Comfortably dumb: Why AI support at Swiggy and Zomato frustrates users

Comfortably dumb: Why AI support at Swiggy and Zomato frustrates users

On a recent weekday night in Gurgaon, 29-year-old marketing professional Rhea Malhotra ordered dinner on Zomato after a long workday. When the order arrived, a key item was missing.

What followed wasn’t a quick fix, but a loop. The in-app chatbot prompted her to confirm the issue, then steered her through a series of preset responses, each one acknowledging the problem, but none actually resolving it. There was no clear way to reach a human.

After 15 minutes, she gave up.

“At some point, you just move on,” she said. “It’s not worth the time.”

Customers across platforms are running into the same problem.

According to several consumers, as well as restaurant operators and analysts Moneycontrol spoke with, the shift to AI-led customer support is making it harder to resolve issues on food delivery platforms, particularly when complaints fall outside standard, pre-defined flows.

In many cases, users are able to raise a complaint, but not resolve it. The system moves but the issue does not.

“It’s like you’re trying to explain something specific, but the system only understands fixed options,” said Gurgaon-based consultant Ankit Sharma. “If your issue doesn’t fit neatly into those, you’re stuck.”

A problem that shows up only when things go wrong

The friction doesn’t show up in every order. It only appears when something goes wrong. That is partly why it has not yet translated into a visible drop in demand.

“Although these chatbots are a friction point in consumer experience, the percentage of customers having these problems is on the lower side, around 5–10 percent of orders,” said Satish Meena, founder of e-commerce consultancy Datum Intelligence.

While it may seem low, food delivery companies, including Swiggy and Zomato, do roughly 4.3-4.5 million orders per day. Even a 5–10 percent issue rate would mean roughly 215,000 to 450,000 orders facing problems each day.

Most orders go through without issues. But when they do not, the experience of fixing them is increasingly defined by how far the system can stretch.

In categories like food delivery, a large share of complaints follow predictable patterns. Missing items, wrong orders, damaged packaging, or quality issues tend to fall into clearly defined buckets where platforms already have standard playbooks.

“There are a bunch of predictable complaints where automation works really well,” said Srinivasan Subramani, VP - Growth and AI at Clevertap. “If you receive the wrong item or a damaged product, there is a clear playbook. It’s actually inefficient to route those to a human agent, because it adds time and cost for issues that can be instantly resolved through predefined workflows.”

The problem emerges when complaints fall outside those predefined scenarios. Food delivery issues, for instance, are often situational. A partially delivered order, a delivery gone wrong mid-route, or a quality complaint that depends on context does not always map neatly to fixed options.

“The system is only as good as the scenarios it has been trained for,” Subramani said. “If the customer’s issue falls outside that, it becomes difficult for the system to capture it.”

A system that works, until it doesn’t

For restaurant operators, the shift to AI-led support has created a more layered trade-off.

“Earlier, refund policies were lax, and a lot of that cost came onto restaurants,” said Pranav Rungta, founder of Churchgate Hospitality - a Mumbai-based premium dining operator. “For any frivolous complaint, the customer would escalate and ask for a refund. Making the system more robust benefits the restaurants.”

But that correction has introduced a different problem.

“When there is a genuine problem and you need to find a solution, these chatbots are causing an increase in friction,” Rungta said.

That friction is not always visible in metrics, but it shows up in how complaints move through the system. For some restaurants, it also changes where those complaints land.

“It has become difficult for customers to escalate issues now. If you want to speak to someone, it’s very hard to get through to customer care,” said Arun Adiga, managing partner of Vidyarthi Bhawan. “Even for us, though we have account managers, they are not always available. And smaller restaurants have to rely entirely on the standard complaint process.”

In some cases, that pushes customers back to restaurants, even when the issue sits with the platform or delivery.

“We do get such cases where the customer reaches out to us directly, even though the issue is with the delivery or the platform. At that point, we have to explain that they need to raise it with the aggregator,” Adiga added.

Not yet an industry flashpoint

Despite these shifts on the ground, the issue has not yet escalated into something the industry is formally taking up with platforms.

“At this stage, this is not something the NRAI is taking up with platforms as an industry-level issue. There are several other challenges that are more immediate for restaurants,” said Sagar Daryani, co-founder and CEO of Wow! Momo and president of the National Restaurant Association of India (NRAI).

In recent years, restaurant bodies have been engaged in more fundamental disputes with platforms, including commissions, discounting structures, visibility on apps, access to customer data, and concerns around private labels.

In that context, customer support, while increasingly visible as a friction point, has not yet emerged as a priority area for collective action.

That creates a gap: while platforms optimize for cost and restaurants negotiate for margins and control, the burden of navigating broken support systems falls largely on the customer.

The friction is visible. The complaints are consistent. But for now, it remains a user-side problem, not an industry battle.

Why chatbots are here

The shift towards chatbots is, at its core, a cost decision.

At scale, handling millions of customer queries through human agents becomes expensive. Automating a large share of these interactions reduces that cost while speeding up resolution for standard, repeat queries.

This push towards automation is now being backed by deeper investments in AI systems across platforms.

Zomato, for instance, has been scaling up its support stack through Nugget, a platform it launched in February 2025 after developing it internally over three years. The system now processes over a billion messages and emails, resolves more than 15 million tickets each month, and delivers an average query resolution rate of 85 percent, as per its website.

The company says it has also reduced resolution time by over 40 percent, improved agent efficiency by 40 percent, and cut support costs by more than $10 million.

Swiggy, meanwhile, is extending AI beyond support and into the ordering journey itself. Through integrations with tools such as ChatGPT, Claude, and Google Gemini, users can place orders conversationally, without navigating the app.

Powered by what Swiggy calls the Model Context Protocol (MCP), the system allows users to search, compare, and transact across services including Instamart, which hosts over 40,000 products. Given that the feature was rolled out recently, the company has not yet disclosed usage or scale metrics for these integrations.

Swiggy and Zomato did not respond to queries sent by Moneycontrol.

Not just a food delivery problem

But this shift is not limited to food delivery.

“I think you’re seeing this most in high-frequency categories like e-commerce and food delivery, where there are a lot of repeat queries,” said Srinivasan Subramani. “Even in sectors like airlines or banking, chatbots are now the first layer of interaction, though the level of decision-making may differ.”

Across sectors, the logic remains consistent. Routine queries are handled through automation, while more complex issues are expected to move to human support.

To be sure, these systems are beginning to deliver measurable gains. At Bajaj Finance, for instance, AI systems have analysed over 20 million customer calls, generated around one lakh new offers, and contributed to loan disbursals of about Rs 1,600 crore, all within a single quarter, as Moneycontrol reported earlier.

But cost efficiency is only one part of the design. Platforms are also trying to manage misuse. Making every complaint easy to raise and resolve can lead to abuse, especially in categories where refunds are frequent.

“In some cases, systems are deliberately designed to add a bit of friction,” said Srinivasan Subramani. “If you make it too easy, you risk misuse. So you try to balance genuine complaints with filtering out abuse.”

While the economic logic makes the direction irreversible, the real question is not whether chatbots will replace human support, but whether they can replicate it.

For now, the answer is: not quite.

“What is being called AI today is not really intelligence. It is a series of predefined commands. If a customer says this, the system is trained to respond in a certain way,” said Gagandeep Sapra, founder of Glisco Kitchens, which operates brands such as Tadka Rani and Meal Combo Box Company.

“These systems are designed to acknowledge and respond, but not necessarily to understand. They work when the problem fits into a defined flow, but break when it does not.”

That gap becomes more visible in food delivery, where issues are rarely uniform.

Unlike categories with predictable queries, complaints here are often situational — a missing item in a bundled order, a swapped delivery, or a quality issue that depends on context.

“The expectation is that AI should behave like a human. It should understand nuance, react to different situations. But what is being deployed right now is not there yet,” he said.

That mismatch, between expectation and capability, is where the experience begins to fall apart.

The next lever: pricing access to support

If the current phase is about replacing human support, the next phase may be about how that support is structured, and who gets access to what.

One possibility is that faster or human-led resolution becomes more differentiated over time, potentially tied to subscription programs or higher-value users.

“Over time, you could see this being priced,” said Datum’s Meena. “If you want faster resolution or access to a human, that could become part of a premium offering, while the rest continues to be handled by chatbots.”

Whether or not that plays out, the underlying direction is unlikely to change.

Whatever the strategy platforms adopt, the objective remains the same: to keep costs under control in a business where margins are thin and price sensitivity is high.

“Costs are going up, and there is pressure on consumption,” NRAI’s Sagar Daryani said. “To keep food delivery affordable, these kinds of changes become necessary.”

That leaves the system in a familiar place.

Efficiency is improving. Costs are being managed. The model is holding. But for customers trying to resolve a problem, the experience is still catching up.