India's Multi-Crore AI Illusion: Are Brands Paying for Technology That Doesn't Exist?

Documented cases of overstated AI capabilities and fresh marketing data raise a harder question for Indian brands: what evidence should come before the purchase?

At the American restaurant drive-through, the promise was automation. Behind the technology, however, were people entering food orders, including workers in India and the Philippines.

That gap sits at the centre of a January 2025 enforcement action against Presto Automation. According to the US Securities and Exchange Commission, the original version of Presto’s proprietary voice technology, commercially deployed in September 2022, required human intervention for every order. A more advanced version piloted from June 2023 still needed people to enter orders approximately 70% of the time through at least December that year. 

The SEC found that Presto had misleadingly claimed its technology eliminated human order-taking. The company settled without admitting or denying the findings. No civil penalty was imposed, reflecting its cooperation and remedial efforts. 

The case was American. The question it raises belongs in every Indian boardroom approving an AI purchase: what exactly is operating behind the interface?

For brands buying customer service tools, personalisation platforms and marketing automation, the danger is paying for a capability that has been exaggerated, poorly defined or never demonstrated. A working screen can conceal a much less impressive system.

The premium attached to a promise

India’s appetite is real. An EY-CII study published in November 2025, covering 200 organisations, found that 47% had multiple generative AI applications in production, while 10% were scaling them across the business. Speed of deployment influenced buy-versus-build decisions for 91% of respondents. 

That urgency creates a difficult purchasing environment. The buyer wants something operational quickly. The seller has an incentive to make implementation sound simple. The expensive questions concern data, integration, supervision and the work that remains after the demonstration ends.

The same tension appears in Gartner’s 2026 CMO Spend Survey. Its 401 respondents, drawn from North America, the UK and Europe, allocated an average 15.3% of marketing budgets to AI initiatives. Only 30% reported mature or fully developed AI readiness. These are overseas findings, rather than an estimate of Indian marketing expenditure.

Gartner’s Ewan McIntyre, VP Analyst and Chief of Research for marketing, warned of investment outrunning the foundations needed to support it: “AI maturity is beginning to separate marketing leaders from laggards.” 

BCG’s separate 2026 survey of 300 global CMOs makes the gap more visible. While 96% reported significant transformation across marketing through AI, 42% were still using generative AI primarily to assist people with particular tasks in limited workflows. BCG classified 32% as leaders deploying agents across redesigned workflows.

The money behind these declarations is substantial. In the same BCG marketing survey, 43% of CMOs said their AI investment exceeded $15 million that year. Those are global respondents, but the figure captures the financial stakes when a loosely defined transformation programme becomes a major purchasing commitment. 

These studies do not calculate Indian brands’ losses from fake AI. Nor does limited deployment establish wrongdoing. They show how easily the language of transformation can run ahead of operational reality.

The distinction matters because disappointment has several possible explanations. A genuine model can be deployed against unsuitable data. An effective tool can sit inside a process that nobody redesigns. A supplier can also promise functionality it cannot deliver. The financial result may look similar, but the evidence needed to diagnose each problem is different.

Consider the difference between generating an advertisement and improving advertising performance. A tool might produce copy instantly. That says little about whether the copy increases profitable sales, survives legal review or performs better than the brand’s existing process.

A similar distinction applies to customer service. A bot can acknowledge a refund request without having permission to issue the refund. A system that answers questions and one that resolves the underlying problem offer different commercial value.

The demonstration is where those distinctions can disappear. A polished answer to a familiar question shows that the system worked on that occasion. The harder test comes with an unfamiliar query, incomplete purchase history or a request that requires action in another system. That is where the distance between a presentation and a product becomes measurable.

For Indian brands, evaluation also has to reflect the actual customer: language, accent, incomplete records and the systems already in use. Performance on a carefully prepared English demonstration cannot establish readiness for every market a national brand serves.

The newest vocabulary can make those differences harder to detect. In June 2025, Gartner described vendors relabelling assistants, chatbots and robotic process automation as agentic AI without substantial agentic capabilities. It estimated that only about 130 of the thousands of vendors making such claims were genuine.

“Many use cases positioned as agentic today don’t require agentic implementations,” Anushree Verma, Senior Director Analyst at Gartner, said in that announcement.

Gartner also forecast that more than 40% of agentic AI projects would be cancelled by the end of 2027 because of costs, unclear value or inadequate risk controls. That remains a forecast, not a count of failures already recorded.

When the claim becomes the product

The clearest cases of AI washing involve a specific promise that the underlying technology cannot support.

In March 2024, the SEC announced settlements with investment advisers Delphia and Global Predictions over misleading AI claims. Delphia had claimed capabilities involving client data that it did not possess. Global Predictions had falsely presented itself as the first regulated AI financial adviser.

The firms agreed to combined civil penalties of $400,000, without admitting or denying the findings.

“Investment advisers should not mislead the public by saying they are using an AI model when they are not,” Gary Gensler, then SEC Chair, said when the actions were announced. 

Those findings concern particular companies and historical conduct. They establish that AI washing can involve demonstrably false representations. They do not establish that every ambitious product claim, outsourced task or disappointing deployment is fraudulent.

Human involvement itself is an especially poor test. People may review model outputs, approve sensitive transactions and handle unusual cases. Those controls can make a product more useful. The issue is whether customers understand the arrangement and whether the commercial promise matches it.

The Presto order illustrates why definitions matter. Its reported non-intervention figures concerned orders completed without restaurant staff stepping in. They did not necessarily exclude assistance from workers elsewhere. Moving labour away from the counter could therefore appear to an investor like removing labour altogether. 

For an Indian brand evaluating a support platform, that difference changes the economics. A claimed reduction in human handling should include the vendor’s people as well as the brand’s own staff. Otherwise, the business may be comparing two differently packaged labour models.

Ownership claims deserve equally close examination. Building a useful service on another company’s model is legitimate. The buyer still needs to understand what the supplier owns, what it licenses, and which capabilities depend on an outside provider. These distinctions affect pricing, continuity and accountability.

A brand’s reputation can also become entangled with its supplier’s claims. If the brand promises customers immediate automated resolution, an undisclosed dependency on manual processing becomes its customer experience problem. The customer encounters the brand’s promise, while the limitations of the technology supplier remain out of sight.

The evidence behind the invoice

There are concrete Indian examples of how to describe an AI deployment more clearly. In his 2025 AGM address, Hindustan Unilever Chairman Nitin Paranjpe described AI-powered personalised advertisements within the company’s Shikhar retailer platform, alongside its ordering and customer relationship tools. 

That identifies a function, an intended user and a place within an existing business process. It gives an evaluator something specific to examine, although the company’s description alone does not independently establish its financial return.

Speaking at the ISA CEO conference in February 2025, Rohit Jawa, then HUL’s CEO and Managing Director, stressed the foundations required. “So, a clear strategy and a robust data foundation are absolutely essential for a successful AI implementation,” he told exchange4media.

The wider evidence also challenges a blanket verdict that AI is an illusion. BCG’s September 2026 Applied AI Index, based on more than 1,300 executives and senior leaders, placed 7.5% of companies in its most mature category and another 41% in a scaling group already creating some value.

For buyers, the useful dividing line is evidence. A proposed personalisation engine should face a controlled comparison with the existing approach. A service agent should be assessed on completed resolutions, including escalations and repeat contacts. Creative automation should be measured after editing and approval costs.

Mahesh Makhija, EY India’s Technology Consulting Leader, put the organisational challenge plainly in the November 2025 report announcement: “For enterprises, the focus must now move from building pilots to designing processes where humans and AI agents collaborate seamlessly.” 

The purchase decision should follow the same discipline. Contracts can specify the promised function, acceptable error rates, human involvement and the evidence required before expansion. A pilot becomes useful when the buyer can test the claim and decline the next cheque.

At the drive-through, the hidden detail was who entered the order. In a marketing department, it may be who makes the decision, what the software actually does, and whether it improves the result. The invoice is real. The capability should be equally easy to establish.

Disclaimer: All data points and statistics are attributed to published research studies and verified market research. All quotes are either sourced directly or attributed to public statements.