Infosys is a global leader in consulting, technology and outsourcing solutions. We enable clients, in more than 30 countries, to stay a step ahead of emerging business trends and outperform the competition. We help them transform and thrive in a changing world by co-creating breakthrough solutions that combine strategic insights and execution excellence.
A few days ago, some Facebook bots defied the rules to create their own language, lending credence to the view that without the right checks and balances, Artificial Intelligence (AI) can go out of control. Big businesses, while mindful of the risks, are betting on the technology, believing there is enough evidence that it will work, and work well.
Cautious by nature, even banks are slowly coming round to this view. Most banks today are implementing AI, or plan to, in areas such as customer service, fraud detection, personal financial management, trading and compliance. The goal is not only to automate routine, repetitive transactions to make them more efficient, but also to automate some decisions using artificial intelligence.
Just to make things clear, artificial intelligence is a broad term denoting a group of related technologies, namely, big data analytics, machine learning, deep learning, predictive and prescriptive analytics, natural language understanding, visual recognition, smart virtual assistants and chatbots, and robotic process automation. Let us take a closer look at some of these components of the AI technology stack and their use cases in banking.
Chatbots and Smart Virtual Assistants Among the most popular AI technologies, a chatbot is a software application found on websites or messaging platforms that can converse with human beings in natural language. At Swedbank, a chatbot called Nina handles about 40,000 customer calls each month and resolves 81 percent of them successfully. When RBS launched Luvo at the end of 2016, they expected the chatbot would straightaway take over 10 percent of online customer queries to give their advisors more time to attend to complex issues. India’s first smart virtual assistant, Eva, made a spectacular debut, answering more than a hundred thousand queries from HDFC Bank customers from 17 countries within the first few days. While easing the burden on customer service agents can be one goal, banks should be thinking ahead about using the learning from chatbot-customer interactions to build further intelligence, and leveraging it to present cross-sell offers, make proactive suggestions and even initiate timely action when necessary.
Banks have mined data for insight for decades. However, the emergence of machine learning has exploded the analytical possibilities. There is no area of banking operations where machine learning cannot add value. In retail banking, machine learning can refine customer understanding by identifying patterns from hundreds and thousands of variables, ranging from demographic indicators and spending patterns to transaction frequency and personal consumption preferences. Patterns learned from the behavior of a cohort of customers can be progressively extended to a segment, product, or geographic region as the machine expands it own knowledge over time. The bank can use this to offer new value and services to customers and thereby improve engagement. Take a simple example of a monthly rental payment that is made to a landlord’s bank account before the 10th of every month. So far, banks have never taken any notice of a default. But machine learning misses nothing, and spotting a deviation from the regular pattern of activity, can immediately alert the customer to investigate the matter.
Deep learning is a subset of machine learning that uses artificial neural networks to make intuitive connections between various types of data, while taking into account the present context. Where machine learning can identify patterns from vast quantities of data, deep learning is capable of more complex non-linear analysis in a particular context. For instance, it can expand the analysis of customers’ spending data by associating similar transactions with the customers’ choice of channel, device, location, authentication mechanism, point of purchase, time of purchase, transaction history, and so on, to discover subtle insights which can be leveraged in many ways. Deep learning also makes predictions based on what it has analyzed using deep neural networks.
Robotic Process Automation:
The most popular roles for Robotic Process Automation within a bank are improving cost efficiency and productivity, and reducing fraud. RPA takes over routine tasks and processes so that bank employees can attend to bigger things, which are outside the current capabilities of AI. From the customers’ point of view, the use of robotic automation in common processes, such as account opening, means faster turnaround, lower cost, and happier experience.
There are innumerable examples of RPA in banking, of which we will mention just two. JPMorgan Chase employs a program called COIN (Contract Intelligence) to interpret thousands of commercial loan agreements, which an army of lawyers and bank officers would take 360,000 hours each year to do. COIN reviews documents in seconds, and what’s more, makes fewer errors than human beings. Since the past one year, India’s ICICI Bank has used RPA in more than 200 business processes to reduce response time by as much as 60 percent and take accuracy up to 100 percent.
The next step is to take RPA outward, and extend it to interactions with partners, regulators and the larger financial ecosystem to maximize benefits.
Although the beginnings of Artificial Intelligence can be traced to the 1950s, its evolution is much more recent. AI is still in its early stages, which means banks – traditionally tech-laggards – have a chance to plunge right in and demonstrate their leadership. While there are risks to AI, banks cannot be deterred by them. Actually, with AI, they have only one choice, and that is to go right ahead.
By Rajashekara V. Maiya, Associate Vice President & Head – Finacle Product Strategy