Nobody wrote the stopping rule

Gartner put its 2027 CIO agenda on stage in Orlando on Monday. The survey behind it ran from May to July and covered 2,002 CIOs and technology executives. The headline: only 13% say their enterprise has realized significant or transformational value from AI so far. Use cases have multiplied for two years. The chance that any one of them succeeds is still about one in five.

One in five. If a dispatcher booked loads at that rate, you'd be having a conversation by Wednesday. But the 13% isn't what stopped me. What stopped me was the thing that separates the CIOs who are getting value from everyone else.

The winners have an exit

Gartner calls them vanguards: CIOs who deliver better business outcomes and whose CEOs also rate the AI value as high. What they do differently is boring. They write down, in advance, when to stop. Organizations with guidelines for when a pilot should not move to production are 2.3 times more likely to be in that group. Organizations with guidelines for abandoning low-value AI that's already in production are 2.6 times more likely. And only 31% have the first rule. Only 17% have the second.

Read that again. Eighty-three percent of these companies have no written rule for switching off an AI system that is live and not paying. They have a start button and no stop.

Alicia Mullery's line from the stage was "Volume is not value." The rest of the sentence matters more: AI is creating a reckoning for organizations that lack fundamental disciplines, such as defining use-case value ahead of time and establishing criteria for when to walk away. Fundamental disciplines. Not better models. Foundations.

Why nobody writes it

Because a stopping rule is a confession written before the fact. It says: this might not work, here is how we'll know, and here is who gets to call it. In most companies nobody wants their name on that paragraph. So the pilot gets extended, re-scoped, relabelled as learning, and eighteen months later it's a line item nobody can explain. Gartner published its CEO survey the next day, 203 CEOs and senior executives: 44% struggle to demonstrate AI-driven financial returns to stakeholders. That's what the missing paragraph costs.

And the money is getting tighter around it. 58% of CIOs report increased pressure to deliver cost improvement from AI. 51% expect AI to raise the total cost of ownership of their technology. Both at once. Mullery's explanation is the one I'd give: AI costs are probabilistic and determined at run time. You can't budget a thing you can't stop.

What a stopping rule looks like in an operation

In transportation and logistics this isn't exotic. Every dispatcher carries a version of it. A lane that stops paying gets dropped. A broker who pays late three times gets cut. Nobody calls a strategy meeting; the threshold was set before the load was booked. The same thing works for an intelligent system, and it fits on an index card.

The number it has to move, measured before it starts. For our inbound system that number was reaction time: three-plus days before, about a minute after. If it hadn't moved, the system would have gone. The date by which it has to move. The name of the person who calls it, who does not need anyone's permission to. And what "stop" means in practice: who takes the work back, what gets switched off, what gets archived.

Four lines. That's the 17% club.

My contrarian bit

The industry frames walking away as failure. It isn't. One in five is the base rate. If you run ten pilots and kill eight on schedule, you haven't failed eight times. You've paid the going price for the two that work, and you paid it in weeks instead of years. The company that can't stop anything isn't bold. It's the one paying for all ten.

And notice what the rule actually is. A named human, a threshold, and permission to say no. That's the gate. A stopping rule is the human gate pointed at your own project instead of at the model's output.

One thing to do this week

Take the AI system you're least sure about. Write the four lines. If you can't fill in the first one, you've found the real problem, and it isn't the model.

Thirteen percent got the value. The difference wasn't a better model. It was a paragraph that says when to stop, written before anyone had a reason not to.

Sources: Gartner, "Gartner Survey Finds Only 13% of CIOs Say Their Enterprise Has Realized Significant or Transformational Value from AI" (press release, October 19, 2026; 2027 CIO and Technology Executive Agenda survey of 2,002 CIOs and technology executives, May to July 2026) · Gartner, "Gartner Says 44% of CEOs Struggle to Demonstrate AI Financial Returns as Technology Becomes Central to Growth" (press release, October 20, 2026; 2027 CEO Survey of 203 CEOs and senior business executives, April to May 2026)