What can you do with AI in Financial Services?

Artificial Intelligence (AI) is probably the biggest buzzword in the business and finance world. Yes, even more than blockchain. AI is not particularly new and can trace its roots back to the mid 1950s. However, it is true that developments in computing over the past 20 years have led the field to progress tremendously. It also helped to move AI from the academia to the business world. This has happened through practical applications, including in financial services. It is often misused and confused with simple rules-based mechanisms. However, there are plenty of use cases for AI in financial services. The applications are almost endless but let’s have a look at the most developed and popular ones.

First, let’s take a step back. What do we actually mean when we talk about AI? Spoiler alert, we are not talking about Terminator here. Or some kind of evil computer taking over the world to dominate submissive humans like in the Matrix. Artificial intelligence is a wide-ranging branch of computer science. It is about building smart machines capable of performing tasks that typically require human intelligence. Although recent developments have been incredible, we have not reached general artificial general intelligence yet. That is a computer being equal in every way or surpassing a human’s cognitive abilities. As you can imagine, artificial intelligence can become very complicated, very quickly. But from a business perspective, what can you do with AI in financial services?

Risk management

As the financial industry became more complex, so did the job of managing the risks. Especially when it went from being a mostly paper-based business and a few guys shouting on a trading floor, to automated computer programmes working 24/7. Accurately predict and manage risks has become very hard without relying on automation and sophisticated artificial intelligence models. 

And that’s normal, there is only so much data a human can analyse. Unless you throw staff at the problem, but banks cannot really do that as they desperately try to cut costs to compete with nimbler fintechs. That’s where artificial intelligence can come in handy.

AI models can be really useful to large financial institutions to build complex models able to go through a large amount of data. On the long run, it is much cheaper than having hundreds of risk people. Machine learning allows for models that can predict with much more accuracy than a human what will happen in a multitude of scenarios. They learn from being fed a huge amount of risk data to try and forecast what will happen next. Natural language processing (NPL) models can also read through legal documents and identify the clauses that may be a problem. Or look through a loan book to identify potential weaknesses. AI is particularly relevant for sprawling FIs that are operating across several segments of the industry and many countries.

Credit

Credit is central to an individual or a business life. Building credit is really important as explained in the 5 finance skills to better understand fintech. The way financial institutions have historically assessed credit is using a limited set of data. Banks needed a standardised process that could be carried out by many credit analysts at the same time. But what it really means is that you are not looking at the full picture. Someone can do well on 4 or 5 criteria, however the underlying risk could be higher because you do not take into account something that does not fit in a box. Conversely, a lot of people have been marginalised because their situations do not fit well in the current credit assessment models.

AI can be used to improve underwriting decisions by utilising a variety of factors that more accurately assess a customer’s financial situation. Instead of relying on traditional data points, you feed the credit model with alternative data. For instance, instead of using years of financial history, use only a few months and predict the likely customer behaviour. AI models can also incorporate things like employment opportunities or ability to earn, which traditional scores do not capture. This is particularly crucial to traditionally underserved borrowers in the credit decision-making process.

Chatbots

Controversial is maybe not the right term but there are advocates and opponents of using chatbots. Using an AI-powered chatbot can be a powerful tool to help simultaneously many customers and deliver a constant quality of services. However, it can frustrate even more customers that have a problem and are looking to speak to someone and not a computer. So we actually have seen companies using the fact that they have real-life people in customer support as a selling point. On the other hand, the machine learning model will also need a huge amount of data to be really helpful to customers. Thousands of words and their synonyms to be able to understand customers queries. But the more it answers questions, the better it gets at it.

Fraud / AML / KYC

Every day, huge quantities of digital transactions take place as users move money, pay bills, deposit checks, trade stocks and do more via online accounts and smartphone applications. More and more data is created every day. It is exponential: over 90% of all data on the internet has been created since 2016. 

An increased complexity that financial institutions have to deal with. Because as technology evolves, so do the criminals using it. Money laundering schemes become more elaborated, it becomes easier to falsify an identity, so forth and so on.

The need to ramp up cybersecurity and fraud detection efforts is therefore now a necessity for any bank or financial institution, and artificial intelligence is playing a key role in improving the security of online banking. Again, AI models can learn many possible fraud scenarios. You can partly automate KYC through machine learning, for instance to match someone’s ID with a selfie. It is lowering false positives and human error, which again has the possible to make finance much more inclusive.

Advisory

Personal Financial Management (PFM) tools can be greatly improved through the use of artificial intelligence. At the moment, most of these tools are focusing on using past data to predict when for instance you are likely to run out of cash. Many falsely claim to be AI, when they are in fact simple rules-based mechanisms. But others do use machine learning.

Investments is an area that is being transformed by AI. Heard of robo-advisors? Well, they are all rage these days. And for a simple reason: it is bringing advisory and investment management to the masses. Very logically, most people were not getting financial advice because, well, they had not enough money. Maybe that’s not obvious to everybody but if it is not economically viable, it is quite normal that a bank or wealth manager would not do it. They are usually paid a small percentage of your overall investment. And you need to pay the guy who is investing the money, the infrastructure, etc. But through automation and using machine learning models to invest, it suddenly becomes incredibly cheap to mimic the best stock investors.

What next?

There are many more use cases with some being very specific and niche. Overall, AI is having a tremendous impact on the financial services industry. And what we have seen is only a tiny fraction of what is really possible. The cost savings potential for banks from AI are huge. And fintechs leverage the technology much more than incumbents. As they slowly eat market share, we will see more AI in financial services.

The applications are limitless because in theory, AI is. However, there are some limitations. At any given time, we should be able to have control over it and know how a decision has been made. Not because we think that some computer is going to take over the world. It is more about audit. The infamous “black box” dilemma: in financial services, you need to be able to explain how you came up with a decision because the regulator will eventually ask you. That’s normal, financial decisions have a huge impact on one’s life. Therefore, AI models have to be developed carefully, and we need to avoid biases at all costs. It can be an issue as inherent bias can be incorporated into a model unconsciously by the developers. Ethics at the centre of AI, that will be the next challenge.

Fintech Review