Chief information officers who achieve the strongest results from artificial intelligence are the ones who know when to hit the brakes. A presentation by Gartner analyst Alicia Mullery at the firm's IT Symposium/Xpo 2026 in Orlando revealed that only 11% of IT executives are both delivering superior business outcomes and proving AI's value. These elite leaders share a common discipline: they've created formal stopping points in the AI lifecycle, both at the pilot phase and during production.
The numbers tell a striking story about discipline and resource allocation. CIOs who've written guidelines for when a pilot shouldn't advance to production are 2.3 times more likely to land in that elite 11%, Mullery reported. Those who've adopted policies to shut down underperforming AI efforts already running in production are 2.6 times more likely to join the top tier. Yet only 31% of the 2,002 CIOs surveyed have set guidelines for halting pilots, and just 17% have policies for abandoning production initiatives. The stakes have climbed sharply: organizations managed an average of 10 AI use cases in 2024, but that figure jumped to 42 use cases by 2025. Nearly 60% of survey participants face mounting pressure to demonstrate cost savings from AI, while over half expect AI to drive up total technology ownership costs.
According to Mullery, the problem is clear: "We are getting very good at launching AI. We are not good at stopping AI." The top-performing CIOs are "shifting from just scaling everything to scaling value," she explained, "which means they have a very strong discipline of quitting and redirecting resources to what is more valuable." The Gartner survey suggested that IT leaders achieving better outcomes are more inclined than their counterparts to build these two decision gates into their AI operations.
The reason elite CIOs are pumping the brakes lies in how organizations budget for AI. Many firms treat AI spending the same way they handle traditional technology purchases like cloud services or software subscriptions, Mullery noted. But AI demands a different approach: CIOs need to account for costs across the technology's entire lifespan, not simply the upfront buy. That includes ongoing expenses like token costs, which some institutions are tracking with weekly executive oversight. Mullery pointed to Citigroup and Bank of New York Mellon as examples of organizations "putting in precise measures on how to manage these token costs." With North American CIOs expecting just a 2.9% IT budget bump in 2027—a figure that will feel like a cut once inflation is factored in—managing AI economics becomes even more urgent.
Looking ahead, Mullery urged CIOs to partner with chief financial officers to "co-create new rules of AI economics" in the year ahead. That collaboration should include deciding which teams bear responsibility for token expenses and securing full lifecycle cost transparency from vendors. As AI budgets swell and return-on-investment scrutiny intensifies, knowing when to walk away from an initiative will separate the leaders from the laggards. The takeaway is straightforward: success in AI isn't about launching the most projects—it's about keeping only the ones that deliver real value. Organizations that master the art of strategic abandonment will find themselves with leaner portfolios, sharper focus, and budgets that stretch further in an era when every dollar counts.