AI voice calls have come a long way from sounding like a robot to being able to have a conversation. But as they begin to sound more human, businesses face an operational blind spot: ensuring that these AI voice agents understand the customers correctly and do not make promises that haven’t been authorised.When human executives handled debt recovery or customer support, they exercised judgment and empathy—gauging whether an overdue borrower was facing genuine bereavement or financial distress or understanding what the customer is really looking for. Autonomous bots strip that discretion away.A Y Combinator-backed startup, Cekura, founded by IIT Bombay alumni is working to address a critical gap. The conversational AI reliability startup tests voice bots before they are deployed and monitors their live interactions to identify errors and improve performance. These AI agents often fail to understand either the language you speak or misinterpret what you are trying to say. Customers may shift between languages even mid-sentence, have different dialects or may be calling from a very noisy or busy street. There is a chance that the AI agent will misinterpret or misunderstand the customer’s need.“Cekura is the reliability layer that helps banking, financial services, and insurance (BFSI) companies scale their AI voice agents with confidence. We test every agent against real-world conditions before it goes live and monitor every call after, so banks and insurers can expand to new languages and higher volumes knowing their agents will get it right,” said Janhvi Nandwani, founding member at Cekura.A bot picking up ambient chatter and offering an unauthorised offer can be a major issue. The risks are way beyond poor customer experience. A voice bot that mishears a response or makes a commitment its employer has not authorised could lead to customer complaints, financial losses or regulatory scrutiny.As a company owner who has an AI agent making thousands of calls a day, this is a serious concern. “Now imagine (as a company owner) you have a system up and running which is making thousands of calls a day, but you don’t know what this robot has said,” Nandwani pointed out. “If you have to sit and listen to those calls, it’s counter-productive,” she added.A reliability tool can solve these issues, says Nandwani, an IIT-Bombay alumnus. The stakes are particularly high in India’s BFSI sectors, where heavy call volumes are expanding beyond English and Hindi into regional languages like Odia, Tamil, Telugu, Kannada, Bengali and Assamese, among others.The next challenge, Nandwani said, is to reach a point where customers are no longer hesitant to talk to an AI agent, because it sounds human, understands what they need and doesn’t make the mistakes that erode trust.Over the next three to four years, she expects voice agents to become fluent enough in local accents and human conversational cues to make that distinction increasingly difficult.
About The Author
