CIOs winning the AI race know when to slow down
This has direct implications for operating model and accountability. Many organizations have approval paths for launching pilots but weaker mechanisms for shutting them down once sponsors, vendors, and business teams are invested. CIOs should define who owns the decision to halt an AI use case, what thresholds trigger review, and how quickly funding can be reallocated. Without those decision rights, AI portfolios can accumulate cost and complexity faster than they accumulate business outcomes.
The budget point is equally important: AI economics are not well served by traditional project funding alone. Token consumption, model tuning, integration effort, monitoring, risk controls, and change management can all shift total cost upward after initial approval. Leaders should push for lifecycle-based costing, vendor transparency, and finance partnership that assigns spend ownership clearly across IT, product, and business units.
Practical next questions for CIOs:
- What are our formal exit criteria at pilot and production stages?
- Who has authority to stop an underperforming AI service, and on what evidence?
- Do we track full-run costs and realized benefits at the use-case level?
- How often does executive governance review AI investments for continuation versus reallocation?
ORLANDO, FLA. — CIOs who are achieving the best outcomes from their AI deployments are the ones who know when to press pause.
That might seem counterintuitive, given the pace at which AI is evolving. But Gartner’s recent CIO survey suggested that IT executives who are delivering on better business outcomes while proving AI value are more likely than their peers to establish two stopping points in the AI lifecycle — at the pilot and production stages.
“They are shifting from just scaling everything to scaling value, which means they have a very strong discipline of quitting and redirecting resources to what is more valuable,” said Alicia Mullery, vice president analyst at the consultancy, during a presentation Tuesday at Gartner IT Symposium/Xpo 2026.
There aren’t many of them. Gartner said just 11% of the 2,002 CIOs and IT executives who participated in its annual survey are currently threading that needle.
Yet Gartner found that CIOs who have established guidelines for when an AI pilot project shouldn’t move into production are 2.3 times more likely to be part of that 11%; likewise, CIOs who have established policies on when to abandon low-value AI initiatives already in production are 2.6 times more likely to be among the elite group.
Still, only 31% and 17% of CIOs, respectively, said they have established these types of policies and guidelines, even as the stakes get higher.
In 2024, organizations juggled on average 10 AI use cases; a year later, that number more than quadrupled to 42 use cases , according to Mullery.
“We are getting very good at launching AI,” she said. “We are not good at stopping AI.”
Indeed, AI budgets are increasing, but so is the pressure for CIOs to demonstrate ROI. Almost 60% of survey takers reported an uptick in pressure to deliver cost improvement from AI. Yet more than half predict that AI will increase the total cost of ownership of technology.
To better track those increases, CIOs need to adjust how they build their AI budgets. Mullery explained that many organizations budget for AI the same way they would traditional technology investments, including cloud and SaaS, when they need to consider the cost of AI over the technology’s full lifecycle, not just the initial technology purchase.
Understanding when to pause AI deployments will be critical for CIOs as they plan for next year’s budgets. While CIOs in North America expect a 2.9% IT budget increase in 2027 , Mullery noted that this boost will likely feel like a budget decline due to inflation.
To better manage AI budgets, Mullery encouraged CIOs to work with CFOs to “co-create new rules of AI economics” in the coming year. Those new rules include identifying which teams are responsible for token costs and obtaining transparency about the full lifecycle cost of AI from vendors, for example.
“Organizations like Citigroup, like Bank of New York Mellon, specifically called out how they are putting in precise measures on how to manage these token costs, and in some cases, they are putting in weekly executive governance for how to manage these investments,” Mullery said.
Enjoyed this article? Sign up for our newsletter to receive regular insights and stay connected.