Only 13% of CIOs See Real Value From AI. The Gap Is in the Decisions.

The number that should change the conversation

Gartner surveyed 2,002 CIOs and technology executives between May and July 2026. Only 13% said their enterprise had realised significant or transformational value from AI so far. The same research puts the chance of any single AI use case succeeding at about one in five.

That is not a technology finding. The models are better than they were two years ago. The tools are cheaper and easier to use. Use cases have multiplied.

Volume went up. Value did not follow.

The more useful part of the research is what separates the few who are getting value from the rest. Gartner calls them vanguards: CIOs who deliver better business outcomes and whose CEOs also see high value from AI. What sets them apart is not a better model, a bigger budget or more pilots. It is a set of decisions most organisations never make.

Three decisions most organisations skip

Read the research closely and three decisions stand out.

The first is the decision to stop a pilot. Organisations with guidelines for when an AI pilot should not move to production are 2.3 times more likely to be vanguards. Only 31% of CIOs said they had them.

The second is the decision to stop something already live. Organisations with guidelines for abandoning low-value AI already in production are 2.6 times more likely to be vanguards. Only 17% had them.

The third is the decision about people. Organisations with explicit plans to redesign roles for a workforce of people and AI agents are 2.4 times more likely to be vanguards. Only 4% of CIOs said they had such plans.

None of these is a technology decision. Each is a management decision that someone has to own, make and live with.

These are survey correlations. They do not prove that writing stopping rules causes value. But they point the same way as much of what I have seen inside enterprise programmes.

I have seen this pattern in enterprise transformation: technology delivery progresses, but value stalls when nobody makes the explicit decision to retire a legacy system, stop a low-value activity or redesign a process around the new capability. The work continues because no one owns the decision to stop.

We got good at starting

Most enterprises now have a well-worn path for starting AI work. There is an intake form, an innovation fund, a centre of excellence, a vendor shortlist and a steering group that reviews a pipeline of use cases.

Each of those makes starting easier. Very few of them make stopping easier.

So the portfolio grows. Every use case has a sponsor. Every pilot shows something promising. Every live tool has a few enthusiastic users. Nothing is clearly failing, so nothing is clearly stopped.

This is not a new problem. It is the same pattern that turns a transformation portfolio into a source of fatigue: starting is rewarded, stopping is treated as failure, and capacity is spread across more work than the organisation can absorb.

AI makes it worse because starting is so cheap. A team can stand up a pilot in weeks. The cost of the pilot is small. The cost of never deciding about it is not.

The pilot gate is only the first decision

I have written before that a pilot should end in a decision: scale, redesign, stop or defer, with the criteria agreed before the first test. Gartner's finding supports that. Fewer than a third of CIOs have pilot stopping rules in place.

But the research also shows something less discussed. The weakest discipline is not at the pilot gate. It is after go-live.

Only 17% of CIOs have guidelines for walking away from AI that is already in production. Most organisations have no agreed way to retire an AI capability that is live, used and not producing enough value.

That matters for three reasons.

First, live AI keeps costing money. Much of its cost is incurred each time it runs. Gartner describes AI costs as probabilistic and determined at run time. A weak use case left running is not a sunk cost. It is a recurring one.

Second, live AI keeps carrying risk. Every capability in production needs monitoring, access controls, someone to handle exceptions and someone who notices when performance drifts. Low-value AI consumes that oversight as much as high-value AI does.

Third, live AI shapes the work around it. Once people have built their routines around a tool, removing it gets harder each quarter. The longer the decision waits, the more political it becomes.

The go-live date is not the end of the investment decision. It is where the evidence finally becomes real.

What a production stopping rule looks like

A stopping rule does not need to be elaborate. It needs to be agreed in advance, owned by a named person and tested on a date.

For each AI capability in production, I would want four things written down:

• Outcome: the business result it exists to improve, with a baseline and a target. Not usage. Not time saved.
• Full running cost: including review, correction, support and monitoring, measured per successful outcome.
• Threshold and date: the level below which it is redesigned or retired, and when that will be tested.
• Decision owner: the person accountable for making the call, who is not the person who championed it.

Then review the live AI portfolio the way you would review any other investment: on a fixed rhythm, against the agreed evidence, with retirement treated as a normal outcome rather than an admission of failure.

A stopping rule matters only when the organisation acts on it. That means naming the decision owner, setting a review date and agreeing in advance what evidence will trigger continuation, redesign or retirement. The test is whether the organisation is prepared to act when the agreed conditions are no longer met.

The people decision almost nobody has made

The 4% figure is the one I find most striking.

Most AI business cases assume the work will change. Few say how roles will change, who will do the redesigned work, what new skills are needed or how the capacity AI releases will be used. That is the decision that turns time saved into value.

Gartner also expects AI to raise overall demand for technology work rather than simply reduce it: 40% of respondents expect IT headcount to grow in 2027. The question is not only how many people. It is what each role now covers, and who decides.

Without that decision, AI is layered on top of the existing operating model. People do their old jobs plus a new tool. The efficiency number improves on a slide. The organisation does not change.

The gap is in the decisions

The 13% figure will be quoted as evidence that AI is overhyped. I think that reads it wrongly.

The technology is doing roughly what it was asked to do. The gap is in what organisations decide around it: what value is expected before work begins, when to walk away from a pilot, when to retire something already live, and how roles change once the work changes.

Those decisions are uncomfortable. They mean saying no to sponsors, admitting that a live tool is not worth its cost, and redesigning jobs rather than adding tools to them.

They are also where the value is.

The question to take into the room

For every AI capability we run today, do we know what evidence would make us switch it off, and who would make that call?

If the answer is no, the organisation is not managing an AI portfolio. It is accumulating one.

Source

Gartner (2026). Gartner Survey Finds Only 13% of CIOs Say Their Enterprise Has Realized Significant or Transformational Value from AI. Press release, 19 October 2026. 2027 CIO and Technology Executive Agenda survey of 2,002 CIOs and technology executives, May to July 2026. Accessed 29 October 2026. https://www.gartner.com/en/newsroom/press-releases/2026-10-19-gartner-survey-finds-only-13-percent-of-cios-say-their-enterprise-has-realized-significant-or-transformational-value-from-ai