Everyone is watching the models. Sachin Bhatia thinks the real action is one layer below.

In this episode of Front Page, Sudhi Sachdev sits down with Sachin Bhatia, Co-Founder and Chief Growth Officer at Exotel, for a sharp take on where voice AI is actually headed. Sachin observes that the models running voice bots are commoditising faster than the infrastructure that carries them. “Models eventually will be like electricity,” he says. “It doesn’t matter if you’re buying it from company A or company B.”

Exotel powers more than 25 billion interactions a year across 7,000-plus enterprises, including Apollo 24/7, HDFC Securities, and JSW MG Motor. Almost 40% of India’s voice AI conversations already run on Exotel’s rails: your bot, a competitor’s bot – everything travels the same highway.

That highway is the story. Exotel built India’s first agent streaming platform on a full-duplex websocket architecture that delivers audio in frames under 20 milliseconds, with no call-setup lag. Get this layer wrong, Sachin argues, and no model will make the conversation feel human. Get it right, and you can replace the dreaded IVR with something a caller actually thanks you for.

The conversation runs wide:

• India needs frontier-minus-one models, not frontier ones; because the unit economics of a sub-10-lakh agent are nothing like a $50,000 one in the western countries.
• Harmony flips the script: one human supervises many AI bots, not the other way around. The industry’s bot-first, human-as-backup model? “I think that model is fundamentally flawed.”
• NPS is outdated. “When you listen, you don’t need to ask.” AI conversation intelligence reads how customers actually feel, at scale.

It is a rare, grounded look at voice AI from someone who has spent 14 years building the plumbing and has a strong, well-argued opinion about what comes next: from India’s 22-language challenge and sovereign AI to Exotel’s expansion across the globe.

Watch the video for the full conversation.