MCKINSEY RESEARCH · INDUSTRY EVIDENCE
From AI experimentation to governed infrastructure value
Across its recent research, McKinsey identifies a consistent pattern: AI adoption is accelerating, but capturing measurable value at scale depends on reliable data, effective governance and integration with real operating workflows.
Nearly two-thirds of enterprises worldwide have experimented with AI agents, yet fewer than 10% have scaled them to deliver tangible value. Eight in ten cite data limitations as a barrier to scale.
Governance represents another major constraint. In McKinsey’s 2026 AI Trust Maturity Survey, nearly two-thirds of respondents identified security and risk as the leading barrier to fully scaling agentic AI. Seventy-four percent cited inaccuracy and 72% cybersecurity as highly relevant risks, while only about 30% reported maturity level three or higher in strategy, governance and agentic AI controls.
In asset-intensive infrastructure, the potential benefits are measurable. McKinsey reports that advanced analytics–enabled predictive maintenance can support typical yearly savings of 10–20% in maintenance operating expenditure and 20–30% in capital expenditure for water utilities when implemented effectively.
For public-sector infrastructure investments, McKinsey estimates that digital twins have the potential to improve capital efficiency, service accessibility and operational performance by 20–30%.
In manufacturing environments, McKinsey reports that predictive maintenance typically reduces machine downtime by 30–50% and increases machine life by 20–40%.
Taken together, these findings indicate that enterprise AI value does not come from isolated models alone. It depends on governed data foundations, operational integration, continuous monitoring and clearly accountable decision processes.
Sources
The figures presented in this article are external industry benchmarks reported by McKinsey & Company. They do not represent SINAIGROUP-AI project results, McKinsey endorsement or guarantees of future performance. Actual outcomes depend on data quality, system scope, implementation and operating context.
