MOST chief executive officers are struggling to turn artificial intelligence experiments into broader business transformation, with 82 percent saying their AI programs are achieving only some or less of their intended results, according to a Bain & Company report.
The report, Proprietary Intelligence: How to Win with AI, said only 18 percent of CEOs surveyed reported that their AI transformation had achieved most or all of its intended results. The findings were based on the Bain CEO Survey 2026, which covered 100 CEOs.
“Most CEOs think they're leading an AI transformation,” the report said, “but they're managing a portfolio of pilots.”
The report said about 85 percent of companies were not executing their AI programs well, with the problem stemming largely from how businesses were approaching transformation rather than limitations in AI technology.
The leading barriers cited by CEOs were a lack of in-house expertise and tools, cited by 43 percent; a focus on local pilots rather than broader transformation, 41 percent; and company data and platforms that were not ready for AI adoption at scale, 39 percent. Concerns about unproven returns on AI investments were cited by 36 percent, while 34 percent cited risks and legal concerns.
“Proprietary data sharpens the agents, agents sharpen the people,” the report said, describing a cycle in which employees redesign work and generate better data that can improve subsequent AI deployments.
Bain said companies advancing with AI are building what it calls proprietary intelligence, combining unique data, encoded workflows and learning systems.
The report identified seven decisions that distinguish these companies: committing to a multiyear AI strategy; concentrating investments on three to five areas where AI can change business economics; developing proprietary data and a semantic layer; building an enterprise technology architecture; redesigning workflows and workforce structures; creating systems that allow AI deployments to learn from one another; and establishing governance with clear executive accountability for AI risks.
Bain said agentic AI makes these choices more important because AI agents can plan multistep tasks, interact with business systems through application programming interfaces, maintain state over extended interactions and act with delegated authority.
“AI is not behaving that way,” the report said, referring to the ability of companies to catch up with competitors after delaying adoption of earlier technologies.
Unlike earlier technology shifts, Bain said, companies that delay AI adoption may find it harder to close the gap with early movers because advantages from proprietary data, workflows and learning systems can compound from the start.
The report cited financial technology firm Ramp and Brazilian bank Bradesco as examples of organizations developing internal AI capabilities. Bradesco rebuilt its initial agentic AI architecture after testing showed it was too slow, costly and difficult to scale safely. The reset took roughly five months, after which the bank deployed customer-facing AI in several core banking functions serving 22 million customers.
Bain said successful AI transformation also requires direct CEO involvement, focused investment and sustained spending on systems that allow organizations to learn from each deployment.
The report said companies should move beyond measuring the number of AI pilots and assess whether their investments are changing critical workflows and building capabilities that become more effective over time.
