NEW YORK – – November 12, 2026 -- Only 15% of enterprises have fully tied AI investment to P&L outcomes, even though 79% already run AI agents in production, according to a Genpact survey of more than 2,000 executives. On average, just 49% of AI initiatives funded in the past 24 months have delivered the expected value.
Agents reach production faster than operating models change
Genpact (NYSE: G) published the findings in its report "Autonomy in Practice." The survey covered senior executives at companies with annual revenues from $1 billion to more than $50 billion, across 17 industries and 14 business functions. Genpact also conducted in-depth interviews with enterprise leaders.
Only 20% of respondents use agents broadly and consistently across functions. Just 9% have completed significant operating model transformation. Genpact calls the unrealized upside the "autonomy dividend," meaning AI gains that translate into business performance rather than only productivity.
AI leads 40% of decisions, but humans still validate at least 87%
AI now leads or fully runs 40% of decisions, more than double the share a year earlier. Humans still decide or audit in at least 87% of cases at every decision stage.
Agents are not displacing staff. Among respondents with agents in production, 92% kept their human workforce the same size or grew it.
The 17% of AI Leaders build one workforce of people and agents
AI Leaders are the 17% of enterprises at the top of Genpact's AI Maturity Index. They are further along in building what Genpact calls Agentic Operations, an operating model that runs people and agents as one workforce. The research measured five building blocks of the model, and most enterprises have not yet connected them. Backed by a CEO-level mandate, AI Leaders are ahead on all five:
- They reinvent work across the enterprise rather than function by function.
- They build the human-agent workforce deliberately, creating roles such as exception specialists and agent trainers.
- They strengthen the data and AI foundation so agents can act across systems, not just retrieve information.
- They govern and secure agents throughout their lifecycle, not only at deployment.
- They treat value capture as a P&L discipline, tying AI-enabled gains to named owners and P&L lines.
"Enterprises have become effective at deploying AI agents. Running them is the hard part," said Sanjeev Vohra, Chief Technology and Innovation Officer at Genpact.
The findings are self-reported, and percentages are rounded.