As Indian enterprises accelerate AI adoption, Arindam Hari, CEO of Acceleron Solutions, argues that scaling beyond pilots requires clear business outcomes, robust governance, trusted data and deeper integration with core enterprise systems.
Ask a room of CIOs whether their organisation uses AI, and almost every hand goes up. Ask how many can point to a line in the profit and loss statement that moved because of it, and the room goes quiet.
The data tells the same story. McKinsey’s latest global survey found that 88 percent of respondents say their organisations use AI, yet only 7 percent say it has been fully scaled across the organisation. Most enterprises have proved that AI can work. Very few have proved that it pays.
The technology is rarely what holds them back. Models are capable, compute is accessible and the tooling improves every quarter. What separates the organisations that scale AI from those still running pilots comes down to three questions. Is AI tied to measurable business outcomes? Is it governed with the same discipline as any core system? And are the data, infrastructure and processes beneath it built to carry production load? The CIO sits at the centre of all three.
India is moving fast. Discipline has to keep pace.
Indian enterprises have moved faster than many of their global peers. The EY-CII report The AIdea of India: Outlook 2026 found that 47 percent of Indian enterprises have multiple GenAI use cases live, and another 23 percent are at the pilot stage. Only one in ten, however, is scaling those use cases across the business.
That speed is a real strength. Indian organisations have shown an appetite for experimentation that many larger markets lack. Speed without a unifying plan, though, carries its own risk. A company can end up with a dozen live use cases spread across procurement, HR, finance and customer service, each built by a different team on a different stack and measured in a different way. Every one of them works on its own terms. Together, they add up to very little.
In my conversations with technology leaders, this is the pattern I see most often. Whether Indian enterprises will adopt AI is a settled question. The open question is whether the use cases they have already launched will ever amount to enterprise value.
”The real measure of enterprise AI is not how many pilots an organisation launches, but how much measurable business value it delivers when AI becomes part of everyday operations.”
Arindam Hari, CEO, Acceleron Solutions
A pilot is designed to succeed. It runs on curated data, a small and willing group of users, and conditions the team can control. Production offers none of that. Data is incomplete and inconsistent, some systems are decades old, and someone is held accountable when the output is wrong.
Most initiatives falter at exactly this transition, and the failure points are familiar.
The first is placement. AI is often built beside core systems such as ERP, CRM and supply chain platforms rather than inside them. Users have to leave their workflow to use it, and within a few weeks, most stop.
The second is measurement. Without a baseline taken before the pilot began, nobody can say with confidence what improved or by how much.
The third is ownership. When the initiative stays with IT or an innovation team, the business function treats it as someone else’s project, and the way work gets done never changes.
The cost of these mistakes is rising. Gartner predicts that over 40 percent of agentic AI projects will be cancelled by the end of 2027 because of escalating costs, unclear business value or inadequate risk controls. None of those reasons points to a failure of the technology.
A pilot answers one question: can this work? Adoption answers a harder one: will the business change how it works? Most organisations only ever ask the first.
Start with the outcome, not the model
The organisations getting real value from AI select use cases the way they would evaluate any capital investment. They begin with a business problem and the value of solving it, and only then decide whether AI is the right tool. Technical novelty makes for a good demonstration and a poor selection criterion.
Success has to be defined in business terms before anything is built. Cycle time, cost per transaction, forecast accuracy, working capital and customer resolution time are numbers the CFO already tracks, and they are the numbers AI should be expected to move. An initiative that can only report logins, queries answered or active users has not yet demonstrated value.
McKinsey identifies a small group of AI high performers, roughly 6 percent of respondents, who report significant value and attribute more than 5 percent of EBIT to AI. Their advantage lies less in the models they use and more in where they put them. They embed AI in the workflows where decisions already happen, so the impact shows up in operating metrics rather than on a usage dashboard.
The practical rule follows from this. Every AI initiative should have a named business owner and a measurable target. If either is missing, the initiative is not ready for funding.
Governance has a reputation problem. Many business leaders see it as the step that slows everything down. In practice, it is what allows AI to move out of the sandbox and into systems of record.
No enterprise would connect a new financial application to its general ledger without access controls, audit trails and clear approval rules. AI that makes or informs decisions deserves the same treatment. That means clear ownership of data and agreed quality standards, security and access controls designed in from day one, defined rules on when a human must validate an output, and the ability to trace any AI-driven decision back to the data and logic behind it. For Indian enterprises, it also means building in compliance with the Digital Personal Data Protection Act from the outset rather than retrofitting it later.
McKinsey’s research shows that high performers are more likely to have defined processes for deciding how and when model outputs need human validation. The finding deserves attention. The organisations extracting the most value from AI are also the ones that have thought hardest about where it should not be trusted on its own.
Sequence matters here. Organisations that treat governance as a launch requirement scale faster than those that add it after an incident. Once something has gone wrong, every subsequent use case faces the scrutiny the first one escaped, and lost momentum is difficult to recover.
Build the foundation for scale
AI can only scale as far as the systems beneath it allow. Three foundations matter most.
The first is a unified, trusted data layer. Most enterprises hold data in functional silos, with the same customer or product defined differently in finance, sales and operations. A model trained on one version of the truth will contradict a report built on another, and users lose confidence quickly.
The second is infrastructure that can handle production workloads at a predictable cost. Pilots are cheap to run. Inference across thousands of users and millions of transactions is not, and many programmes stall when the first full-scale cloud bill arrives.
The third is integration with core enterprise applications, so that an AI output triggers an action. A purchase order gets raised, a claim gets routed, a price gets adjusted. A recommendation that waits on a dashboard for someone to act on it delivers a fraction of its potential value.
People complete the picture. Roles need to be redesigned around AI-assisted work, teams need reskilling, and change management needs the same rigour as the technology rollout. An enterprise that deploys AI into unchanged roles should expect unchanged results.
Taken together, these foundations lead to one conclusion. Scaling AI depends less on deploying more models and more on modernising the enterprise architecture those models rely on.
The CIO’s agenda for the next 18 months
The CIO’s role in AI is shifting from running experiments to running AI as a business capability. Four priorities should shape that agenda over the next 18 months.
- Prune the pilot portfolio. Keep the use cases that have a business owner and measurable value, and close the rest.
- Set governance before scale. Agree on data ownership, validation rules and auditability before a use case reaches production.
- Fix the data and integration layer. Invest in the plumbing that lets AI outputs flow into core systems and trigger action.
- Measure AI in business outcomes. Report on cost, cycle time and revenue impact, and retire adoption metrics as the primary measure of success.
Access to AI has stopped being a differentiator. Every enterprise can license the same models and buy the same cloud capacity. The advantage over the next few years will belong to the organisations with the discipline to make AI part of how the business runs.