Only 13% of CIOs See Real AI Value. The Missing Discipline Is Knowing When to Stop.

This week Gartner brought its IT Symposium/Xpo to Orlando, about an hour down the road from us. The headline from its new CIO survey should stop every CEO and business owner in their tracks: only 13% of CIOs say their organization has gotten significant or transformational value from AI.

Thirteen percent. That's after two years of pilots, copilots, vendor demos and "AI strategy" decks.

I don't think the technology is the problem. After 25+ years leading technology organizations, I see a simpler issue: most companies are very good at starting things and very bad at stopping them. AI has just made that weakness more expensive.

What Gartner Found, and Why It Matters for Mid-Sized Companies

• •2,002 CIOs and technology executives were surveyed between May and July 2026
• •Gartner puts the odds of any single AI use case succeeding at roughly 1 in 5
• •58% of CIOs say pressure to deliver AI cost improvements has increased, and 51% expect AI to raise total cost of ownership
• •Organizations with rules for when a pilot should not move to production are 2.3x more likely to be top performers. Only 31% have those rules.
• •Organizations with rules for retiring low-value AI that's already in production are 2.6x more likely to be top performers. Only 17% have them.

A Fortune 500 company can absorb a dozen failed pilots. A $50M–$500M company or a nonprofit can't. Every dollar and every hour of your best people's time spent on an AI experiment that goes nowhere is a dollar and an hour not spent on something that works.

Activity Is Not Value

Here's what I typically find when I walk into a mid-sized company today. I call it pilot sprawl:

• •Copilot licenses that most people stopped opening after week two
• •A customer chatbot marketing launched without a success metric
• •An AI add-on in the ERP that got switched on because it was "included"
• •A data science experiment owned by someone who has since left

Each one had a sponsor. None of them had a finish line.

AI makes this worse because the costs behave differently. Traditional SaaS is a predictable license fee. AI costs are often usage-based and set at runtime, so a pilot that looks cheap with 20 users can get expensive fast with 400.

Set Your Stop Criteria Before You Start

Before any AI use case gets funded, answer these questions in writing:

• •What business metric moves, by how much, and by when? (hours saved on invoice processing, days cut from month-end close, support tickets deflected)
• •What's the baseline today?
• •What's the full cost at production scale? That means licenses, usage, integration, data cleanup, support and training.
• •What result at 60–90 days means we stop?
• •Who on the business side owns the outcome? It shouldn't just be IT.

If you can't answer these, you don't have a pilot. You have an experiment with no budget ceiling.

Prune What's Already Running

Most companies already have AI in production that isn't earning its keep. I recommend a quarterly AI portfolio review, run with the same discipline as a program health check:

• •Inventory every AI tool, add-on and pilot, including "shadow AI" bought on a company credit card
• •Compare actual usage and results against the original promise
• •Classify each one: Scale, Fix or Stop
• •Reallocate the freed-up budget to the one or two use cases that are actually working

The hard part isn't analytical, it's political. Gartner made the point that CIOs need business partners who are willing to walk away from weak investments. That's where an experienced outside voice helps: someone with no attachment to the pilot and no vendor commission riding on it.

Governance Doesn't Have to Be Heavy

Gartner also reported that 71% of organizations expect to deploy agentic AI within 12 months, while only 21% have governance in place for it. For a mid-sized company, AI governance doesn't mean a new committee and a 60-page policy. It means:

• •One accountable owner for the AI portfolio
• •A one-page intake form that includes the stop criteria above
• •Clear rules on what company data AI tools can and can't touch
• •A 30-minute monthly review of AI spend and results

What to Do Next: A Checklist

• ☐List every AI tool, add-on and pilot running today, with its owner and monthly cost
• ☐Write down the original business case for each one, or note that there wasn't one
• ☐Score each one Scale / Fix / Stop
• ☐Cancel or pause at least one low-value item this quarter
• ☐Require a one-page intake with value targets and stop criteria for every new AI request
• ☐Budget AI usage for 2027 as a variable line item, not a fixed license
• ☐Name a business owner for every use case that survives

Related Reading