There was only one thing off in what otherwise turned into a smooth call: the customer\u2019s assumption of the agent\u2019s gender. The voice was not female; it belonged to an artificial intelligence (AI)-powered voice bot.<\/p>\n
That human-like bot was built by Yellow.ai, which manages sports goods retailer Decathlon\u2019s voice and chat AI, while the conversation itself\u2014including how the call connected, streamed and was recorded\u2014ran on virtual telecom operator Exotel\u2019s cloud telephony network.<\/p>\n
\u201cEarlier, our interactive voice response (IVR) could deflect about 25% of our calls. With voice AI, that\u2019s gone up to nearly 55-60%. Customers get the right information faster and our agents can focus on real issues,” says Dominic Raju, customer support project manager at Decathlon.<\/p>\n
\u2018Deflect\u2019 here means that the voice AI agent is able to solve the query and it doesn\u2019t need to be transferred to a human agent.<\/p>\n
The bot now handles everyday queries such as \u2018where is my order\u2019 and \u2018when will I get my refund?\u2019<\/p>\n
Like Decathlon, enterprises across sectors\u2014from e-commerce companies handling delivery queries to banks chasing loan repayments\u2014are rushing to automate their call centres with voice AI. It\u2019s an exercise that involves replacing or augmenting human agents with AI agents that can speak in multiple Indian languages, understand context and respond without any robotic delay.<\/p>\n
In order to enable this transition, a new industry is taking shape, bringing together global model providers, Indic language startups and India\u2019s decade-old cloud telephony companies. Exotel, Ozonetel and Knowlarity, known more for powering India\u2019s call-routing systems, have found a key role in this transition. These companies, which often come with deep integrations with telecom networks and regulatory clearances (from the department of telecom and the Telecom Regulatory Authority of India (Trai) for routing, number masking etc.), also bring in years of experience managing real-time voice traffic.<\/p>\n
The shift also marks a disruption in and evolution of India\u2019s large contact centre industry, which employed over 4 million people until 2024. From powering the world\u2019s customer calls, Indian enterprises are now evolving to build the infrastructure to power AI agents.<\/p>\n
![\"India\u2019s](\"https:\/\/datawrapper.dwcdn.net\/TIvXg\/full.png\")
![\"Shivakumar](\"https:\/\/www.livemint.com\/lm-img\/img\/2025\/11\/25\/600x338\/2-0-679955329-shivakumar-ganesan21-20160614-hm-0_1680347605016_1764067189748.JPG\" "\\"Shivakumar")
\n\t\t\t\t <\/source><\/picture>\n
\n\t\t\t\t\tView Full Image<\/span>\n\t\t\t\t<\/p>\n<\/p><\/div> To add to that, traditional phone calls run at the frequency of 8kHz, which is good enough for human ears but too coarse for AI speech models, which expect crisp 24kHz audio. Ganesan explains that Exotel\u2019s system automatically \u201cup-samples” the sound in real time, ensuring that the AI at the other end receives studio-grade input even from a basic mobile call.<\/p>\n These engineering changes are invisible to users but critical to make AI calls sound human. If latency\/delay eats even half a second, the voice bot\u2019s reply feels unnatural. If the audio fidelity drops, the speech recognition also fails. Also, if each minute of AI interaction becomes too expensive, as it is globally at \u20b9<\/span>7-9 per minute, India\u2019s mass-market use cases such as delivery confirmations or payment reminders are likely to collapse under cost pressures.<\/p>\n Traditional phone calls run at the frequency of 8kHz, which is good enough for human ears but too coarse for AI speech models, which expect crisp 24kHz audio.<\/p>\n Exotel\u2019s AgentStream, which was in the works for a year and launched this August, focuses on addressing these challenges. It powers companies such as Meesho, Cars24 and Sarvam AI, enabling thousands of AI-driven calls every day, says Ganesan.<\/p>\n \u201cWe went from zero to 120 to 130 customers in six to eight months and about $3 million in annual run rate (ARR). My guess is that close to 25% of all AI calls in the country today are probably running on Exotel,” Ganesan told Mint. \u201cThere is one NBFC (non-banking financial company) for whom we collected about \u20b9<\/span>3 crore a month, completely automated.” The founder calls this Exotel\u2019s \u201cNvidia moment” for voice AI.<\/p>\n From reducing latency to reducing cost, each company in India\u2019s cloud telephony ecosystem is solving a different piece of the voice AI puzzle. Shalil Gupta, managing director of Ozonetel, another early mover, points out that the rise of voice AI is changing the nature of the contact centre in India. \u201cIt\u2019s no longer a contact centre,” says Gupta. \u201cIt is now an experience centre.”<\/p>\n \n\t\t\t\t\tView Full Image<\/span>\n\t\t\t\t<\/p>\n<\/p><\/div> Gupta explains that customer interactions have expanded far beyond service related queries. Every touchpoint, starting from pre-purchase marketing to post-sale feedback, is now part of an ongoing conversation with a customer and has to be treated that way. Voice AI is becoming the first layer in this new architecture, he adds, filtering and resolving routine calls, while humans step in only for high-value, complex or emotional conversations.<\/p>\n Voice AI bots can fully handle simple customer requests, such as rescheduling an order. On the other hand, AI copilots listen to a conversation between two humans and help human agents respond faster and more accurately. For instance, during a bank support call, if a customer asks why a transaction failed, the copilot can pull up the customer\u2019s last three transactions, detect a possible limit breach and provide the agent the exact reason. Or, if a customer calls about a missing package, the AI assistant can show the human agent the order history, flag a delay notification from the courier and suggest options to reschedule or offer compensation.<\/p>\n At a higher level, AI-driven quality analytics can now evaluate 100% of calls instead of the 1-2% that used to be manually reviewed. \u201cAI is helping agents work smarter, not just cheaper,” says Gupta.<\/p>\n Like Exotel, Ozonetel also has a product dedicated to address the voice AI needs of its customers. Gupta says CXi Studio, the company\u2019s no-code platform, allows businesses to map out their entire customer journey, blending AI and human interactions. A delivery company can automate order confirmations and reschedules, while a bank can route loan queries or repayment reminders through voice bots trained in multiple Indian languages.<\/p>\n Gupta says Ozonetel saw this shift coming two years ago, when clients began asking for 30-40% cost optimization without compromising on experience. \u201cWe realized our customers weren\u2019t looking to replace people,” he told Mint. \u201cThey were looking to scale without adding headcount and that\u2019s where AI fit naturally.”<\/p>\n Ozonetel now has one of the largest order books for voice-AI deployments in the country, claims Gupta, with use cases spanning e-commerce, financial services and real estate. He estimates that AI products could contribute nearly 40% of the company\u2019s revenue within the next three years.<\/p>\n Ozonetel now has one of the largest order books for voice-AI deployments in the country, claims Gupta, with use cases spanning e-commerce, financial services and real estate.<\/p>\n Meanwhile, Gupshup-owned Knowlarity, a cloud telephony company founded in 2009, is positioning itself at the intersection of voice and messaging, which are increasingly becoming the two channels that define how Indian users communicate.<\/p>\n \u201cKnowlarity started as a cloud telephony provider, but we have evolved to layer AI-driven agents directly on top of that infrastructure,” says Beerud Sheth, founder and chief executive officer (CEO) of Gupshup, which acquired Knowlarity in 2022.<\/p>\n \u201cOur strength lies in how these layers work together, so telephony and AI aren\u2019t siloed but natively integrated,” Sheth said. \u201cThis makes it easier for enterprises to deploy voice AI without complex integration issues.”<\/p>\n \n\t\t\t\t\tView Full Image<\/span>\n\t\t\t\t<\/p>\n<\/p><\/div> The integration between telephony and AI has delivered a measurable impact for Knowlarity customers, he adds. In one such deployment, with skincare brand Dot & Key, Knowlarity\u2019s voice AI handled nearly 80% of calls without human intervention, reduced the call handling time by half, and cut support costs by about 4%.<\/p>\n To imitate real-life workflows, these platforms are also blending communication. For instance, if a customer misses a voice call, the system can trigger an automatic WhatsApp or SMS follow-up to create a hybrid experience that matches how Indian users prefer to communicate. \u201cInstead of forcing businesses or customers to choose between voice or messaging, we are building seamless hand-offs,” says Sheth. \u201cIt respects customer preferences and drives better engagement.”<\/p>\n Industry observers point out that the rise of cloud telephony companies in the voice AI stack signals the maturity of India\u2019s software as a service (SaaS) ecosystem. Few others, they say, could have built these systems with the same blend of technical depth and regulatory awareness.<\/p>\n \u201cBoth Exotel and Ozonetel figured out an India-centric product for the Indian customer base and regulatory architecture,” says Sharad Sharma, co-founder of software product industry association iSPIRT Foundation. He believes the next wave of disruption will come from AI-first software built in India, and that voice AI will be at the forefront.<\/p>\n Sharma points out that Western voice AI systems will struggle to work \u201cout of the box” in India\u2019s multilingual and small-business-heavy context. \u201cThis is an opportunity that these tech-savvy entrepreneurs can exploit,” he says. \u201cThey can embrace new data rails, like DEPA, to get ahead of their foreign competitors.”<\/p>\n DEPA, or Data Empowerment and Protection Architecture, is a consent-based framework in India for user data, created by Niti Aayog. It allows individuals and businesses to share their data securely and with consent across platforms.<\/p>\n This opportunity is already being seen in how Indian companies are building voice systems tuned to the country\u2019s diversity. It\u2019s a challenge that\u2019s forcing them to innovate beyond Western templates.<\/p>\n Sheth believes India\u2019s linguistic complexity makes it the ideal testing ground for the world\u2019s next generation of conversational AI. \u201cIndia\u2019s diversity and call volumes create a real-world lab for voice AI,” he says. \u201cWe\u2019re building models that can handle code-mixing, regional dialects and local nuances with high accuracy. If it works here, it can work anywhere.”<\/p>\n According to a report by research firm IMARC, the cloud telephony services market was globally valued at around $23.9 billion in 2024, and is projected to reach nearly $47.8 billion by 2033. In India, the market for AI agents, the software layer backing these voice bots, was estimated to be worth $0.28 billion in 2024 and expected to balloon to $3.55 billion by 2030, according to a report by Grand View Research.<\/p>\n
\n\t\t\t<\/strong><\/figcaption><\/figure>\n<\/div>\nBuilding experience centres
<\/h2>\n![\"Shalil](\"https:\/\/www.livemint.com\/lm-img\/img\/2025\/11\/25\/600x338\/shalil_gupta_1764065395538_1764067219580.jpg\" "\\"Shalil")
\n\t\t\t\t <\/source><\/picture>\n
\n\t\t\t<\/figcaption><\/figure>\n<\/div>\n![\"Beerud](\"https:\/\/www.livemint.com\/lm-img\/img\/2025\/11\/25\/600x338\/2-0-548526124-beerudsheth-0_1680594856604_1764067238700.jpg\" "\\"Beerud")
\n\t\t\t\t <\/source><\/picture>\n
\n\t\t\t<\/figcaption><\/figure>\n<\/div>\nA uniquely Indian advantage<\/h2>\n
![\"Giant](\"https:\/\/datawrapper.dwcdn.net\/tHU4s\/full.png\")