Why Most Successful AI Leaders Have a Stop List
Knowing when to stop an AI pilot starts with knowing what success looks like.
It’s month nine of the AI pilot, and the steering committee is meeting again. Adoption is sitting roughly where it was in the spring. The vendor has brought a new roadmap and a confident slide about what the next release will fix. Someone suggests giving it one more quarter to prove itself, and heads nod around the table. Very few people in that room believe the next quarter will look different, but nobody wants to be the one who says it aloud.
The project got done, but done was never the outcome the organization was paying for.
None of this started with AI. Sponsors have always found it hard to end their own initiatives, and change programs that ran into trouble were rarely cancelled outright. More often they were quietly descoped, with the most meaningful deliverables trimmed back until whatever remained could be finished and declared a success. The project got done, but done was never the outcome the organization was paying for. The same behavior is now showing up in AI, and it’s producing what people have started calling zombie pilots.
Why do the leaders getting value from AI plan their exits?
Gartner released new data as its IT Symposium this week that should change how executive teams look at their AI portfolios. Only 13% of CIOs say their enterprise has realized significant or transformational value from AI so far, and the odds that any single use case succeeds remain about one in five.
The more useful finding sits underneath that headline. Organizations with clear guidelines for when a pilot should not move into production were 2.3 times more likely to be in Gartner’s top-performing group, and organizations with guidelines for shutting down low-value initiatives already in production were 2.6 times more likely. Yet only 31% and 17% of CIOs, respectively, have those practices in place. - Gartner, October 2026
If four out of five use cases won’t succeed, the ability to stop one cleanly is part of how an organization wins with AI.
If four out of five use cases won’t succeed, the ability to stop one cleanly is part of how an organization wins with AI. The strongest leaders keep a stop list, and they write it before anyone gets attached.
How will you know when it’s time to stop?
A stop list only works if you can see clearly whether a pilot is delivering. That means defining the target outcomes up front, in business terms, with KPIs attached to each one. Adoption rates and hours saved tell you people are using the tool, but they say very little about whether the organization is better off. As Gartner’s Alicia Mullery put it at this week’s Symposium, “Volume is not value.”
“Volume is not value.” - Gartner
The Wharton School’s 2026 AI Adoption Report shows how much the choice of measure matters. Among enterprise leaders tracking at least one growth measure, 35% report strongly positive ROI from AI, compared with 13% of those tracking efficiency alone.
Outcome KPIs and stop criteria belong together. Without the KPIs, any pilot can claim progress, and without the stop criteria, nobody is obliged to act on what the numbers show.
Why is topping so hard for capable leadership teams?
Most of the reasons live in the system around the pilot. Many organizations launch AI pilots without a defined business value target, so by month nine there’s nothing concrete to measure against. Gartner also points out that many organizations still fund and govern AI with models built for cloud and SaaS purchases, where costs are predictable and the decision ends at approval. On top of that, in most cultures a stopped project reads as a failed project, and someone’s credibility is attached to it.
Then the brain does the rest. Our brains register losing something we already have as more painful than gaining something of equal value. Once a pilot has a sponsor and a team invested in it, ending it feels like a loss, while the slow cost of continuing barely registers. Psychologists Daniel Kahneman and Amos Tversky called this loss aversion, and it’s operating in every person at that steering committee table at the same time.
When the system offers no stop criteria and no clear owner for the decision, loss aversion fills the gap. The default becomes one more quarter.
What are the signs your AI pilot should have ended months ago?
The definition of success has shifted since launch. If the target has quietly been redefined to match what the pilot is delivering, the pilot is now grading itself.
Status updates report activity rather than outcomes. Logins and features released show motion. Executives should be hearing about cycle time, cost, quality, or revenue.
The case for continuing rests on the next release. When the argument for another quarter is mostly about what the vendor plans to ship, the pilot has stopped being judged on its own results.
The old process is still running in parallel. People have built workarounds so the real work still gets done, which tells you the new workflow never took hold.
Nobody can say who has the authority to end it. This one sits hidden underneath all the others.
What would a stop list look like in your organization?
Start at launch, while everyone is still neutral and loss aversion has nothing to hold onto. Agree on the outcome KPIs that define succes, then agree on the threshold for those KPIs, by a specific date, that will trigger a stop decision.
Build in the two decision points Gartner identified: one before a pilot move into production, and a scheduled review once it’s live. Make the call a shared one, because Gartner is clear that CIOs eed busness oartners who are prepared to walk away from low-value investments too.
Treat a clean stop as a reported success. Show the board where the freed budget and people are going next. Over time, that’s how stopping becomes a capability your organization trusts.
“A stop only becomes a failure when nothing is learned from it.”
A stop only becomes a failure when notheing is learned from it. Every pilot that ends should leave something useful behind, starting with what the data showed and which assumptions didn’t hold. When that learning is captured and shared openly, organizations start building a cultural muscle many of them have let weaken: the wilkingness to take smart risks and learn from the results. Leaders build that muscle by going first. When a senior sponsor explains what a stopped pilot taught the organization, everyone watching learns that trying smething worthwhile and ending it honestly is safe.
A stop list on paper won’t change much by itself, because people are the oness who have to act on it. Sponsors need enough safety to say out loud what the room already suspects, and the teams who built a pilot need to see a stop decision treated as a judgment about the work rather than about them.
Making those calls in evidence, with loss aversion pulling hard in the other direction, takes a specific set of leadership skills. Leaders need the critical judgment to read what the KPIs are telling them and the emotional intelligence to recognize when their own attachment is doing the talking. They also need the adaptability to change course without treating it as defeat. Most IT and business leaders were never trained for this. Yet I still hear CIOs say the next step is sending their teams to AI training, when the capability gap that decides whether AI pays off sits in leadership. I made a relatd case in an earlier edition, Your Managers Need More Than AI Training, and the same gap shows up here at the point of decision.
The World Economic Forum’s Future Of Jobs Report points the same way. Analytical thinking is the core skill employers value most, with seven in 10 companies considering it essential, followed by resilience, flexibility and agukity. The report reads that ranking as underscoring “the critical role of adaptabikity and collabortion alongside cognitive skills.”
So here’s a question for your consideration: which AI pilot in your portfolio would you stop today if no one’s reputation were attached to it?