that the AI agent was trying to displace happened to have a bias data set?

In fact we saw this playing out when apple tied up with Goldman Sachs to issue credit cards to their customers. There was this now very famous case of a couple who had applied for the credit card. However despite the fact that they both had a similar financial background the wipes credit card application was rejected.

This is simply because traditionally banks have been very apprehensive of lending out especially unsecured loans to minority population.

And AI Agents, trained on such biased data sets, would only work towards enhancing and multiplying those biases at scale.

How AI can level the credit playing field??

But that was 2020 and we are now in 2024. In fact there is this player called Zest AI and they have been implementing some very amazing use cases in the lending space.

Zest AI has helped lenders increase approvals by 49% for Latinos, 41% for Black applicants, 40% for women, 36% for elderly applicants, and 31% for AAPI applicants, all while holding risk constant.

And some of their customers like Suncoast Credit Union have nothing but praises to sing for the underwriting speed, ability to identify good credit customers by using alternative data points, ability to help in the collection process of potential defaulters.

Generative AI : Servicing Banking and Financial Services customers

And now finally we are at the generative AI use case for banking. This I think was a no brainer. Using Gen AI for customer service requests.?

In fact in October last year, Erica, Bank of America I chatbot past 1.5 billion customer conversations .

Born to make banking a breeze, Erica was one of the OG AI assistants in the financial world. She can do everything from checking your balance to finding the nearest ATM, and even handling all your other banking chores. Erica has been chatting it up with over 37 million Bank of America customers since she burst onto the scene. And last year, users spent a whopping 3 million hours shooting the breeze with Erica, a 31% jump from the year before. Now, instead of just handling transactions, users are turning to Erica for advice and insights over 60% of the time. She's not just a virtual assistant, she's the ultimate banking sidekick.

These days, Erica is mainly busy keeping an eye on your subscriptions and bills, as well as studying your spending habits. Just to give you an idea, people are constantly checking for refunds from merchants around 863,000 times a month, and keeping tabs on their FICO scores around 267,000 times. And as if that wasn't enough, Erica can also direct you to specialists for any intricate problems that go beyond the realm of self-service. Making her a multitasking digital assistant!

Explainable AI: A use case for AI in Financial Literacy programs

Now onto the path that I'm most passionate about. Financial literacy! Often times new 2 financial services people can find the jargon and the long terms and conditions to be very intimidating.

And if there is someone who is new to finances and looking for a loan at the same time they? may ?be, ?tempted into signing and agreement that's not truly beneficial for them.

And as I keep thinking of this, wouldn't it be amazing, if and explainable AI model with multimodal capabilities could explain the financial product without any jargon in the applicants native language?

How are regulator and central banks approaching AI?

But, if you've been following this newsletter for a while, then you would know that no new technological prowess is possible without the blessings of the regulators and the central bankers.

This month, the European parliament has given its approval to the European Union's Artificial Intelligence Act, marking a significant milestone in establishing legal guidelines for AI globally. With the EU AI act set to come into effect by year-end, businesses that operate within the region and utilize AI technology must ensure compliance with the law, lest they face hefty fines.

The EU AI legislation is so all-encompassing that it doesn't just affect European organizations, but? global players trying to sneak their AI systems into the EU. The EU is basically saying, "We run this tech town, Bros!"?

So, here's the deal: This new Act is gonna label a lot of important AI systems used in different industries like finance, insurance, healthcare, and more as "high risk." I mean, no long can you get away by pointing to AI and claiming it gave you the instruction to execute a command. If you can't explain how AI arrived at that answer, well then, you'll have to answer to the authorities.

That means companies will need to really look into these systems and come up with ways to make sure they're still reliable and secure. If they don't follow the rules, they could face some serious fines. Insurance companies will have a bunch of new legal responsibilities on their plate, while banks will need to change up how they do credit assessments and make sure they're following the conformity standards. Basically, everyone needs to act fast and make sure they're not using any AI in a way that's not allowed before the December deadlines roll around.

AI for Central bankers

And EU wasn't alone. Even BIS released a paper in mid 2023 about how central bankers should be incorporating AI into their workforce. And they covered? broadly 4 areas: (i) information collection and statistical compilation;

(ii) macroeconomic and financial analysis to support monetary policy;

(iii) oversight of payment systems; and

(iv) supervision and financial stability.?

This means, all those Fintech which were operating in grey areas, can presumably no? longer hide. Or even if they do manage somehow to avoid scrutiny, they won't be overlooked for much longer. Because now with AI, central bankers can identify, in real time,? who is circumventing which guideline, and what the impact of that could potentially be on the system.?

That is it for this edition of the Fintech Chronicler. I will be back next week, and we shall dive into the world of card networks, Visa and Mastercard, to make out what the anti trust settlement means for them as a business, and for their global customers as well. See you next week !