Changes are underway in the financial services sector: a fusion of Wall Street smarts with Silicon Valley technology. India's largest private sector lender, HDFC Bank, has ambitions too.
One way of understanding the importance of artificial intelligence and machine learning in Indian banking is to talk to the country’s largest lender, the State Bank of India. The 211-year-old bank has been through epochal changes from 1806 when its earliest predecessors were formed.
Arundhati Bhattacharya, who led the bank until two months ago, thinks artificial intelligence is the most important change staring at banking, as we know it. “It helps in how to interact with whom in which way from giving responses to queries. It helps in ensuring certain patterns,” she says. “Analytics, robotics and AI will go hand-in-hand.”
Bhattacharya’s comments lend context to the changes underway in the financial services sector: a fusion of Wall Street smarts with Silicon Valley technology. From Robo advisers to trading floor robots, underwriters and automated assistants, banks have already deployed leading edge tech.
Changes are afoot at India’s largest private lender as well. The HDFC Bank is doubling down on machine learning and artificial intelligence to change the way the bank functions internally and with its customers.
HDFC Bank, India’s second-ranked bank by assets, has divided its digital ambitions into three: build capabilities for a comprehensive digital bank; make it easier for its customers to do business with them; and to increase its digital offerings.
“The next step was to make it more personal, contextual – that is where the role of AI comes in,” Nitin Chugh, country manager of digital banking at HDFC Bank says. “With AI, we are building a full stack – both within the backend and for customers. Once you have the stack you can play in whichever way you want,” says Chugh.
By stack, he means that almost all services — about 90% of them like disbursing loans, payments, tracking cyber attacks, hiring, customer experience, personal banking, among others — will have components of AI and ML.
In March this year, the bank started deploying chatbots to interact with customers. But Chugh says that a chatbot is useless without AI built into it. “Can we have a conversational chatbot, which does not only do QnA, but is more personalised, and can that be used for transactions and have a voice,” says Chugh.
To begin with, Eva, the bank’s chatbot now has a voice that’s powered by Amazon’s Alexa. The prototype built on Alexa, Amazon’s AI-based assistant, was shown to Managing Director Aditya Puri in October. With a voice bot, Chugh says that options are unlimited. “The same thing can be simulated on the dashboard of a car, Echo at home, wherever you want to have a voice interface,” he says.
For now, Eva only reads content from a website and is able to tell you if there is a bank at a particular location or give information on loans, mutual funds and fixed deposits. “Quickly it will recognise you, and ask you what do you want to do today,” said Chugh.
Other banks aren’t far behind when it comes to bots. ICICI Bank is deploying 10 different bots — for NRI customers, remittances, finance, sales, lead generation, among other functions. Others like SBI, Yes Bank, HSBC are looking at various use cases of chatbots and AI. SBI’s bot, Intelligent Assistant, will be used to answer customer queries. In January Yes Bank partnered with GupShup to launch its own version of a chatbot. HSBC had introduced its chatbot Olivia in the UK in 2011, but in India, they launched it recently.
For HDFC Bank, the next generation bot will be intelligent enough to understand human language, in English and in Hindi, even if there are spelling mistakes. If a customer writes pay, it will ask what do you want to pay: an electricity bill or transfer money to another account. “They need to be supported by machine learning and deep learning, and they should have natural language interface. All four of them need to come together,” says Chugh.
Eva is not only limited to paying a bill or transferring money or fetching information. It is can also remember when you paid a bill the last time. In the future, it will remind you when the time comes to pay the bill, and if you don’t have enough balance it will ask you to transfer money to the account.
The bank is working with many startups, especially in the fintech space to deploy some of these solutions. Its chatbot on the Facebook messenger is powered by Niki.AI and Bengaluru-based Senseforth powers the bot on its website. IRA, its humanoid was built by Kochi-based startup Asimov Robotics.
Shridhar Marri, the founder of Senseforth says that Eva started as an experiment but since its deployment early this year, it has answered close to four million queries for 750,000 unique users. He adds that Eva’s accuracy is 86% with an uptime of 99.9%. That would be better than most humans.“In its second phase it will be able to make a demand draft, transfer money…,” says Marri.
The bank is still early in its journey to tap into newer technologies, but stock analysts like the sound of it. “Other banks also have similar technologies but HDFC Bank has much implementation and execution compared to its peers… HDFC has a wider customer base so it can use it better than other banks,” says Alpesh Mehta, banking analyst with Motilal Oswal brokerage firm. The brokerage firm has given HDFC a ‘buy’ rating, which means the analysts believe that the stock price will go higher up.
Mehta says that instead of the customer going to the bank, using data analytics, HDFC finds out which customer might need a loan and approaches them. This it does by looking at the search and enquiry patterns while the customer is interacting with the bank through the website or mobile app or even on social media.
In some cases, Chugh explains, where the loan is pre-approved depending on the customer’s’ profile, the loan can be disbursed in 10 seconds. In other cases, it can be done in 48 hours after taking the right approvals. Five years ago the same process would take 10-15 days. “AI is playing a big role in making this happen,” says Chugh without disclosing much. It took 18-24 months to make this happen, he adds. Previously, a loan would happen when someone asks for it, but with now using data, the bank is able to look at the transaction patterns and tell how much loan can be given to a person. It does not require any additional paperwork, and the loan is automatically linked to the bank account.
There are a large number of use cases in lending ranging from cracking down on fraud to managing risk better. These were mostly done manually with some help from computers before. Now, it has a layer of artificial intelligence. In lending, AI is being used for pulling out information from big data, making sense of it and try to improve our own credit score of customers.
For example, in two-wheeler loans, there are some models where the default rate is much higher. The traditional practice is to reduce the exposure in those models, but that is also a loss in business. “So you can say, a particular model, in a particular geography for a kind of customer profile, has a lower or a higher default rate, so let me use different variations,” says Chugh.
With data, Chugh says, a lot of AI can be used to sharply identify customers, create buckets where fraud can happen more, understand which segments to avoid and assess risk in certain segments. These get better as the machine solves more of these queries and problems.
The bank also uses AI in non-core areas like hiring. It uses AI in screening and psychometric analysis of the applicants. For example, if 100 people have applied for a job, the AI-assisted bot would weed out anything between 70-80 people. Previously every application would have to be seen by a human, which would delay the process of the final interviews. After the bots do the screening, final interviews happen. But hiring is not completely done by the intelligent virtual machines. Chugh said that the filtration is done by an AI bot, which goes through their the application form, it also uses a video and social media filter for psychometric analysis, and filters down to the right set of candidates.
Aditya Puri, managing director of HDFC Bank said in an interview to Boom News earlier this year that he wants his customer’s experience to be like using Netflix and Uber — that everything should be one click away. Chugh demystifies the analogy. “Netflix means different things to different people. It knows what shows I watch, things I can choose, I can watch a certain category of content. It tells me what shows are coming up, and it’s very easy to navigate, it easy to find. That is exactly the journey we are on,” says Chugh who has set up a “fintech team” to develop new features for the bank.
“Our intelligent notification engine, which just uses personalisation and looks at things you are doing on the digital channels,” Chugh says. By digital channels, he means, using a Facebook Chat application or the net-banking or mobile-banking platform.
The bank has also deployed an assistant for the knowledge bank, for employees. Knowledge bank has all the circular and government regulations. Many times it becomes impossible for an employee to find the circular that is required. The assistant, in form of a bot, pulls out the data and the required circular — it comes in very handy if a customer is waiting at the table.
There is more. “We are also using AI to make sense of what customers are doing in portfolio management, which we would have done traditionally also, but on a statistical basis. Now we are moving to an algorithmic basis, and then predict what should be the next best thing for the customer,” Chugh says.
Here is how it would work. For example, three people might be coming from the same strata of society, with similar customer profiles, but might need different things from the bank. AI will help HDFC make those differences. For example, in credit cards — if customer “A” is a loves dining out, HDFC will be able to give a card which is offer discounts on food, a Jet Privilege card to a traveller, and a regular card for a regular user.
“AI helps to cut down manual intervention, thereby reducing errors. Add to that, the neural network trains itself every time a transaction or an interaction happens. And then AI helps in reducing cost,” says Neil Shah, research director at Hong Kong-headquartered Counterpoint Research.