It is ironic that while banks are closing branches and moving to automated, self-service channels they are also trying to offer personalised services to their customers at an individual level. In the past, relationships were developed and deepened with local bank branches staffed by people whose role was to get to know their local customers. However, as customer preferences evolved and technology improved, branch closures were justified by reducing customer usage, increasing numbers of digital transactions and the need to optimize operational cost. As the “human element” fades away, replaced by digital, how can banks enhance their service quality and deliver personalized service while keeping costs down?
The use of Artificial Intelligence (AI) in banking and financial services could prove to be a game changer. As big data continues to gain traction in many industries and as the focus on effectively leveraging micro-level insights increases, traditional approaches for acquiring and servicing bank customers are being transformed. Consumers’ growing use of digital channels is creating ever larger digital footprints, which can be used to identify meaningful patterns and suggest new approaches. AI has been deployed in wide range of areas, including, autonomous credit scoring, digital assistants (Chatbots), enhanced expense reporting and personalized servicing. In a recently released research report, UBS calculated that even in an “optimal” scenario where disruption from AI is limited, banks could benefit from a potential uplift in revenue of 3.4 per cent and cost savings of 3.9 per cent by 2020.
Let us consider the case of lending. For example, by taking data from traditional credit bureaus and non-traditional sources such an individual’s banking transactions, information from their social media accounts, earning and spending patterns, friends and family history, telecom bills and processing it through an AI system, extremely comprehensive credit profiles can be created. Lenders can use these kinds of credit scoring approaches to offer easy credit to sub-prime segments of the market, but crucially, do it while reducing their own exposure. With Deep Learning, machines can be “trained” to make autonomous credit decisions after they are exposed to a huge number of diverse data points including past decisions, credit policies, risk appetite, various rules, regulations, eligibility criteria and complex scenarios. Using their own intelligence and based on an individual’s AI driven credit profile, the intelligent machines can now make credit decisions at speed and accuracy which humans can hardly match. These machines learn continuously and improve their performance while handling newer cases. Deriving meaningful insights from data can help identify exact requirements at the most opportune time and other preferences, which in turn can be shaped into tailor-made loan products and services. Also, these machines would be far better at picking up the anomalies in applications, data points and behavioral patterns – leading to the early identification of all kinds of possible frauds. And of course, the AI never sleeps.
The chairman of the UAE Banks Federation recently mentioned that Artificial intelligence and digitization can save up to 30 to 50 percent costs for the banking and financial system. According to an Accenture report, 86 percent of bank executives agree that the widespread use of AI provides for a competitive advantage beyond cost. Which bank would not love to have an intelligent machine, which can take perfect loan decisions on its own, 24 hours a day, 7 days a week, in nano-seconds? A new study by Juniper Research predicts that unsecured consumer loans that use Artificial Intelligence or Machine Learning technology will jump to 960 percent over the five year period (2016-2021) and potentially becoming a US$17 billion business for FinTechs by 2021.
However, as with any new technology, we can’t be certain that Artificial Intelligence will make “perfect” decisions. Can these super-fast machines really never make a mistake? As the algorithms become more complex, will the chances of error increase or decrease? And as the complexity increases, how can humans be sure that the logic is accurate? Even if they are “perfect”, will these intelligent systems be able to sustain the two-way ‘trust’ that exists in banking today? The machines are only as good as the data they are fed, and what if that data is wrong, or out of date or incomplete? Imagine a situation where an individual is deemed not creditworthy by the machine based on certain specific actions which took place a long time ago, actions that a human would consider and then discount? Alternatively what happens if someone stage manages specific actions to fool the system and boost his scores. Is there a possibility that the machine may inherit some biases based on the personal preferences of the individual/team that was involved in the learning exercise? Will there be transparency in the logic used for these decisions (very difficult in Deep Learning)? How will regulators ensure that organizations have safeguards against any kind of discrimination in the criteria used by these machines? What about the cyber security risks associated with AI? What if people running various models don’t really understand why their models are behaving erratically in certain situations? What happens when people feel this widespread usage of data invades into their private lives? How will a threshold be defined to avoid getting into this situation? What happens if regulators impose restrictions on private data usage in AI? In the words of Financial Stability Board, “The rapid use of artificial intelligence in banking could trigger financial stability risks and some unexpected surprises unless proper testing and training is put in place”.
While it is obvious that AI has tremendous potential for banking and financial service industry, there are still many aspects which need to be carefully considered and analyzed before it is given free rein. Relying on machines for sensitive decisions which impact humans critically is a delicate issue and banks would like to trudge extremely carefully here. Having said that, almost every breakthrough in technology goes through this cycle and this is no reason to cast doubt on the value that AI could unlock for both the banks and customer.