If you’ve tried to get help from an Indian bank’s chatbot in the past year, you probably have a strong opinion. And that opinion is probably negative.

Banking chatbots have been around for almost a decade in India, but until recently most of them were glorified FAQ search engines. You’d type a question, get a response that was vaguely related but not actually helpful, and end up calling the customer service number anyway. The chatbot was a speed bump on the way to a human, not a solution.

That’s starting to change — for some banks. Others are still stuck in 2018. Here’s the current state of play.

The Good: HDFC Bank’s Eva (and Its 2025 Upgrade)

HDFC Bank’s Eva chatbot was one of the first in Indian banking when it launched in 2017. The original version was fine for simple queries but couldn’t handle anything complex. The 2025 overhaul, however, made it genuinely useful.

The upgraded Eva handles account balance inquiries, transaction history, bill payments, and basic product information without needing to escalate to a human. More importantly, it understands context across a conversation. If you ask about your credit card balance and then say “pay the minimum,” it knows which card you’re referring to.

Eva’s success comes from HDFC’s significant investment in natural language understanding models trained specifically on Indian English (and increasingly Hindi). The bank reportedly spent over INR 500 crore on AI capabilities between 2023 and 2025, and the chatbot was a major beneficiary.

The main limitation is that Eva still can’t handle complaints well. It can log them, but for anything requiring judgment or investigation, you’re transferred to a human. That’s probably the right approach — complaint handling requires empathy and flexibility that chatbots don’t have.

The Decent: ICICI Bank’s iPal and SBI’s YONO Assistant

ICICI’s iPal has improved steadily and now handles most transactional queries competently. It integrates with WhatsApp, which is a smart move in the Indian market where WhatsApp is essentially a utility. The ability to check your balance or request a mini-statement through a WhatsApp message feels natural in a way that opening a separate banking app doesn’t.

SBI’s YONO app includes an AI assistant that’s functional if not inspiring. It handles the basics — balance queries, transaction search, branch locator — and does so in multiple languages. Hindi support is solid; regional language support is improving but still patchy.

Both iPal and YONO’s assistant struggle with the same limitation: multi-step problems. If you need to do something that involves two or three sequential actions (like blocking a lost card, requesting a replacement, and updating the linked auto-debits), these chatbots can’t handle the full workflow. They’ll manage step one and then transfer you.

The Bad: Most of the Rest

I’m not going to name every underperforming chatbot, but the pattern is consistent across several public sector and smaller private banks. Their chatbots:

• Misunderstand basic queries regularly
• Provide generic responses that don’t address specific questions
• Loop endlessly when they can’t understand the user
• Lack integration with core banking systems (they can’t actually do anything, just provide information)
• Default to “please visit your nearest branch” for anything beyond trivial questions

The root cause is usually underinvestment. A banking chatbot that’s genuinely useful requires continuous training, regular updates, integration with multiple backend systems, and ongoing monitoring of failure modes. Banks that treat their chatbot as a one-time project rather than an ongoing investment end up with a tool that degrades over time.

Several banks have started working with external specialists in AI automation services to modernise their chatbot infrastructure, and the results from early adopters suggest that the technology gap can be closed relatively quickly with the right approach and investment.

What Customers Actually Want

This is where the industry often gets confused. Banks assume customers want chatbots because chatbots are cheaper to operate than call centres. But customers don’t care about the bank’s operating costs. They care about getting their problems solved quickly.

A KPMG survey of Indian banking customers in late 2025 found that:

• 72% had used their bank’s chatbot at least once
• Only 34% described the experience as “satisfactory” or better
• 61% said they’d prefer to speak to a human for anything beyond basic queries
• 83% said the most important factor was getting their issue resolved in a single interaction, regardless of channel

That last point is critical. Customers don’t inherently dislike chatbots. They dislike chatbots that waste their time and then send them to a human anyway. If the chatbot can resolve the issue, they’re happy. If it can’t, they’re frustrated because they’ve wasted time on the chatbot before reaching a human who could have helped from the start.

The Road Ahead

Large language models have dramatically improved what chatbots can do. The technology is no longer the constraint — implementation, integration, and investment are.

The banks that are succeeding with chatbots share a few characteristics:

They invest continuously. The chatbot is a product with a team, a budget, and an improvement roadmap. It’s not a project that was “done” three years ago.

They use their own data. The best banking chatbots are trained on actual customer interactions, not generic conversational datasets. They understand the specific ways Indian banking customers phrase requests.

They know when to escalate. Smart escalation — recognising when the chatbot can’t help and seamlessly transferring the customer to a human with full context — is more important than trying to handle every query automatically.

They measure what matters. Not “number of conversations handled” (a vanity metric) but “percentage of issues resolved without human intervention” (an actual measure of value).

Indian banking chatbots are getting better. But “better than terrible” isn’t a high bar. The banks that treat their AI customer service as a serious capability investment, rather than a cost-cutting gimmick, will be the ones whose customers actually prefer using it.