Tech Leaders Say AI Is Delivering But Few Track Results

Agentic AI
,
Artificial Intelligence & Machine Learning
,
Next-Generation Technologies & Secure Development

Economist Enterprise Research Reveals the Gap Between AI Optimism and Real Returns

Jennifer Lawinski •
May 14, 2026    

See Also: AI Agents Introduce a New Insider Threat Model

Four out of five executives said their AI programs are beating expectations, but fewer than half actually track whether that’s true, the report found. Economist Enterprise, which surveyed more than 1,200 senior technology executives from 18 countries including 296 CIOs, conducted qualitative interviews with technology leaders at enterprises including Disney, Mercedes-Benz, Nasdaq, Atlassian, Takeda and others.

“Everyone is jumping on the train of saying that they’re ahead of AI because the board expects this,” said Eddie Milev, who led the research. “Industries generally are quite inflated around the subject, but in practice, that isn’t always the case.”

The report introduces a benchmarking framework to evaluate which companies are generating actual AI returns and which are still stuck in pilot purgatory, such as strategy, technical foundations, governance and workforce transformation.

Among those identified as AI leaders, 84% said returns are better than expected, but only 43% require teams to measure business impact. This chasm is also visible when the survey queried strategy-oriented CTOs and vice presidents, working closer to the technology. Nearly 90% of CTOs said their AI rollouts were ahead of schedule, but only three in four senior vice presidents concurred. In IT, three in five C-level tech leaders said AI was fully embedded at scale, while two in five vice presidents agreed.

The study also found that companies are still struggling with moving AI pilot projects into production. The timeline for 58% of enterprises is still seven to 12 months, and only 40% have established AI development life cycles.

“Process is the key word here,” Milev said. “A surprising, concerning amount of firms tell us that they either have a framework but don’t apply it in full, or they just don’t have one that is developed for the whole life cycle of AI projects.”

When it comes to data governance, the survey found that 97% of firms with unified data architecture said they were seeing ROI ahead of schedule. For those without unified data architectures, that number fell to 77%. And data storage, movement and duplication were identified as the biggest ongoing AI cost by 59% of respondents. Infrastructure and compute costs, in contrast, were cited by just 25%.

“When you ask firms what is the biggest thing that worries you, the greatest share tells us it is data storage, movement and duplication,” Milev said. “That’s the consistently top thing that just sinks costs because it kind of goes hidden. The CIO should really be thinking about this, because that’s strictly within their limits.”

Executive feedback supports the data. “We have done work to retire 99% of our legacy or fragmented data, and that now gives AI the ability to answer questions much more definitively and makes the insights of AI agents more valuable,” Tal Saraf, senior vice-president, engineering and CIO, Atlassian, told Economist Enterprise.

While using clean data leads to better outcomes, governance enforcement isn’t consistent throughout project life cycles. About 59% of respondents said they conduct security reviews during development and before deployment, but only 39% continue after a system goes live. One in eight admits to reviewing governance only when something goes wrong.

Milev said this hands-off approach is driven by the ways companies have historically deployed other types of enterprise software, but it’s not a process that works for AI.

“Companies often approach AI the way they would approach conventional enterprise systems. With conventional enterprise systems, you deploy them, they don’t really change. But AI is a technology that is constantly evolving. With use, AI models actually change their behavior sometimes,” he said.

Governance around agents is also a patchwork system that companies are figuring out as they go. Three in five leading AI adopters said they now have autonomous agents doing real work, but fewer than half mandate a formal governance framework for them.

The report identified what it called a “lethal trifecta” for agentic security risk where agents have access to untrusted outside content, sensitive corporate data and the ability to communicate externally. This creates a problem as large language models can’t distinguish between underlaying data and instructions, leaving them vulnerable to malicious commands hidden in text.

“It goes to the crux of what agents need in order to work well, but also what makes them risky,” Milev said. “They need to be connected with other enterprise systems and have access to your broad data, but simultaneously be able to scout the web. And the merger of those two realities is not always a pretty picture.”

He said leading firms are creating systems to making this risk, including deploying AI gateways, giving business owners kill-switch authority and giving monitoring agents authority over other agents.

One thing the survey data didn’t fully capture, but that was consistent across interviews, Milev said, was the issue of how culture determines the success of AI projects.

Task-level job redesign, meaningful training and the right incentives matter more than the sophistication of AI systems, the report said. At the same time, while half of respondents said human review is a top ongoing AI cost, only 4% said employee upskilling was a significant expense.

“You can put the process in place and get the technology right, but it won’t be enough unless you get the culture right.” said Chas Murphy, senior vice-president for direct-to-consumer data and analytics at Disney, in the report.