In two years, India's national AI Adoption Index moved from 2.45 to 2.47, on a scale that runs to 4. That's the aggregate score NASSCOM tracks across sectors, and it barely shifted despite what feels, anecdotally, like every company claiming an AI strategy in the same period.
The flat national number hides a sharper split underneath it. Adoption isn't spreading evenly, it's concentrating in a handful of industries and use cases while the rest experiment without much to show for it. Globally, tech and software companies report AI use in 88% of organizations, financial services follows at 79%, healthcare has climbed to 62%, and retail sits at 53%, mostly for demand forecasting and personalization. Sixty-two percent of companies worldwide are at least experimenting with AI agents for workflow tasks, according to McKinsey's most recent global AI survey. Only 39% report that activity actually moving enterprise profit, and just 6% of companies qualify as true AI high performers.
Where AI adoption is highest, by industry
NASSCOM's research on Indian tech SMEs explains why so much of that activity stalls before it compounds. Most companies here follow a project-specific approach to generative AI rather than building any real strategic capability, adopting reactively when a specific problem forces the issue instead of planning ahead. ROI stays uncertain because nobody measured a clean before-and-after on the first project, so the second one never gets funded with confidence. That uncertainty is the actual bottleneck, not the technology itself.
The projects that do compound share a pattern: a narrow, bounded job with a number attached. AI-driven predictive maintenance in manufacturing has cut equipment downtime by 45% and maintenance costs by 25% in reported deployments, a concrete swing a plant manager can point to in a budget meeting. Invoice extraction that used to eat three re-keying sessions a week now runs unattended in minutes. A support inbox that once needed a person triaging every ticket now routes the routine ones on its own. Each of these has a clear before, a clear after, and one owner who can vouch for both.
Where to start, based on what's actually working
None of this requires an enterprise AI roadmap. It requires picking the task that's already costing the most hours and proving the swing on paper before moving to the next one. That's the sequence separating the 6% pulling ahead from the 94% still running pilots, and it's the same sequence we walk clients through, starting with a short audit of what's actually eating their team's week.