May 17, 2022

How AI is changing the finance sector forever

Geoffrey Shenk

People often describe new technology as “transformative” or “revolutionary”. Yet all too often, it simply doesn’t live up to the hype. AI is a definite exception. Yes, there has been a lot of frankly absurd hype around AI. But AI truly has the power to transform your business for good. In this article, I will look at how AI is revolutionizing the consumer finance sector and changing it for good.

A short history lesson

Way back, consumer finance was the preserve of the banks. The local bank manager had the ultimate power over who did or didn’t get offered credit. Invariably, you needed some form of security to raise any sort of credit. That began to change in the 1950s and 60s with the emergence of credit cards. Now, you could get unsecured credit that you could use to purchase anything you wanted. The consumer credit market was born and began to boom.

The basic economics of consumer credit are simple. You lend money to people who repay it with some level of interest. The trick is in selecting the right people and setting the appropriate interest rate. If you select people who are poor risks, they will default on some or all of the loan. If you set the interest rate too high, you may put off people who would be good risks. The risk is that you get your calculation wrong and then the bottom drops out of the market as we saw in 2008.  

Crunching the credit numbers

For decades, selecting who is a good risk has been done based on their credit score. This is a pseudo-scientific concept that combines a consumer’s repayment history, credit exposure, and various other factors. For instance, have they applied for credit recently? The result is a number reflecting the supposed risk of lending to that individual. There are a few problems with this approach, not least that you need existing credit in order to receive credit.  

Some puritans are very against the idea of credit. They instinctively believe that money lending is morally suspect or even sinful. This view is reinforced when financial institutions lend money to people that can ill-afford to repay it. But the big lesson of the 2008 credit squeeze was that without credit, money doesn’t flow around the economy. And without a flow of money, the economy simply grinds to a halt—a classic example of Keynesian economics.

So, how can banks and other financial institutions ensure they lend to the right people? Who even are the right people? This is just one case where AI can transform how the finance sector works.

A bit of background

In this context, AI isn’t about computers taking over the world, or sitting back while your car drives itself. Rather, it means systems that take your data and identify hidden patterns and insights. In turn, these insights allow you to make more informed decisions. These systems rely on a technique called machine learning or ML. In fact, this is the basis of almost all real-life applications of AI.

The three types of machine learning

I’m definitely no data scientist, so here’s my simplified take on how ML works. Machine learning comes in three flavors.  

  • Supervised learning. In most cases, ML uses a technique called supervised learning. Here, you use known data to teach the computer to identify interesting patterns or features. It can then go on to identify the same features in data it hasn’t seen before. This is basically how we teach children to speak or to read.  
  • Unsupervised learning. By contrast, unsupervised learning is trying to see if there are any features that might be interesting. For instance, does the data seem to form clusters? Or are there obvious outliers? This is similar to the way that scientists come up with new hypotheses.  
  • Reinforcement learning. Finally, reinforcement learning is when the computer learns by trial and error. When it gets things right, it is “rewarded”, but when it gets it wrong it receives a “punishment”. This is really similar to a child learning to speak.  

In practice, many real-world AI systems will combine more than one of these approaches to deliver better results. When you apply one of these ML approaches in real life, you get “artificial intelligence”.

How can AI help in the finance sector?

All AI systems rely on data to work properly. However, not all data is equal. Often, it is too noisy or messy to be really valuable. Finance is something of an exception. There is a vast quantity of data and it all has to make sense in order to comply with regulations and to pass audits. As a result, the whole sector is ripe for AI transformation. Here’s just a handful of examples.

Credit scoring

As we saw above, credit scoring is basically a black art. So much so that many companies choose a much simpler approach. Calculate the overall risk of loan defaults and make sure your interest payments more than cover this cost. Then lend to anyone that meets a basic set of sanity checks e.g. they have a job, they pass KYC checks, etc. AI can help in two ways here. Firstly, it can help you more accurately assess a customer’s individual credit-worthiness without needing to know their detailed financial history. Secondly, it allows you to correctly set your interest rates and charges to ensure you will maximize profits and minimize losses.

Loan default prediction

Credit scoring is all about knowing who is a good credit risk in the current climate. But most credit institutions make loans that last for years if not decades. Over that time, many things can change. In order to understand their exposure, these institutions have to be able to predict which loans are at risk of becoming delinquent and which may even default completely. AI allows you to accurately identify loans that are at risk by taking into account all the factors that might come into play. These include macroeconomic factors like the overall state of the economy, interest rates, etc., as well as the repayment history. However, the AI can also look at similar loans and use these to help its predictions.  

Fraud detection

Fraud is one of the biggest risks in the finance sector. This ranges from fraudulent use of stolen credit cards right up to fraudulent stock trading and market manipulation. AI is perfect for identifying transactions that might be fraudulent. Its real power comes from being able to spot patterns across a diverse range of data. This means it is able to spot correlations that humans simply couldn’t. Or couldn’t without investing significant time. For instance, credit card companies are suspicious when a cardholder makes a transaction in a foreign country. Often they will issue a temporary block on the card. But an AI system would have known that the cardholder had recently purchased flight tickets and paid for travel insurance. Both things that make it more likely to be a valid transaction.  


Finance has become the most heavily-regulated sector in the whole economy. Every year, thousands of pages of new regulations and guidance are produced by governments globally. Banks and other institutions have the unenvious task of keeping track of all these regulations and ensuring they abide by them. This is truly a super-human task since it requires absorbing such huge amounts of data in order to understand the regulations. As a result, we have seen a new branch of AI known as RegTech. This employs AI models to process and understand the regulations. Transactions can then be checked by the AI to establish if they are allowed or not. Unlike the other examples, this application is pushing AI right to its very limits. As a result, it still isn’t completely reliable.

The problem of AI debt

The issue for most companies is overcoming their AI debt. This comes in three forms. Firstly, data AI debt happens because your data is spread across numerous systems in different formats and structures. Getting all this data into one place and then actually understanding what it shows is a challenge. Secondly, technical AI debt comes when you have developed an AI solution, but then face the issue of deploying and monitoring it. This is often called MLOps and, like DevOps, typically comes with a large price tag. Thirdly, human AI debt arises because developing and running AI applications requires a large and highly skilled team. You need data scientists, mathematicians, AI engineers, MLOps specialists, to name but a few. Building a team like this takes time and (lots of) money. Many companies will try to sell you pre-built AI applications, AI platforms, or MLOps solutions, claiming they solve all your AI problems. The reality is that these only go a little way to addressing your AI debt problem.

How can Sonasoft help?

Here at Sonasoft, we have worked with some of the largest financial institutions. We have built a range of AI applications that deliver real results. This has given us a deep understanding of what AI can achieve in the finance sector. Our aim is to work alongside you to develop your own custom end-to-end AI applications and eliminate your AI debt for good. This will result in AI solutions that address your specific needs. We aim to make the process as painless as possible.  

Early on, our data scientists and AI engineers will sit down with your team to discuss how we can help and to do a sanity check on your data. If it all looks good, we will offer you a data feasibility study. This will explore all your data in detail and identify exactly how AI can help you. After that, we will develop a proof of concept. This will deliver you a production-ready AI application that can be deployed straight to your backend or can run in Sonasoft’s own infrastructure. If you like what you see we then license the application to you along. This includes a license for our AI-powered smart monitoring system. This helps solve the problem of MLOps, ensuring your application is reliable and continues to deliver good results. If this all sounds good to you, reach out to me today.

White Paper

SAIBRE AI Ecosystem

End-to-end AI applications that solve any business problem