The gap between putting AI agents into production and making them pay off is still wide, according to new Genpact research. A survey of more than 2,000 enterprise executives found that 79% have AI agents in production. But only 15% have fully connected AI spending to profit-and-loss outcomes.
Production Is Not the Same as Scale
Deployment numbers can look strong without showing real operating change. Only 20% of surveyed businesses said they use AI agents broadly and consistently across functions. The report covered 17 industries and 14 business functions, with executives at companies earning from $1 billion to more than $50 billion a year.
The financial link is weaker still. Just 15% of enterprises have tied AI investment fully to profit-and-loss results. An average of 49% of AI initiatives funded in the prior 24 months delivered expected value, and only 9% reported completing substantial changes to their operating models. That suggests many companies are buying or building tools faster than they are changing how work gets done. Perhaps the harder task is not deployment but day-to-day operation.
Agents Take More Decisions, People Still Check Them
AI now leads or fully handles 40% of decisions, more than double the share from a year earlier. Human oversight remains heavy. At each decision stage, people make or audit the decision in at least 87% of cases. Workforce numbers also do not show a simple replacement story. Among respondents with agents in production, 92% reported human headcount either unchanged or higher.
Genpact’s chief technology and innovation officer, Sanjeev Vohra, said enterprises have become effective at deploying AI agents, but running them is the hard part. That points to a practical question: which tasks should agents own, which should people decide and validate, and how should both be measured against business outcomes?
Leaders Tie Operations to Financial Ownership
Genpact identifies 17% of respondents as AI Leaders. These companies are further along in mixing people and agents into one operating model. They have CEO-level mandates and lead across the five building blocks the research assessed, according to the report. The study calls the wider model Agentic Operations, combining AI, process intelligence, human judgment, and governance.
Most enterprises have not connected all five building blocks yet. The result is an autonomy dividend that remains out of reach for many: AI activity rises, but financial performance does not follow at the same pace.
