Enterprise AI adoption grows rapidly, yet linking AI spending to profits remains rare. Discover key findings from Genpact's 2026 survey.
Enterprise AI adoption is moving faster than the business changes needed to turn it into financial results, Genpact research released on November 12, 2026, finds. While 79% of surveyed enterprises have AI agents in production, only 15% have fully connected their AI spending to profit-and-loss (P&L) outcomes.
Summary
Key takeaways
- Just 20% of surveyed enterprises use agents consistently across functions.
- On average, 49% of funded AI initiatives met value expectations.
- Among respondents running agents, 92% maintained or expanded their human workforce.
Based on a poll of over 2,000 enterprise executives, Genpact’s report, Autonomy in Practice: How agentic operations can turn AI progress into business value, presents its findings. Its central finding: deploying agents has outpaced the redesign of enterprise operations, leaving most businesses short of what Genpact calls the “autonomy dividend”—AI-enabled improvements that translate into business performance rather than productivity alone.
Enterprise AI adoption is widespread, but broad use is not
Production deployment does not yet mean enterprise-wide use: only 20% of surveyed businesses apply AI agents broadly and consistently across functions. That contrasts with the 79% reporting agents already operating in production.
The study was conducted in 2026 across 17 industries and 14 business functions. It surveyed senior executives at companies with annual revenue ranging from $1 billion to more than $50 billion, alongside in-depth interviews with enterprise leaders.
AI investment outcomes lag deployment
Only 15% of enterprises have fully linked AI investment to P&L results, while an average of 49% of initiatives funded during the preceding 24 months delivered their expected value. Just 9% reported completing substantial changes to their operating models.
Genpact frames this gap as a problem of running AI within business operations, not simply installing it. “Enterprises have become effective at deploying AI agents. Running them is the hard part,” said Sanjeev Vohra, Genpact’s Chief Technology and Innovation Officer.
AI takes a larger decision role alongside people
AI leads or entirely handles 40% of decisions, more than twice the share a year earlier, according to the research. Human involvement remains extensive: at each decision stage, people make the decision or audit it in at least 87% of cases.
The workforce findings also show agents scaling alongside employees. Among respondents with agents in production, 92% reported either unchanged or increased human headcount.
Genpact describes autonomy in practice as a deliberate division of responsibility: defining where agents act, where people decide and validate, and which business outcomes the combined workforce delivers.
AI leaders connect operations, governance and financial ownership
The enterprises Genpact classifies as AI Leaders—17% of respondents, at the top of its AI Maturity Index—are further along in integrating people and agents into one operating model. Backed by CEO-level mandates, they lead across the five building blocks the research assessed.
Genpact calls this model Agentic Operations, combining AI, process intelligence, human judgment and governance. The report identifies five practices that distinguish the leading group:
- Redesign work across the whole enterprise rather than separately within each function.
- Deliberately develop a combined human-agent workforce, including exception specialists and agent trainers.
- Build data and AI foundations that let agents act across systems instead of only retrieving information.
- Apply security and governance throughout agents’ lifecycles, beyond their initial deployment.
- Assign named owners and specific P&L lines to AI-enabled gains.
Genpact’s research found that most enterprises have not yet connected all five building blocks.
Read also: Anthropic news today (10 November): S-1 filing targets $1.4T IPO valuation
Article produced with the assistance of artificial intelligence and reviewed by the editorial team.
