How & Why Banks Use Artificial Intelligence
Artificial intelligence has already become a part of our daily life. Every day we use the most modern developments, without thinking about how it works. And these are not only such obvious things as searching for information on the Internet and smart feed in social networks. AI has penetrated many areas of our lives, including financial ones. Now banks are actively using this technology to communicate with customers and make their services more convenient and reliable. We will tell you exactly how credit institutions use AI and what the banking of the future will be like.
The banking sector has always sought to use the most advanced developments. So, back in the 50s, banks began to use special mathematical models for credit scoring, or assessing the creditworthiness of customers. By the 80s, technology stepped forward and computers appeared with systems that allow solving financial problems. Then banks began to use them, for example, to analyze and predict the dynamics of the Dow Jones industrial index and determine the optimal investment strategy.
70 years after financiers became interested in mathematical models, banks are everywhere using AI algorithms in their activities, analyzing data and predicting risks.
Artificial intelligence is a very broad concept. Today it is customary to divide it into two major areas. The first is General AI (or general AI). This is exactly what films about the future like The Matrix and The Terminator are made about: a kind of superintelligence that independently generates tasks and solves them on its own. But science has not yet come to this – although it is actively moving towards this and, according to various estimates, will reach the goal in about 30-40 years. The second direction is specialized artificial intelligence (or “weak” AI). Just it is used today in various tasks.
In this case, we are talking about technologies with the help of which companies solve individual problems in a narrow area. And when today experts pronounce the phrase “artificial intelligence”, in fact, they are talking about specialized AI.
Banks are actively developing and applying artificial intelligence technologies in various business areas. One of the main and complex tasks is risk management. Credit scoring systems, or credit rating systems, have become one of the key tools for managing risk in the banking industry.
Let’s see how AI scoring works in retail banking:
- Client fills out an application for one of the loan products (consumer loan, credit card, car loan or mortgage);
- Customer data is enriched by various sources (credit history, transaction activity, activity on banking products, etc.), and various statistical features are generated;
- The trained AI model receives enriched client data as input and forms a decision on issuing a loan at the output.
There is no universal credit scoring system; each bank independently creates AI models to perform this task. For training, a training dataset is required, such as historical data on loans that the bank issued in the last few years, and payments or delinquencies on them.
When trained on hundreds of thousands of historical orders, the AI model identifies various linear and more complex, non-linear patterns in the data, which makes it possible to predict the probability of default on new orders with high accuracy.
The more accurate the resulting model, the better the separation of “good” and “bad” applications occurs. And this allows the bank to issue more loans at an optimal level of risk, which brings additional profit. Therefore, banks are striving to develop more and more accurate models for solving the problem of credit scoring.
AI work is not limited to credit scoring. Another global challenge for AI is the task of anti-fraud, or fraud detection and counteraction. And here, modern AI models also allow you to effectively solve the problem.
Fraud can be divided into two types: application fraud (when a loan application is made without the intention of repaying it) and transactional fraud (when a fraudulent transaction occurs with a client). And if the first task is somewhat reminiscent of the credit scoring task (the same data are used, and the task itself is solved as a binary classification task), then the second one is very different from the data that is analyzed, and also has some additional difficulties in solving.
The task of recognizing transactional fraud has additional complexities.
Data volume
To train the model, transactions of bank customers are used, the size of this data is several orders of magnitude larger than the size of the data in the credit scoring problem.
Computational complexity of features for the model
To train the model, various statistical features are generated for the client’s transaction activity (average amounts of transactions by category in various historical sections (3/6/12 months), average, minimum and maximum values, and hundreds of other features).
Strong class imbalance
Fraudulent events make up a very small percentage of the total volume of transactions – less than one thousandth of a percent.
The bank must constantly maintain the high quality of the model in order to minimize the number of false positives. Since erroneous suspensions of operations during payments bring tangible discomfort to customers.
Such models make it possible to recognize fraudulent transactions with high accuracy and minimize false positives, which ultimately saves clients’ money.
Interbank Race
Why are credit institutions at the forefront of AI technology development? For two reasons: first, the banking environment today is extremely competitive. In addition to competing with each other, traditional lenders these days also have to compete with neobanks and technology companies.The second reason for the widespread introduction of artificial intelligence into many business processes within traditional banks is that technology allows them to get additional profit.
According to a McKinsey study last year, artificial intelligence could add up to $1 trillion in additional value to global banking every year.
The banking industry is a commercial field that aims to improve business performance and earn more money for shareholders. If the processes in the company are inefficient, then it becomes uncompetitive and cannot offer the client market conditions. As a result, he goes to other banks that work better. Artificial intelligence helps prevent this.
Thus, the introduction of AI in banks is also beneficial for customers of credit institutions. For example, the already mentioned scoring task, which helps the bank to identify unreliable borrowers, on the other hand, allows creditworthy clients to receive the loans they need faster and on better terms. In addition, AI increases the comfort of the client in interacting with his bank. And it’s not just the use of text and voice assistants. AI-based models enable better understanding of customer needs and preferences. For example, they determine at what time and in what channel it is most convenient for the client to interact and receive recommendations on the bank’s products.
Banking of the Future
Although AI is now widely used in the financial industry, there are still many complex and large-scale tasks facing banks, for which AI will be involved.In particular, it is necessary to integrate technology into financial services in order to make personalized offers to customers.
Probably, in the near future, a significant part of the processes in banks will take place without human intervention. The widespread introduction of AI is already underway, the level of automation of business processes and the increase in the efficiency of the bank as a whole are growing, and this vector of changes will continue to grow. In 5-10 years, banks will be automated as much as possible and AI technologies will be involved in absolutely all banking processes.
On June 22, 2022 the Gartner company published a research in which allocated 4 main rules of implementation of the artificial intelligence (AI) in the financial sector which allow to reach or exceed the expected effect and to provide achievement of critical results in the field of finance and business processes.
The use of artificial intelligence in finance is still in its infancy, and most companies have only begun to do so in the last two years. Most also fail to quickly get the expected returns from such projects. Given the early days of AI in the financial industry, CFOs lack a clear definition and strategy to successfully apply it. To help CFOs, Gartner has identified four critical rules for AI success in finance.
Hiring External AI Experts
Typically, there are three options for hiring people with AI skills and experience: hire new people, upgrade the skills of existing people, or bring in people from an existing IT department. Organizations that focus their talent acquisition strategies on attracting external AI-skilled employees are significantly more likely to become top financial institutions using AI.
IT professionals have indispensable professional skills in working with various technical nuances of AI, which allows the company to overcome the difficulties in working with AI applications and reduce the technical learning curve of other employees. Conversely, while improving the skills of finance staff may be less costly, it has the potential to slow progress and increase the likelihood of errors. In addition, new IT professionals provide an opportunity to go beyond the usual processes and settings, bringing new ideas to promote the adoption of AI.
Investing in Software with Embedded AI
Purchasing software with embedded AI makes it easier for companies to experiment with AI and use it in more areas of the financial industry; they can make it easier to create pilot projects to solve unique business problems. In contrast, building your own AI solutions for all financial processes creates a lot more work and reduces the opportunity to explore new pilots or use cases.
Conducting Exploratory and Varied Experiments at an Early Stage
Leading financial companies are taking an experimental approach to implementing AI at the start, rather than making a few big bets. With more early pilots, there are more use cases for AI, and adoption is faster as the organization can focus on the most successful pilots.
Typically, the most successful organizations are still exploring the same use cases as the less successful ones, with three being the most common: accounting processes, back office data processing, and cash flow forecasting. The only exception is customer payment forecasting, which is used by about half of the top companies, but very few of the less successful ones.
Choosing a Leader in the Implementation of Analytical AI
To realize the benefits of AI, CFOs must select the appropriate person to lead the implementation of AI. In particular, it may be the head of the financial planning and analysis (FP&A) department or the head of the financial analytics department, which will be engaged in the implementation of AI, and not the persons controlling them.
The leaders of FP&A and financial analytics are successfully meeting the challenge of implementing AI due to their strong analytic background and work with data. They rely less on knowledge of traditional financial processes and more on understanding the complexities of AI in the business environment.
Examples of Using AI in the Financial Sector in 2023
At the design level: forecasting the demand for banking products, predicting changes in demand, automated risk assessment.
At the production level: automation and optimization of interaction with existing and potential customers. Automation of document processing and loan approval.
At the promotion level: providing personalized offers at the right time. Automatic adjustment of interest rates depending on the history of the client.
At the level of service delivery: development of automated systems and self-service interfaces in all communication channels.
In the near future, Boosty Labs will begin providing banking solution development services.