AI in banking industry: The relentless rise of 'machine bankers'

You call an hdfc bank call centre.

Lady AI responds: Namaskar, welcome to HDFC Bank.

You: Mera credit card balance batayein. (What is my credit card balance?)

Lady AI: You want to know your credit card outstanding balance, is that correct?

You: Haan. (Yes.)

After confirming your identity, Lady AI announces your card balance to you. If the call drops midway and you redial, you are reminded of the last interruption. You want to pay bills? Lady AI will instantly send you a payment link via SMS.

Job done. No human intervention.

Customer service in banks has evolved from traditional call centres to interactive voice recordings (IVR) to AI-powered bots. And the latter are getting smarter by the day. Lady AI (an imagined name) is part of a new generation of AI bots with natural language skills. They are trained on large data sets of customer interactions, they work 24x7, and they are learning fast with ML algorithms to recognise you and your banking needs. “Today, a small percentage of calls are exceptional and complex, and get directed to human agents,” says Anjani Rathor, Chief Digital Officer at HDFC Bank.

Clearly, the foundation of a strong AI system is good data, and so the banking industry is working on a war footing to consolidate data from various sources into a single location. “Banks are now creating enterprise data lakes for data consolidation, processing and modelling on a real-time, on-the-move basis,” says Balaji V.V., CTO of ICICI Bank. “This will not only enhance our AI tool’s capabilities but also simplify data management.” Banks are already using AI to personalise offers, underwrite credit, manage risks, and for analytics.

While AI helps a bank make personalised offers, pre-approved- and pre-sanctioned loans to existing customers, work is on to offer the same to new customers or even people with no banking history. How does that happen? “There are currently thousands of customer profile categories created by banks for identifying customer segments having similar requirements when making offers, which group a particular set of customers into specific profile buckets based on data available to the bank,” says Balaji. The acceptance or rejection of an offer provides additional data points to refine the system. “It continuously learns on the go,” says Rathor of HDFC Bank, adding that as AI tools evolve, they will likely incorporate external sources of data as well.

AI is also used in decision-making for loans, where algorithms combine a bank’s conventional underwriting model with its intelligence based on credit data to take decisions. “This technology helps reduce the risk of default and enables banks to make more informed lending,” says Sriram Srinivasan, Chief Digital Officer of Ujjivan Small Finance Bank. At the same time, a private banker points out that private sector banks have reached around 50–60 per cent automated decision-making. The rest are still based on traditional methods. Flagging of risks is also on the radar. Let’s say you usually swipe your credit card to buy food or groceries. One night, if the system notices, say, a late-night swipe to buy gold, something you haven’t done before, you will immediately get a verification call.

AI is helping insurance companies, too. AI tools use customer-uploaded images of a car, for example, to give a first-level assessment of an insurance claim. “We are working with many insurance companies, and the adoption is decent,” says Geeta Gurnani, IBM Technology CTO and Technical Sales Leader, IBM India & South Asia. “The efficiency of damage claims under this AI model has improved by 40-70 per cent in terms of accuracy.”

Will it impact jobs? Gurnani says only certain mundane tasks will shift to AI in terms of job requirements. Adds Rathor of HDFC Bank: “Low-skilled, repetitive work and very simple tasks will move on and get done by systems, computers, and technology. And people will evolve into bigger things.” Banks are already hiring data engineers, data scientists, UX designers, etc., which was not the case some years ago. “Any technology is not going to replace people just like that. But the workforce may not increase proportionately,” says the tech head of a private bank.

AI tools also come with their own biases. They are like black boxes, and no one knows how the machine decides. “Consistency [in outcomes] is critical in building trust and confidence in AI technology, as it helps to ensure that the tool is reliable and accurate,” says Balaji. Tomorrow, there will be a requirement for an audit of AI models. “We have launched a few products that are explainable by AI. This falls under the entire area of AI governance, which means whatever models you are creating can be audited and investigated at any time for their robustness, fairness, and explainability,” says Gurnani.

AI still has some way to go in banking. They can’t enable transactions, for instance. Plus, a banking tool equivalent to a ChatGPT, which learns from everything available on the internet, would help banks know their customers better. “The more data we have about you, the more intimate we will be as a bank,” says Rathor.

@anandadhikari