getty

For CIOs, the AI performance of their enterprises has become personal.

They believe their success in turning AI projects into real business value could make or break their careers, according to new research from AI platform Dataiku*.

Almost all say AI outcomes shape their careers, and nearly three-quarters (74%) believe their job is at risk if measurable gains don't materialize. AI is now a standing agenda item at the board level, and 98% say board pressure to show measurable ROI has increased since 2024.

“Eighteen months ago, the conversation in most boardrooms was about getting started,” says Dataiku CEO Florian Douetteau, “today it is about proving the spend.”

But demonstrating AI’s value is only half the problem. Boards and regulators aren't just asking whether AI is delivering results — they're asking how AI tools reached their decisions. Nearly three in 10 (29%) CIOs say they’ve been asked six or more times in the past year to justify or defend an AI outcome they couldn’t fully explain.

“The CIO has become the person accountable for what AI is doing in the business in real time,” adds Douetteau.

Dataiku’s Three-Part Formula

What separates AI success from failure, explains Douetteau, isn't a technology decision. It comes down to three interdependent things: ensuring the right people are building AI, establishing orchestration that connects every data source and system in use and governance with no blind spots. If one of these components is missing, AI will reliably fail.

People

The most valuable AI use cases often originate outside of the IT department. They come from the underwriters, the supply chain planners, the analysts — the people who know the enterprise’s problems and opportunities best. But many organizations still build AI the other way around: technical teams develop the solution, hand it over and then see adoption stall.

"The people who understand the work were consulted, not involved," says Douetteau. "People defend what they helped build. They tolerate, at best, anything else."

Giving non-technical teams the ability to create their own solutions — without depending on an IT queue — is how AI capability scales across the enterprise. Ninety-four percent of CIOs see low- and no-code capabilities as critical to doing exactly that. But expanding who builds creates new risk if what gets built can't be inspected, validated or defended.

Orchestration

Enterprise AI will never live in one place. Data sits across multiple systems, models vary by use case and agents multiply faster than any single team can track. The CIO's job isn't picking the right tool — it's making all of them work together as one coherent system.

Nearly all (93%) agree that different LLMs perform better for different use cases, and 81% expect to rely on two or more LLM providers in 2026 just to stay competitive.

"The orchestration that matters now is across kinds of intelligence," says Douetteau — large language models, predictive models trained on the company's operational history, analytics on governed data, business rules and human judgment at the points where it still needs to apply. "The CIOs who got this right understood that no single kind of intelligence is enough on its own, and that the discipline is in putting them together … Visibility and control across everything running matters more than any one tool underneath.”

Governance

Free AI tools are everywhere, and employees are using them. Fifty-four percent of CIOs have found unsanctioned AI running inside their organizations, and 82% say employees are creating AI agents and apps faster than IT can govern them. The costs are already showing up, with 85% saying that traceability gaps have delayed or killed projects before they reached production. For CIOs, the challenge isn't whether to enable — it's how to do it without losing control.

"The instinct to restrict is reasonable," says Douetteau. "Something feels risky, you build a fence around it. The trouble is that AI breaks the fence by design. The tools are free. They sit on every employee's phone." The CIO's only real lever, he argues, is making the sanctioned path easier than the unsanctioned one. If it is, governance becomes possible. If it isn't, the organization spends its energy chasing employees who are simply trying to do their jobs faster.

The sanctioned path only works if it comes with the controls to back it up. That means one place where the organization can see every model, every agent, every dataset — regardless of which cloud it runs on or which LLM produced the answer.

Looking To The Future

AI has become a performance system with direct consequences for the executives responsible for it. The organizations that pull ahead won't necessarily be the ones that moved fastest; they'll be the ones that get all three fundamentals right.

Dataiku is built around that formula — bringing domain experts into the AI building process, connecting models, agents and data across multi-vendor environments and providing the governance and visibility leaders need to keep every AI decision traceable and defensible.

The job is no longer building AI systems. It is operating them. The CIOs who treat that as a discipline, with people, tools and budgets attached, will end the year in a different position than those who treat it as an afterthought.

*Research was conducted online by The Harris Poll on behalf of Dataiku (December 2025–January 2026), surveying 600 CIOs at companies with annual revenue of $500M or more across the US, UK, France, Germany, UAE, Australia, Japan, South Korea and Singapore.