AI took over 65% of the conversations: Here is the 35% that stayed human, on purpose

By Akhil Gupta, Founder of ConvoZen.AI and Co-Founder, NoBroker

There is a fear going around in almost every boardroom in India right now. Someone in the meeting says it out loud, or someone thinks it and does not say it. If AI can now answer calls, reply to chats, and solve customer problems on its own, do businesses still need people in customer service at all.

Inside one of India’s largest proptech platforms, that question has already played out in practice, not as a hypothesis. The company deals with an enormous volume of customer interactions every single day, calls about property listings, chats about payments, questions about service bookings. When AI agents were first introduced into that ecosystem, they handled a small fraction of the load. Today, AI manages a large share of it, and within the next couple of years, that share is expected to reach close to 80% of all customer conversations.

By most measures, that should be the whole story. Faster answers, lower costs, fewer people waiting on hold. Businesses across banking, insurance, travel, healthcare, and e-commerce in India are seeing the same shift play out. But what stands out from watching this transition up close is not the enormous 65% that has moved to AI. It is the 35% that has humans retained, deliberately.

What AI is genuinely great at
It is worth being honest about where AI wins, because it wins clearly and completely in a large part of customer service.

Most customer queries are repetitive. Where is my order? When is my EMI due? Can you book this slot? These are questions that have already been answered a thousand times before, in more or less the same way. AI is built for exactly this. It responds instantly, in Hindi, Tamil, Telugu, or whichever language the customer is comfortable in, without making anyone wait or repeat themselves.

This is not a small win. The difference between waiting in a queue for fifteen minutes and getting an answer in under a minute is enormous for a customer. It is exactly why AI adoption has moved as fast as it has inside large Indian consumer platforms, and why the same pattern is now repeating across nearly every industry that deals with high volumes of customer conversations.

The 35% percent nobody should automate away
Here is where most conversations about AI and customer service tend to stop. AI is efficient, so it will simply keep expanding until there is nothing left for people to do. Real deployments at scale tell a more complicated story.

A large part of customer service is not really about information. It is about emotion, trust, and judgement. A customer whose delivery is delayed and who is genuinely angry is not looking for a status update. Someone whose insurance claim just got rejected is not looking for a policy clause read out to them. They want to feel heard first, before anything gets solved.

AI can be trained to sound polite, even empathetic. But there is a difference between sounding empathetic and actually reading between the lines, sensing that someone needs a small exception made even though it is not strictly in the rulebook, or simply staying quiet for a second while someone vents before jumping to a solution. That instinct is still deeply human. Every contact centre has cases that do not fit any script, situations nobody has seen before, and those are exactly the moments where a person needs to think on their feet in a way AI still cannot.

This is the twenty percent that companies leading on AI adoption in India are choosing to protect rather than shrink further, not because the technology cannot get there eventually, but because the cost of getting it wrong in these moments is a customer who feels processed rather than understood.

The job is changing, not disappearing
What is actually happening across these contact centres is not job loss. It is a job change, and that distinction matters more than most people give it credit for.

A customer service agent’s day used to be filled with the same five questions asked on repeat, all day, every day, work that is exhausting and rarely lets anyone grow. As AI takes over that repetitive load, the human role is shifting toward the conversations that actually need a person: the angry customer, the confusing edge case, the negotiation that needs tact. Agents are becoming less like switchboard operators and more like problem solvers and relationship managers.

Businesses that understand this end up with both happier customers and happier employees. Few people enjoy repeating the same answer three hundred times a day. Given the choice, most would rather spend their time solving problems that genuinely need a human brain and empathy.

The accent problem nobody talks about
There is one more piece of this that does not get talked about enough. A lot of AI used in Indian contact centres today was originally trained on polished, accent-neutral, mostly Western speech data. That is a problem, because the way people actually speak on a call centre floor in Patna, or in a WhatsApp chat from a small town, sounds nothing like that.

Indians switch languages mid sentence. Accents vary hugely from one state to another. A customer might start in English, slip into Hindi halfway through, and drop in a regional word without even noticing.

Building AI that can genuinely follow this kind of conversation, not just handle a clean, single language script, is not a small technical detail. It is the difference between AI that feels helpful and AI that constantly frustrates the very people it is meant to serve. This is part of why the case for India building its own AI, trained on Indian voices in all their variety, has grown so much stronger.

Where this actually leaves us
AI will keep growing its share of customer conversations in India, and that is a good thing on the whole. It means faster answers, lower costs, and fewer people stuck waiting on hold. But the idea that this replaces human involvement entirely misses what is actually happening on the ground. The real shift is AI and humans working side by side, each doing the part they are genuinely better at.

AI will keep growing its share of customer conversations in India, and that is a good thing. But the goal was never to remove people from the equation. It was to free them up for the conversations that actually need a human. That is the real story here, not the 65%, but the 35% companies chose to protect.
Ultimately, conversational AI isn’t meant to eliminate human connection, but rather designed to handle the repetitive heavy lifting so teams can focus their energy where human judgment and empathy matter most.