Dataiku’s latest research finds that enterprise AI is entering a stage where getting projects into production is no longer enough on its own. Companies are paying closer attention to what those investments are actually delivering and whether the results can justify continued spending.
That expectation is becoming part of how technology leadership is evaluated.
In a Harris Poll survey of 685 enterprise CIOs, 88% said their professional reputation or career trajectory will be shaped by their success with AI, while 97% reported more board pressure to demonstrate measurable AI ROI compared with 2025.
The difficulty is that measuring those results gets harder as AI spreads across the organization. Agents are being built across business teams and platforms, leaving many companies to figure out how to track costs and performance while maintaining oversight of the systems already in use.
Why It Matters: Companies are asking more of AI while many are still building the systems needed to manage it. Agents can operate across business systems, make decisions and complete tasks with some autonomy, making visibility into their performance and ownership more important. The report shows a gap between how much responsibility organizations are giving AI and how well they can measure, govern and explain what those systems are doing.
- AI Results Are Carrying More Weight: 86% of CIOs said their CEO has told them that job security depends on AI outcomes, with 56% hearing that message explicitly. 76% believe their role could be at risk if their company fails to produce measurable business gains from AI by the end of 2027. Funding is also tied to performance, with 72% saying their AI budgets could be cut or frozen if targets are missed by the end of 2026.
- Measuring the Return Remains Difficult: Only 24% report mature ROI measurement across all or most AI initiatives, while 72% cannot consistently measure operational performance and business outcomes across every agent. Tracking the spending behind those results can be difficult too. Only 21% have near-real-time visibility into agent and advanced AI workload costs by business unit, team or use case, and token and compute expenses have changed how 39% plan, approve or execute AI initiatives.
- Technical Performance Only Tells Part of the Story: 79% have experienced an agent that violated business intent, policy or expectations while still operating within its technical parameters. 31% said an incident resulted in customer, financial, compliance or operational impact. Responding can also take time. Around 60% cannot contain a problematic agent within the same day, while 10% can do so within 1 to 2 hours. The findings show why monitoring an agent’s technical performance alone may miss whether its actions actually match what the business intended.
- More AI Is Being Built Outside Central IT: 84% said employees are creating AI agents and applications faster than IT can govern them, reaching 94% in the U.S. The report also found that 81% lack complete oversight of shadow AI created outside approved systems or formal submission processes. The management challenge continues after an agent is identified, with 83% lacking a standardized approach to agent lifecycle management and 47% having already decommissioned more than 20 agents this year.
- Agent Ownership Is Still Divided: CIOs gave several answers when asked who should carry primary responsibility for harmful agent decisions. 23% placed responsibility across multiple teams, 21% with central IT, 20% with the data or AI team, 18% with security, risk or compliance, and 13% with the business owner of the workflow. That division comes as 91% support allowing business teams to build AI within a governed environment. With development spread across the organization, companies need to establish who owns an agent throughout its lifecycle and who has the authority to step in when something goes wrong.
Go Deeper -> Global AI Confessions Report: CIO Edition, 2026 – Dataiku
