Foundry's 25th annual State of the CIO survey, reported by Beth Stackpole and based on responses from 662 IT leaders and 249 line of business participants, lands on an uncomfortable number: only 19% of organizations say their AI initiatives have met intended goals, and another 18% concede that fewer than a third of their use cases hit targets. After three years of pilots, board pressure, and budget reshuffling, the CIO conversation has moved from ambition to accounting.
The 2026 edition reads less like a victory lap and less like a panic, and more like a procurement memo. Steering committees are forming, approval gates are hardening, and the metrics that matter are the ones a CFO would recognize. The survey frames this as CIOs taking the wheel on AI ROI rather than chasing every model release.
The 19% problem and what sits behind it
The headline figure is blunt. Per the Foundry survey, just 19% of respondents report that AI investments have delivered on the goals set for them, while 18% acknowledge that under a third of attempted use cases reached target outcomes. Sastry Durvasula, CIO at TIAA, tells CIO.com that the gap is not a technology problem so much as a discipline problem: scoping, sequencing, and shutting things down on schedule.
The barriers respondents name are equally pragmatic. Forty percent cite a shortage of in-house expertise, 32% point to ill defined ROI metrics, and 31% admit their AI strategy itself is unclear, according to the 2026 results. None of those are model selection issues. They are operating model issues.
Steering committees and approval gates become standard kit
Governance is catching up faster than capability. The survey reports that 83% of organizations either run or plan to run cross functional AI steering committees, and 53% have formal AI project approval processes in place. That is a notable jump from a year when most AI work was still moving through innovation budgets and CIO discretion.
Sean McCormack, CIO of First Student, describes the shift to CIO.com as moving from curiosity funding to portfolio management, with finance, legal, and operations sitting at the same intake table as IT. Andrea Ballinger, CIO at Rensselaer Polytechnic Institute, makes a similar point about pairing academic and administrative stakeholders so that experiments do not silently become production dependencies.
Stage gated funding at 90, 180, and 270 days
The sharpest operating prescription in the coverage comes from Thomas Prommer, a CTO and CIO advisor quoted in Foundry's companion piece on staying on the human side of AI transformation. Prommer recommends stage gated funding at 90, 180, and 270 day checkpoints, with explicit kill criteria at each gate. In his portfolios, that discipline retires roughly one in three projects before they consume a full annual budget.
The mechanics matter. At 90 days the question is whether the use case has a measurable baseline and a willing business owner. At 180 days the question is whether the pilot has produced a signal worth scaling. At 270 days the question is whether unit economics survive contact with production volume. Each gate has a default of no, not a default of continue.
The metrics CIOs are actually tracking
When the survey asked how AI success is being measured, the answers were unromantic. Forty percent named operational efficiency, 34% pointed to employee productivity, and 30% cited cost reduction. Revenue lift, customer experience, and new product creation appear further down the list.
Operational efficiency: 40%, per the Foundry data.
Employee productivity: 34%, reflecting copilots and workflow automation.
Cost reduction: 30%, typically tied to contact center, code generation, and shared services.
Durvasula's framing in the CIO.com writeup is that productivity gains only count when they show up in headcount plans, cycle times, or vendor spend. Anything softer than that tends to evaporate by the next budget cycle.
How we are rewriting intake to match the 270 day clock
In our own delivery work we are treating Prommer's 90/180/270 cadence as the default intake shape for AI initiatives rather than an advanced option. The change is structural, not rhetorical. Every new AI request now enters with a named business owner, a baseline metric pulled before any model is touched, and three pre committed review dates on the calendar.
The decision we have made is to refuse intake without those three artifacts. A request without a baseline becomes a workshop, not a project. A request without an owner becomes a research note. A request without scheduled gates does not get a budget line. That mirrors the 53% of organizations the survey identifies as having formal approval processes, and it is the cheapest way to stop the silent accumulation of half live pilots.
The signal we are watching internally is the kill rate. If fewer than 20% of AI projects are retired at the 180 day gate, the gate is decorative. If more than 50% are retired, intake is too loose. Prommer's one in three is a reasonable target and a useful benchmark when reporting to a steering committee that, per the Foundry data, 83% of peers either have or are building.
What to watch through the FY26 planning window
Most enterprise budget cycles will lock FY26 AI spend between November 2025 and February 2026. The State of the CIO 2026 data suggests two concrete things to watch in that window. First, whether steering committees actually own the kill decisions or merely review them, because committees that cannot stop work do not change outcomes. Second, whether the 19% success rate moves materially in next year's edition, since a flat number after a year of governance investment will force boards to ask whether the problem is execution or whether the underlying use cases were ever real.
The 25th edition of this survey is, in effect, the first one where AI is treated as a normal capital allocation question rather than a strategic narrative. CIOs who walk into their FY26 planning meetings with baselines, gates, and a published kill rate will be operating from the same playbook the Foundry research describes. Those who walk in with a roadmap of pilots will spend 2026 explaining the 19%.
