Two research reports published five months apart, both about Indian companies and artificial intelligence, tell stories that cannot both be comfortable at the same time.
In March 2026, Deloitte's State of AI in the Enterprise found that Indian organisations lead their global peers in at-scale AI deployment. Forty per cent of Indian respondents reported significant or full usage of AI, against a global average of roughly 28 per cent. At-scale deployment ran strongest in product development (62 per cent), strategy and operations (56 per cent), marketing and sales (55 per cent) and supply chain (48 per cent).
In August 2026, ET Edge's CIO&Leader published State of AI in Indian Enterprises 2026, drawn from 300-plus senior technology leaders surveyed in May and June. It found that 60 per cent of organisations remain at pilot or exploration stage after two or more years of investment, and that only 12 per cent can point to significant, measurable return.
India is simultaneously the country that deploys AI faster than anyone else and the country that cannot demonstrate it is working. Reading why those two findings sit together is more useful than picking a side.
Two Reports, Two Indias
The contradiction is less severe than it first appears, because the two studies measure different things. Deloitte measured deployment — how much AI is running in production across business functions. ET Edge measured demonstrated return — whether anyone can prove the deployment paid for itself. A firm can score well on the first and badly on the second, and a great many Indian firms evidently do.
| Measure | Finding | Source |
|---|---|---|
| Significant or full AI usage | 40% India vs ~28% global | Deloitte, Mar 2026 |
| Still at pilot or exploration stage | 60% | ET Edge, Aug 2026 |
| Significant, measurable ROI | 12% | ET Edge, Aug 2026 |
| No measurable ROI, or cannot tell | 57% | ET Edge, Aug 2026 |
| Name productivity as the primary motive | 83% | ET Edge, Aug 2026 |
| Allocate under 10% of IT budget to AI | 58% | ET Edge, Aug 2026 |
Deloitte's own framing anticipated the problem. Its India commentary noted that the pace of adoption is running ahead of depth in capability building, and that the dominant playbook is incremental change targeted at measurable operational gains rather than structural reinvention. Read alongside the August findings, that reads less like a caveat and more like a forecast that came true within one quarter.
A Measurement Failure, Not a Performance Failure
The single most important line in the ET Edge report is its characterisation of the 57 per cent who have no measurable return or cannot determine whether one exists. The report calls this a measurement failure rather than a performance failure.
That distinction matters enormously, and it is the part most commentary skips. It does not claim the AI is not working. It claims that a majority of Indian organisations built no mechanism capable of detecting whether it worked. Those are different diseases with different cures. If the tool were simply bad, the answer would be to change tools. If the instrumentation is absent, changing tools changes nothing — the next deployment will be equally invisible.
"The spending arrived. The returns have not. That is not a story about a market failing for lack of vision — it is a lack of process discipline, in how pilots are selected, who owns the business case, and how success is measured." — R. Giridhar, Editorial Director, Technology, ET Edge
The practical implication is uncomfortable for anyone who has already spent money. If you did not record what the process cost before you automated it, you have permanently lost the ability to prove what the automation saved. There is no retrospective fix. The baseline had to be captured first, and for most Indian firms it was not.
The Budget That Was Never Properly Drawn
The funding pattern in the ET Edge data explains a good deal of the measurement gap. Eighty-three per cent of technology leaders named productivity improvement as the primary reason for investing in AI. Yet 58 per cent allocate less than 10 per cent of the IT budget to it, and 55 per cent either experienced AI cost overruns in the past year or do not track AI spending separately from the broader IT budget at all.
Consider what that last figure means in practice. If AI spend is not a separate line, then AI return cannot be a separate line either. The investment disappears into general IT, the savings disappear into general operations, and the board is handed a question it has made structurally unanswerable. The report's argument to boards is precisely this: the combination of an unbudgeted spend and an unmeasured return makes it difficult to hold management accountable for either the investment case or the returns it is meant to deliver.
This is a governance failure disguised as a technology story, and it is worth noticing that it costs nothing to fix. A separate cost centre and a recorded pre-automation baseline are administrative decisions, not capital ones.
What MSME Owners Actually Said When Asked
Enterprise surveys capture the view from the CIO's office. PwC India's Unlocking the AI Edge for MSMEs, published in March 2026, did something more useful: it asked smaller manufacturers directly, and printed what they said.
The owner of a machine-component manufacturing unit, asked about his data readiness, answered: "I don't know if my enterprise is ready to benefit from AI." Another promoter, describing vendors selling digital transformation, said: "Show me the real value AI will deliver and the time frame within which I can realise it." A third, having lived through half-implemented ERP systems and unused software, asked: "Who will bail me out if I am stuck with a sub-optimal solution due to the fault of the tech provider?"
The sharpest of them is the shortest. "Everyone is trying to sell; no one is willing to stand with me."
PwC's reading is that unlike large firms, MSMEs typically lack a supportive ecosystem that can help them make informed choices, validate technology decisions, and intervene when course correction is required. The absence of those guardrails, the report argues, leaves smaller firms feeling unsupported through the whole adoption journey. A worker interviewed for the same study framed the anxiety differently and more plainly: "Will the machine still need me once it becomes smarter?"
These are not the questions of people resisting technology. They are the questions of people who have been sold technology before and are asking, reasonably, for evidence and accountability this time.
Why the Enterprise Numbers Understate the Smaller Firm's Problem
If 60 per cent of large Indian enterprises are stuck at pilot stage, smaller firms face the same barrier with fewer resources to clear it, for reasons PwC sets out with some precision.
The first is margin. MSMEs operate on thin margins and, as the report puts it, more often than not prioritise survival over experimentation. A failed pilot at a large enterprise is a write-off; at a twenty-person unit it can be the year's discretionary spending.
The second is the shape of the cost curve. PwC notes that while software may begin as freemium, scaling its use inevitably entails paid subscriptions, system integration and workforce training. The pilot is cheap by design. Production is where the invoice arrives, which is exactly the point at which a firm without a measured baseline has no argument for continuing.
The third is capability. Limited digital skills, uneven infrastructure, and solutions not tailored to regional contexts or informal business practices all slow adoption and dilute impact. None of these are fixed by a better model.
The scale involved is not marginal. PwC records 7,59,56,661 MSMEs registered on the Udyam portal as of 29 January 2026, and notes that manufacturing MSMEs accounted for approximately 35.4 per cent of India's manufacturing value added in FY 2023-24 and nearly 48.58 per cent of exports in FY 2024-25. Whatever the AI measurement gap costs, it is being paid across a very large base.
There is a more optimistic strand in the evidence, and it deserves stating. The Vi Business MSME Growth Insights Study 2026 found that 57 per cent of surveyed MSMEs regard AI as an important driver of business growth, and that roughly one in four has already integrated AI tools into operations. Interest is not the constraint. Proof is.
The Governance Bill Nobody Costed
The ET Edge findings on risk are, for boards, the most exposed part of the report. Eighty-one per cent of respondents named data privacy and compliance as their leading AI concern — but only 19 per cent described their organisation as highly prepared to meet obligations under India's Digital Personal Data Protection Act. Nine per cent reported no formal AI governance structure of any kind. Close to two in five reported a confirmed or suspected AI-related security incident.
Layered on top of that is speed. Sixty-four per cent of technology leaders said their organisation is actively piloting or has deployed agentic AI — systems that execute multi-step tasks and make decisions autonomously rather than drafting output for a human to approve. Among the report's risk and governance respondents, 46 per cent named agentic AI as the capability most likely to reshape enterprise operations within eighteen months.
"Every AI budget a board approves is also, implicitly, a risk the board has accepted. Data privacy exposure, concentration in a handful of AI vendors, and now systems that act without a human in the loop are no longer operational details." — Jatinder Singh, Chief Editor, Enterprise Tech Publications, ET Edge
The sequence here is the worrying part. Organisations that cannot yet prove their assistive AI delivered value are moving on to autonomous AI that acts without review, while four in five admit they are not prepared for the data protection obligations already in force. Each step compounds the previous one's unmeasured risk.
A Note of Caution on the Famous 95 Per Cent
Any discussion of AI returns eventually reaches the figure everyone quotes: MIT's Project NANDA finding that 95 per cent of organisations deploying generative AI saw zero measurable return. It is worth being honest about what that number is and is not.
The study drew on 150 interviews with leaders, a survey of 350 employees and an analysis of 300 public deployments. It is also explicitly preliminary, has not been peer-reviewed, and has been criticised for a short measurement window and a limited sample. It is a useful directional signal and a poor precision instrument.
That caution cuts both ways, and it applies to the Indian figures too. The ET Edge study is a survey of 300-plus self-reporting technology leaders, not an audit of their books; Deloitte's optimism and ET Edge's pessimism are both, ultimately, what executives said about themselves. What gives the 12 per cent figure its weight is not statistical authority. It is that it agrees with the independent, qualitative testimony PwC collected from MSME owners who described exactly the same thing in their own words — perpetual pilots, unproven value, no one accountable.
What Appears to Separate the 12 Per Cent
Neither report publishes a profile of the firms that can prove their returns. But the four board-level priorities ET Edge sets out are, in effect, a description of what the other 88 per cent did not do, and they are worth reading as diagnostics rather than advice.
The first is evidence of business ownership and data readiness before any pilot is funded — that is, a named person whose objective the project serves, established before money moves. The second is a minimum, rigorously measured ROI benchmark agreed before further investment is authorised. The third is treating data protection compliance for AI systems as a present legal obligation rather than a future initiative. The fourth is documented governance — autonomy limits, audit trails and escalation paths — before any autonomous system scales past pilot.
PwC arrives at a compatible conclusion from the opposite direction, framing AI adoption not as a one-time technology purchase but as a long-term capability-building journey requiring sustained action across government, large enterprises and civil society alongside the firm itself. Its 3A2I framework — access, acceptance, assimilation, implementation, institutionalisation — is explicitly designed so that adoption produces time-bound, measurable value rather than another pilot.
Strip both down and the common finding is unglamorous. The firms that can prove returns are not the ones that bought better AI. They are the ones that wrote down what they were trying to improve, measured it before they changed anything, and named someone answerable for the result. Every one of those steps is free, and every one of them has to happen before the software is installed.
For a smaller Indian firm reading this and wondering whether to start at all, the more accurate lesson from 2026's data is not that AI does not pay. It is that nobody can tell you whether it paid — and the cheapest thing you can do this quarter is make sure that, twelve months from now, you are not in the same position.
Sources
• ET Edge, CIO&Leader — State of AI in Indian Enterprises 2026. Survey of 300+ senior enterprise technology leaders conducted May-June 2026; findings unveiled at the 27th CIO&Leader Conference, Jaipur, 31 July - 2 August 2026. Reported 5 August 2026.
• Deloitte India — The State of AI in the Enterprise, 2026, press release "Indian enterprises lead global peers in at-scale AI adoption across most functions", 24 March 2026.
• PwC India — Unlocking the AI Edge for MSMEs, March 2026. Udyam portal registration figure as of 29 January 2026; manufacturing GVA and export shares for FY 2023-24 and FY 2024-25.
• MIT Project NANDA — The GenAI Divide: State of AI in Business, July 2025. 150 leader interviews, 350-employee survey, 300 public deployments; preliminary and not peer-reviewed.
• Vi Business — MSME Growth Insights Study 2026, AI adoption findings among Indian MSMEs.
