Silent hijack: Meet the post-app middleman terrifying e-commerce giants

When I explored another option, it adjusted again, explaining the trade-offs between the two products. It was nearly like having a conversation with a salesperson in a brick-and-mortar store, one who was paying attention to a prospective buyer.

When I finally made up my mind, the bot redirected me to a payment page operated by Razorpay to complete the purchase.

ChatGPT, operated by Open AI, allows users to connect various applications—apart from Innovist, there are other Indian companies within the chatbot including MakeMyTrip (travel), Zomato (food delivery), Cars24 (pre-owned cars) and Square Yards (real estate). Many more direct-to-consumer (D2C) brands could follow.

Harshil Mathur, Razorpay’s co-founder, expects the top 1,000 D2C brands to begin integrating AI-led commerce journeys over the next six-eight months. Currently, the company has around 15–20 active pilots where some form of agentic commerce has been rolled out.

Until a couple of years ago, India’s e-commerce wars were fought over apps. Investors spent billions trying to get users to open Flipkart, Swiggy or Zomato first because whoever owned the first click often owned the customer. Now, some of India’s largest commerce and payments companies are preparing for a future where consumers may no longer open apps at all.

This has raised deep concerns globally and in India, particularly around the role of marketplaces. In a world where AI becomes the storefront, who would own the customer and their data? And are large horizontal platforms, like Amazon or Flipkart, staring at a disruption to their business models? Let’s dive in.

Context is king

For years, e-commerce firms were obsessed over search engine optimization (SEO) to rank higher on Google. But with AI altering user behaviour, search is no longer just keyword-led. It is becoming contextual and conversational. Behind the scenes, marketplaces are therefore rushing to adapt to a new race: AI optimization.

Recently, listed internet-first firms Meesho and Swiggy have both introduced conversational AI-led shopping and ordering experiences aimed at simplifying transactions for users. Meesho launched Vaani, a voice shopping assistant in March. The same month, Swiggy partnered with Sarvam, an Indian AI platform, to enable multilingual voice-led ordering across food delivery, grocery and dining services.

“We publicly publish catalogue feeds and work with LLMs (large language models) on attribution, measurements, conversions, and improve the quality of information we provide them so that our products rank better and conversions improve,” Balaji Thiagarajan, chief technology and product officer at Flipkart, told me.

Earlier, a consumer searching for ‘black raincoat’ would typically see results matched to keywords alone. “But conversational AI systems increasingly interpret the broader context before recommending products,” Thiagarajan said. In the above example, the system would factor in signals such as weather conditions, travel intent and usage behaviour while surfacing products. Similarly, a consumer searching for “red Nike shoes” would typically see results matched to keywords alone. But in the age of AI chatbots, the searches typically look like “red shoes for a hike on my Goa trip.”

High-frequency, everyday categories such as food delivery, groceries and quick commerce are where conversational commerce could scale first. “The second category is high-assistance commerce where customers naturally seek guidance—travel is one example; large consumer purchases are another,” said Razorpay’s Mathur.

According to the company’s early data, while ticket sizes are lower and more frequency-led, repeat behaviour is beginning to show up. “Early signals suggest that users who try these experiences are engaging with them repeatedly, for specific use cases because the journeys are faster and more contextual than traditional flows,” Mathur said.

New battles

On AI optimization, the key piece for marketplaces is to get AI chatbots to recommend their products first. The earliest signs of disruption are already beginning to show up in search visibility battles.

In December last year, IndiaMART sued OpenAI in the Calcutta High Court alleging that ChatGPT was selectively excluding IndiaMART listings from AI-generated responses while continuing to surface rival platforms. IndiaMART argued that the alleged exclusion was causing commercial and reputational harm by weakening its visibility in AI-led discovery, Mint had earlier reported.

Nonetheless, marketplaces could themselves feel the heat of disruption—from smaller brands even. Here’s why: India’s digital commerce economy is built on app-based discovery. Marketplaces typically capture ₹15–25 or even more on a ₹100 order—they charge commissions, advertising revenue and sponsored placements because discovery and checkout happen inside their ecosystems.

If AI assistants become the primary interface for shopping, comparing merchants and triggering payments, discovery could shift away from marketplace apps to AI agents, impacting their revenue stream in the process.

“This is probably the most critical discussion happening in the industry today. It is still evolving and everyone is learning and adjusting,” said Thiagarajan. “We want products to be discovered through those channels, but the transaction should happen on Flipkart. That is our current view of how a collaborative AI-commerce ecosystem should work.”

Anuvrat Jain, a partner at Lightspeed Venture Partners, the venture capital firm that has backed companies including Razorpay, Sarvam and Zepto, believes that marketplaces, whose biggest strength was discovery, could face greater pressure than platforms built around fulfilment, logistics and supply chains.

“Conversational AI could weaken marketplaces whose moat was primarily discovery, while strengthening those whose moat is fulfilment and operational scale,” he told me.

However, Jain cautioned that a new intermediary layer may emerge altogether if AI interfaces increasingly control shopping journeys. “When AI agents begin handling discovery, they will eventually demand a share of the economics as well,” Jain said. “That creates the possibility of a new intermediary sitting above the marketplace layer—one that the industry has not fully priced in yet.”

Inside global retail and commerce companies, the conversation has already shifted toward these harsher realities. A recent Bain & Co. report echoed Jain’s perspective, describing what it called the ‘agentic economy’—a world where consumers increasingly discover, compare and even purchase products through AI conversations rather than through websites or apps. The report, based largely on US consumer behaviour and global technology trends, warned that marketplaces and consumer internet platforms may no longer fully control the front door to commerce.

“Consumers are finding what they need in summaries and other AI-powered sources, and they are decreasingly likely to click through to brand websites,” Bain wrote in the report, published in March this year.

According to the company’s research, up to 45% of shoppers already use generative AI tools to compare products while searching online, while a “significant chunk” complete purchases with agentic assistance.

Another report, by McKinsey & Co., described agentic commerce as a “seismic shift” that could fundamentally remake shopping itself. By 2030, McKinsey estimates agentic commerce could influence $3–5 trillion in global commerce value, with the US B2C retail market alone potentially accounting for nearly $1 trillion.

This report also carried the same stark warning for marketplaces. It argues that AI agents could increasingly become the interface between consumers and merchants, handling product discovery, price comparison, checkout, loyalty rewards and even post-purchase returns.

So, how far are we from that reality?

“Right now, in India, most companies are building standalone AI launches or assistants. In the next phase, these agents would start interacting with each other directly. That stage has not fully arrived yet, but it is coming fast,” Kapil Makhija, chief executive officer of Unicommerce, a commerce SaaS (software-as-a-service) firm, told me.

When agents begin interacting with each other, AI can independently shop for users without a need for them to visit marketplaces.

When I type the same query for a hair product, an AI assistant that knows my spending preferences could automatically search across multiple merchants, compare prices, apply available coupons, select the most suitable product, and complete the transaction by interacting directly with merchant and payment agents working for them in their back end.

Instead of opening a marketplace app, I might receive a notification saying my preferred shampoo has been ordered and will arrive tomorrow. An agent that understands my hair type, buying patterns and consumption cycle could eventually go a step further—reordering the product before I even realize I am running low.

Convenient, perhaps. Slightly unsettling like a ‘Black Mirror’ plot.

The data question

For years, internet marketplaces operated on a relatively stable formula: bring users into the app, keep them engaged, learn their behaviour and monetize attention through commissions, ads and repeat purchases.

Companies such as Zomato, Swiggy and Flipkart today generate growing portions of revenue from advertising and sponsored visibility inside their apps. Merchants pay to rank higher in searches, appear first in recommendations or stay visible during peak demand windows.

If users begin their shopping journeys inside an LLM interface rather than a marketplace app, the intent data—what they browsed, what they compared, what they almost bought but finally discarded—might start accumulating somewhere else.

So, the bigger question is: who owns the customer and their data?

Thiagarajan said the data question is one the industry has not yet resolved.

In traditional search, Google knows only that a user clicked a link and arrived at Flipkart—no behavioural data beyond that is shared. “We have commitments around data sharing. We tell customers we will not share their data outside the Flipkart ecosystem without consent. But in a deeply integrated AI-native shopping experience, the industry has not solved this yet. That is an ongoing conversation,” he said.

“Right now, the AI companies want engagement, traffic and market share. Monetization usually comes later. Whether the economics fundamentally changes or not is still unclear,” he added.

Thiagarajan, nonetheless, is not ruling out a harder pivot. “Amazon does not share catalogue information with Google. That is a deliberate strategic decision they have already made. We are not there yet,” he said.

Difficult to replace?

As of now, marketplaces believe they remain structurally difficult to replace. Behind every order sits a sprawling operational machine—warehouses, sellers, inventory systems, delivery fleets, returns management and customer support. That infrastructure still belongs largely to the marketplaces. The discovery moat may weaken, but the logistics and fulfilment pieces still remain.

“Unless AI companies decide to become full e-commerce companies themselves, I see them staying at the top of the funnel enabling discovery and shopping initiation, not full commerce operations,” Thiagarajan said.

There are technical hiccups to sort out, too. Razorpay’s chief product officer Khilan Haria told me that the early pilots work better on certain configurations like with a desktop app or a paid subscription of ChatGPT. But that is only a minor and temporary bottleneck. The enduring problem, and one that would keep marketplaces up at night, is the data question, and the nagging thought of a new intermediary that could control discovery.

In the US, according to a Forbes article published in November, ChatGPT is already making a dent on how shoppers interact with marketplaces. Etsy gets more than 20% of its referral traffic from ChatGPT while Target gets about 15%, the report stated.

While Indian marketplaces are currently focusing on the immediate problem of avoiding invisibility within conversational commerce, the larger problem of saving margins is arriving soon.