70% of Indian firms have more AI pilots than they can scale, and 31% have already lost money on AI

Indian enterprises have mastered the art of starting AI projects. Finishing them is proving harder. According to Kyndryl’s third annual Readiness Report, released in Bengaluru on 10 November, 70% of Indian organisations say they have more AI pilots than they can realistically scale, and 58% say innovation often stalls after proof of concept.

The cost of that gap is starting to show. Half of the Indian organisations surveyed have significantly changed their AI strategy in the past year, and 31% report having experienced an AI investment failure.

The report draws on the views of 3,700 senior business leaders across 21 countries, including 200 in India. Its headline finding is a widening divide between the returns organisations are seeing and their readiness for what comes with them. Some 60% of Indian leaders investing in AI and machine learning report positive returns, up three percentage points from 2025. Over the same period, average readiness across external risks fell eight points to 27%.

“As AI creates significant opportunity, organisations are now focusing on how to make early returns sustainable,” said Martin Schroeter, Chairman and CEO of Kyndryl. “This year’s Readiness Report shows a new kind of readiness gap: AI ambitions are moving faster than the systems needed to support them, and the bar for readiness keeps moving as AI becomes more integrated in enterprises.”

Out of the lab, into a bottleneck

Progress is real. Some 85% of Indian organisations are now deploying AI in workflows beyond pilots, and the share still at the experimentation stage has fallen 15 percentage points to 56%. In practice, 80% say they are using AI agents to automate workflows that once required coordination across multiple teams or systems.

But moving beyond experimentation is not the same as scaling, and the data suggests many companies are stuck between the two. When seven in ten organisations are running more pilots than they can take to production, the constraint is no longer ideas or enthusiasm. It is the capacity to industrialise them. Each pilot that stalls after proof of concept consumes budget, talent and executive attention without producing a return, and it leaves behind a portfolio that is expensive to maintain and hard to prioritise.

That helps explain the strategic churn in the numbers. A 50% share of organisations rewriting their AI strategy within a year points to a market still learning what works. The 31% reporting an investment failure shows that learning is not free.

The boardroom disconnect

Part of the problem appears to sit inside the organisation. The report finds that 37% of Indian respondents see disagreement between the CHRO and the CTO or CIO on AI strategy. The finding matters because AI at scale is as much a workforce question as a technology one. Decisions about reskilling, role redesign and how much autonomy to give AI agents cut across both functions. When those leaders are not aligned, pilots can succeed technically and still fail to change how the business operates.

For CIOs, the implication is that governance of AI cannot sit inside the technology function alone. Programmes that depend on people changing how they work need a shared view of priorities from the people responsible for both the systems and the workforce.

Modern is not the same as AI-ready

Infrastructure is the other brake on scale. For the third consecutive year, Indian leaders named upgrading IT infrastructure as the top action they are taking to mitigate external risks, and 71% rate their infrastructure as mostly or fully modernised. Yet 59% say significant re-architecture is still required to support AI at scale.

The report draws a distinction between a modern IT estate and one ready for AI. An environment that has been migrated to the cloud and is well maintained can serve conventional workloads capably, yet still struggle with the data access, integration and governance that AI running across business processes demands. For organisations that have spent years on modernisation, the finding suggests the target has moved. It also offers a plausible explanation for why so many pilots do not scale: they work in a controlled setting but meet resistance when they touch the wider architecture.

Security makes scaling riskier

Security concerns topped the list of external risks for the third year in a row and contributed most to the fall in readiness. Some 68% of Indian leaders identify cyberattacks as a top external concern, yet only 35% say they are completely ready to manage them, down 10 points from last year. Globally, the figure is 26%, down 11 points.

The report also highlights a tension. Three in four leaders say AI helps their organisations identify threats faster, even as confidence in their ability to respond has declined. Better detection is not translating into better response. For organisations adding AI agents that act across systems, the gap matters more. Every workflow handed to automation widens the surface that security teams must govern.

A related pressure comes from outside the enterprise. Among global leaders, readiness for competitive disruption from outside their own industry fell 10 points to 23%, as AI agents and other technologies erode established boundaries. Retailers are weighing how to use AI shopping assistants without giving up direct customer relationships. Software companies face questions about SaaS models as AI shifts the build-versus-buy balance. Financial institutions are being challenged by emerging lending models.

Schroeter’s prescription is to focus on people, technology and governance. For CIOs facing a crowded pilot portfolio, the practical reading is to scale fewer things, more deliberately. That means choosing which pilots deserve production investment, fixing the architecture they depend on, aligning with HR leadership on the workforce implications, and matching the autonomy granted to AI systems with the security posture to oversee them.

The bar for readiness, as Schroeter noted, keeps moving. For the 70% with more pilots than they can scale, the first test may be deciding which of them to stop.