Gartner today announced the results of its annual survey of CIOs and technology executives — 2,002 respondents, surveyed between May and July 2026 — and the headline is sobering. Only 13% say their organisation has realised significant or transformational value from AI. Roughly one in five AI initiatives succeeds. And the pressure is rising rather than easing: 58% of CIOs say the pressure to reduce costs through AI has increased, while 51% expect AI to raise their total cost of ownership.
Gartner VP Analyst Alicia Mullery summarised the lesson in four words: “volume is not value”. Many organisations have spent the past two years launching AI initiatives at pace. The survey suggests that the number launched has little to do with the value realised.
The finding that deserves the most boardroom attention, though, is about stopping, not starting. According to Gartner, the 31% of organisations that have criteria for stopping pilots before they reach production are 2.3 times more likely to be among the “Vanguard” — the group realising the most value. The 17% that have criteria for ending low-value initiatives that are already live are 2.6 times more likely. These are associations, not proof of cause: organisations with stopping rules may simply be better managed overall. But the pattern is consistent with what disciplined portfolio management looks like everywhere else in the enterprise, and it points to a capability most AI programmes have not yet built.
This blog is for boards, executive committees, and transformation leaders who own an AI portfolio and need it to produce outcomes rather than activity.
Why AI Portfolios Struggle To Stop Anything
Every mature investment discipline has a way to stop. Capital projects have stage gates. Product portfolios have sunset reviews. Venture investors decline follow-on rounds. AI portfolios, in many enterprises, have none of these — for reasons that are worth naming.
AI initiatives are often launched as experiments, and experiments feel cheap to continue. Each pilot’s ongoing cost is small enough that nobody is forced to decide, and together they absorb the scarce resources — data engineering, integration capacity, executive attention — that the most promising initiatives need.
Success criteria are frequently set after the fact. Without a measure agreed at the outset, every pilot can report something positive: users liked it, accuracy looked good in testing, the team learned a great deal. None of that is a decision rule.
Stopping also carries a reputational cost for sponsors. When ending an initiative is read as failure, sponsors keep initiatives alive. The organisations that stop well treat a clean stop as a good outcome — capital and talent returned to the portfolio — not as a mark against the people involved.
A Stopping-Rule Framework For The AI Portfolio
Stopping rules are not a brake on ambition. They are what lets an organisation fund its best AI initiatives properly. Five elements make them work.
The first is a value hypothesis written before launch. Every initiative states, in business terms, what outcome it is expected to change, by how much, and by when: hours of processing removed from a workflow, days taken out of a cycle, revenue retained, error rates reduced. The hypothesis is the yardstick every later decision is measured against.
The second is pre-agreed gates with explicit thresholds. At defined points — end of pilot, pre-production, a set period after go-live — the initiative is assessed against its hypothesis. The thresholds for continuing, re-scoping, or stopping are agreed up front, so the decision at the gate is an application of a rule, not a negotiation.
The third is a full-cost view. Gartner’s finding that half of CIOs expect AI to raise total cost of ownership is a reminder that the cost of an initiative is more than its licence or compute line. Integration, data preparation, evaluation, human review, monitoring, and change management all belong in the assessment. An initiative that delivers value at a cost higher than the value is not succeeding.
The fourth is a review for live initiatives, not only pilots. The survey’s 2.6 times association for organisations that end low-value live initiatives suggests this is the rarer and perhaps more valuable discipline. Production agents and AI features should be reviewed on a regular cycle against current value and current cost, and retired cleanly — access revoked, dependencies removed, records retained — when they no longer earn their place.
The fifth is recycling the capacity. A stop should release something: budget, engineering time, integration capacity, sponsor attention. The portfolio process should make the redeployment explicit, so that stopping the weakest initiatives visibly accelerates the strongest. That is how stopping becomes a source of momentum rather than a loss.
What This Means For The Board
For boards, the practical implication is a change in the questions asked of AI programmes. Not “how many initiatives are live?” but “how many have we stopped, and what did we do with the capacity?” Not “what did the pilot show?” but “did it meet the hypothesis we set before it started?” A portfolio report that only ever grows is a warning sign; one that shows initiatives entering, being re-scoped, and leaving is evidence of management.
This is also where cost pressure and value realisation meet. With 58% of CIOs reporting increased pressure to cut costs through AI, the fastest route to a better return is often not a new initiative but a cleaner portfolio — fewer, better-resourced initiatives measured against outcomes that matter.
The Gulf Strategy View
Gulf enterprises have been among the most ambitious adopters of AI, supported by national digital-transformation programmes that set clear, public targets. That ambition is an advantage, and stopping rules are how it converts into value: an ambitious portfolio without stopping rules accumulates initiatives, while an ambitious portfolio with them concentrates resources on what works.
The region’s compliance calendar adds a useful discipline. With UAE e-invoicing becoming mandatory for the first in-scope businesses on January 1, 2027, and ZATCA’s Wave 25 integration deadline on February 1, 2027, many finance and operations teams have fixed-date obligations competing for the same data and integration capacity as AI initiatives. A portfolio review that stops low-value AI work this quarter frees that capacity for obligations that cannot move — and for the AI initiatives that directly support them, such as invoice extraction and reconciliation.
What Inspires Us At Lynt-X Global
At Lynt-X Global, we welcome Gartner’s findings precisely because they are candid. A 13% figure could be read as disappointment. We read it as a sign that enterprise AI is entering its most productive phase: the point at which organisations stop counting initiatives and start measuring outcomes. Every transformative technology has passed through that moment, and it is usually when the real value begins.
What inspires us most is the evidence that discipline pays. The organisations that know when to stop are not less ambitious; they are more focused, and focus is what lets a promising initiative receive the data, integration, and leadership attention it needs to succeed. That is an encouraging message for every team that has felt stretched across too many pilots.
For the Gulf, we see an exciting combination. The region pairs some of the world’s boldest digital-transformation ambitions with a growing culture of clear targets and published timelines. Add disciplined stopping rules, and that ambition can concentrate into results.
What we are watching next: how many organisations publish value hypotheses alongside new AI initiatives, whether boards begin asking for “stopped and redeployed” figures in portfolio reviews, and how Gartner’s Vanguard group evolves in next year’s survey.
The Strategy Read
Gartner’s 2026 CIO survey finds that only 13% of organisations have realised significant or transformational value from AI and that roughly one in five initiatives succeeds — while cost pressure rises and half of CIOs expect AI to increase total cost of ownership. The organisations most likely to lead are associated with a discipline most portfolios lack: defined criteria for stopping pilots, and rarer still, for ending low-value initiatives already in production.
A practical stopping-rule framework has five elements: a value hypothesis written before launch, pre-agreed gates with explicit thresholds, a full-cost view, regular review of live initiatives, and deliberate recycling of the capacity a stop releases. Volume is not value. The enterprises that will turn AI activity into outcomes are the ones that know, in advance, what would make them stop.
“Every mature investment discipline has a way to stop, and most AI portfolios do not. Gartner’s 2026 CIO survey associates the organisations realising the most AI value with defined criteria for ending pilots and retiring low-value live initiatives. Stopping well is not a brake on ambition — it is how the capital, talent, and attention reach the initiatives that will actually change outcomes.”