AI chatbots are taking over customer support—human help now comes at a premium

Flipkart’s Black, Swiggy’s One BLCK and Zomato’s VIP Mode–either requiring subscriptions or additional payments–promise priority support and direct access to human agents. For other users, chatbots are the only way to raise concerns, and users say getting through to consumer service executives is becoming increasingly difficult.

“This happens because people assume that AI’s marginal costs (e.g., per-query inference in GenAI) make it economical for volume, but human agents excel at empathy and complexity,” said Abhivardhan, president, Indian Society of Artificial Intelligence and Law, an industry forum. “Companies try to capture “willingness to pay” for humans, turning support from a cost center into revenue.”

Large language models (LLMs) have transformed chatbots into conversational agents that can provide human-like responses. As online retailers and quick commerce companies focus on profitability, AI agents cut down on the costs of staffing call centres for customer support. They allow these platforms to even process surges in festive orders without ballooning headcount.

A human premium

Flipkart markets its Black membership as offering “personal attention over automated responses,” with direct access to dedicated agents, immediate callbacks and 24×7 priority support, bypassing chatbot queues. Swiggy One BLCK, an invite-only premium tier, lists “priority support from top customer care agents” as a marquee feature. Members get faster, personalized resolutions compared to standard users.

Zomato has tested a pay-per-order VIP Mode, charging ₹50 for priority riders and concierge-style assistance. Its terms of service explicitly reserve the right to charge subscription or membership fees for access to customer support, signalling where the model could head.

Zepto hasn’t marketed human escalation as a perk yet, but is doubling down on automation.

“We’re heavily investing in agentic AI models that will create unprecedented leverage across multiple business functions,” Nikhil Mittal, chief technology officer at Zepto, told Mint in an emailed response. “These intelligent systems should be able to understand context, apply policies, take decisions and appropriate actions–just like a human would, but faster, more accurately, and at scale.”

Swiggy’s FY25 annual report positions “AI-led services” as a growth driver. It points to Pyng, a standalone AI-driven platform that connects users with verified professionals across 100+ specializations, described as “powered by personalised AI assistants”.

Queries emailed to Zomato, Swiggy, Flipkart, and Amazon did not elicit a response until press time.

Customer trust shaken

Yet, customer care’s AI pivot hasn’t been smooth. Last year, Zomato was forced to ban AI-generated food photos after users complained they were misleading, fuelling refund claims and eroding trust. The platform rolled back the feature.

Many customers are struggling to raise complaints and get resolutions in time.

“Some months back, I used to get a callback from a human agent within minutes if something was missing,” said Rohan Mehta, 25, a consultant in Delhi. “Now I’m stuck nearly half an hour in chatbot loops, and most of the time it’s just the bot repeating itself.”

Priya Sharma, 23, a postgraduate student in Delhi, said, “With food delivery, by the time I reach a human, the order is already ruined. It feels like they’ve made human support deliberately hard to reach.”

According to Ananya Rathore, 26, a marketing executive, even a simple return gets stuck in endless bot chats. “Earlier, I could explain the issue to a person in minutes; now it takes days just to escalate.”

These frustrations signal deeper cracks in the service model as users find it difficult to reach human agents even for genuine concerns.

“Without a human who understands the deeper issue, the system often seems designed to block resolution rather than help. That’s where it starts to look like a dark pattern—whether by design or as a byproduct of AI-based systems,” said Farheen, an analyst at the Centre for Critical and Emerging Technologies, Advanced Study Institute of Asia.

“It’s becoming increasingly difficult to have direct interaction for customer service. Zepto uses AI chat instead of a customer service number. Before you can speak with a human, Amazon forces you to go through bot-first flows. It is extremely high-friction, and it seems like escalation to an agent is purposefully hidden,” she added.

What is being monetised “is scarcity, as we are now moving towards a paradigm that views attention and trust as luxury items”.

AI evangelists

AI startups argue that automation isn’t all bad—it clears routine queries instantly and frees up humans for complex cases.

“If a customer explicitly asks for a human, or if a conversation drags on beyond a certain number of turns, we trigger an escalation. The cost of running AI agents rises with every turn, so beyond a point it’s more efficient to hand over to a human who can resolve it in seconds,” said Abhishek Upperwal, founder of Soket AI Labs, one of the startups chosen under the IndiaAI Mission.

Human support will eventually be reserved for premium users because the cost of running AI agents is coming down drastically, and 70–80% of customer support queries are repetitive, according to Upperwal. AI will naturally handle those, while humans will be pulled in only for complex or paid tiers, he said.

Running human support is costly, and the way to balance the books may be to charge for it—a model already taking shape, Upperwal said.

“Paywalling human agents sits in a regulatory grey zone: firms must ensure timely, effective redress, but they retain latitude on whether the first touchpoint is a bot or a person and whether premium tiers get faster human access,” said Sohom Banerjee.

Ankush Sabharwal, founder and CEO of CoRover.ai, which has raised around $4.9 million in funding, with HDFC Bank as its most recent investor, said, “If the system senses frustration, it can automatically route the query to a live agent.”

He cited the example of markets like the US, where it is difficult to reach a human, even with so-called premium support. “It can take days to connect, and you might only get to speak to a live agent once or twice. It’s a real struggle.”

“As India grows, talking to humans should also not be that easy. A service provider is not there just to offer emotional support or be scolded,” he said. “Humans still have a role in sales, but for routine service, AI can handle it just as well.”

Regulatory grey zone

According to Sohom Banerjee, senior research associate at CUTS International, a policy research and advocacy group, “Paywalling human agents sits in a regulatory grey zone: firms must ensure timely, effective redress, but they retain latitude on whether the first touchpoint is a bot or a person and whether premium tiers get faster human access.”

Yet, experts warn that chatbots remain imperfect. “Hallucination is a ‘by design’ problem of LLMs and cannot be eliminated in entirety, so it is advisable to engage in only readily simple tasks for triage,” said Abhivardhan of the Indian Society of Artificial Intelligence and Law.

Companies worldwide are leaning on AI to plug the gap—Uber now uses generative models to field whole categories of queries, while back home Meesho is testing voice bots it claims score higher than human agents.

“The business logic is compelling: when festive order volumes triple, no company can afford to triple its human support headcount,” said Sohom Banerjee, senior research associate at CUTS International, a policy research and advocacy group. Yet, automation may solve the problem of volume, but it does not necessarily solve the problem of trust, he said.