Robotics and AI in the finance industry
The finance industry has always been an early adopter of robotics. The ATM, which has been around for over fifty years, is fundamentally a robotic cash dispenser. Today ATMs can perform multiple functions and recently ICICI Bank in India has claimed to be the first bank to use a robotic arm for counting cash.
Robotic processes, usually referred to as robotic process automation (RPA), are employed in many aspects of finance around the globe. Software robots can do many of the tasks previously performed by people and are designed to reduce the burden of carrying our repetitive tasks while increasing efficiency and eliminating human errors. Almost any office function involving multiple tasks and across various platforms can be delegated to RPA.
RPA systems obey predefined rules created by people and implemented in software. The creative element is entirely human. They are incapable of carrying out any intelligent function or improving the processes they perform. In other words, they are unintelligent.
Used extensively throughout the finance industry including banking, insurance and investment, RPA implement processes such as compliance, credit card processing, reconciliation, credit checking of loan applications, filing, form filling and many other back-office functions. Their benefits are far-reaching and range from increased customer satisfaction, improved employee engagement, increased efficiency, and considerably reduced error rates. Humans, freed from spending their time on repetitive tasks, can devote their time to more creative and strategic tasks.
According to the Deloitte Global RPA Survey, 53% of respondents had already implemented RPA in their financial functions, and 72% stated their intention to use it in the future.
AI plus robotics is a game-changer
So far, the robotic processes we have addressed are merely rule takers. They implement human-designed methods faster and more efficiently than people can. RPA is often criticised for implementing flawed procedures. For instance, if the human-designed process is inefficient, cumbersome, or contains errors, then so too will the robotic process. Once implemented as an RPA, the process is likely to be set in stone with little or no opportunity for improvement.
But what happens if you add intelligence to the package? Can machine learning and artificial intelligence improve on the processes we currently carry out? In many aspects of finance, the answer is a resounding yes. Let’s dig a little deeper.
AI and credit scoring
While RPA is used in credit scoring, it follows pre-determined rules and has little or no flexibility. Your credit risk is assessed on your previous credit history, which can be bad news for people who don’t have one. Around one-third of the global adult population don’t have a bank account, and only 55% have an active bank account. With standard rule-based credit scoring, almost half the world’s population would be unable to raise a loan.
AI can provide a far more accurate assessment of creditworthiness. For instance, it can differentiate between a credit-worthy loan applicant who does not have a credit history, and an applicant with a sound credit history but a high risk of defaulting on the loan.
Such AI can delve into many other records such as smartphone data and social media, assuming the applicant provides appropriate permissions. Using machine learning, such algorithms can fine-tune themselves and improve their accuracy, benefitting both customer and lender. In the US, car finance providers report that the implementation of AI has reduced their losses by up to 25%.
Organised crime and state-sponsored fraud are becoming ever more sophisticated in their ways of committing fraud. According to the Certified Association of Fraud Examiners, they are using machine learning and AI to commit fraud that conventional fraud prevention systems are unable to prevent or even detect. An example of these sophisticated fraud exploits is intercepting the text messages used in 2-factor authentication.
Conventional fraud detection depends on rules and predictive algorithms that are no longer effective at detecting the advanced, sophisticated levels of fraud that are currently emerging. At the very least, fraud detection needs to be at least as effective as fraud attacks. The deployment of AI and machine learning to fraud management is so far low but is likely to increase over the next few years.
Such systems use supervised machine learning to differentiate between legitimate and fraudulent transactions; behavioural analytics to monitor the behaviour of every account holder and detect anomalies; simulated fraud attacks to learns prevention strategies; and adaptive algorithms that learn from additional data that is input.
Some examples of the recent implementation of intelligent fraud detection systems include: the Net Fraud Ensemble AI-powered anti-fraud system, said to be capable of tracking ever-changing fraud exploits in real-time, and Mastercard’s Threat Scan that simulates fraudulent attacks and preventing actual attacks before they occur.
AI and market trading
Given that machine learning algorithms can now beat the best Chess and Go players on the planet, is there any reason why they cannot beat the stock market? Indeed, such systems are the dream of many traders, and no doubt, substantial effort is being exerted on developing them. However, they remain a fantasy, at least for the foreseeable future. To an extent, the system is chaotic; just like the weather, small differences in the starting conditions can lead to considerable changes in the long term. While RPAs can be deployed to significant effect in a wide range of markets, predicting prices reliably, in the long run, is a bridge too far.
Robotics and AI will continue to have a huge impact on the finance industry. RPA will be used to automate a wide range of procedures much more efficiently and effectively than human beings, freeing people from the drudgery of retentive tasks so they can participate at a deeper level within the organisation. At the same time, AI, in conjunction with machine learning, will lead to increasingly sophisticated product and service offerings.