Two surveys released this month looked at AI value from opposite ends of the market, and they arrived at similar findings. Gartner asked 2,002 CIOs and technology executives, and only 13% said their enterprise has “realized significant or transformational value from AI so far.” Lever Talent, a workforce consultancy, asked 484 mid-market leaders, and while nearly all of them have adopted AI, only about one in four calls it extremely valuable.
Most leaders of the organizations getting the most value from AI had made two decisions the others hadn’t. Ask yourself whether, for every AI initiative your business is running, your leadership team could answer two questions. What would make us stop? And who owns the output?
More AI Initiatives Don’t Improve the Odds
Neither survey found a shortage of AI activity. Gartner reports that use cases have accelerated over the past two years, but the odds that any one of them succeeds are still about one in five. In the mid-market, almost every company Lever surveyed plans to spend more on AI next year. The budgets are growing faster than the results.
At their IT Symposium/Xpo in October, Gartner’s Alicia Mullery summed it up as “Volume is not value. 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.”
Taken together, the two reports point to two disciplines that sit upstream of any choice of model or platform, one for each of the two questions. Both are decisions a leadership team makes before the build, and both are within reach this quarter.
Every AI initiative needs an exit and an owner
What would make us stop?
It’s easy to picture how an AI pilot drifts. It shows some promise, nobody wants to be the one to kill it, so it moves into production because stopping would feel like admitting a mistake. And it keeps running, and costing money, because nobody agreed up front on what result would end it.
Gartner found that the CIOs it calls vanguards, those delivering better business outcomes while also reporting high CEO-perceived AI value, were much more likely to have stopping rules at two points. One rule decides when a pilot shouldn’t move to production. The other decides when to abandon AI that’s already running in production. That second rule is the rarer one, and only 17% of CIOs said they have it.
That rule may also be the one that matters most as AI moves into everyday operations, because AI costs don’t behave like traditional technology costs. Traditional software does the same work the same way every time, so its costs are fairly predictable. AI is probabilistic, so what it costs can’t reliably be determined before run time. A pilot gate can catch the idea that shouldn’t scale, but not the initiative that cleared the gate and then cost more to run than anyone expected.
Mullery was clear that ownership of that decision doesn’t belong to just one person. “CIOs need business partners who are willing and ready to walk away from low value investments.” The technology leader can bring the numbers, but the executives who sponsored an initiative need to agree beforehand on what result would end it, and then honor that agreement later. That’s easier to settle before anyone’s reputation is tied to the outcome.
If nothing in your AI portfolio has ever been stopped on purpose, it’s worth asking whether anyone is actively choosing what’s in it.
Who owns the output?
When Lever asked leaders to describe a workflow where people and AI work well together, the answers sounded similar. One described customer support triage as “AI drafts replies and summarizes tickets. But human agents verify and personalize before sending.” Another described software development, “with AI helping with code generation and code review and employees providing requirements, testing and final approval.”
Each response describes a decision someone made about where AI’s work ends, where a person’s judgment begins, and who is responsible for the result.
Lever measured this with a pair of survey questions. Is the division of responsibilities between people and AI clear, and is it clear who catches and fixes a wrong AI result? Among companies that said yes to both, 51% called AI extremely valuable compared to 9% among all others. Lever describes this as work that “costs little more than leadership attention,” yet about half of the companies it surveyed haven’t done it.
The gap is widest where AI works on its own. Most of the companies using agents for multi-step work couldn’t say that a person reviews every one of those workflows before the work moves on. Gartner sees the same unmade decision from the enterprise side, where only a small fraction of CIOs have explicit plans to redesign roles for a workforce of people and AI agents.
Why these decisions go unmade
When Lever asked respondents about their biggest obstacle, one leader described “coming to a consensus of which AI to use and how it will be used, which department or teams will do what.” Another put it more simply: “People just don’t know where to start.”
Both of those describe a decision with no owner. Nearly half of the mid-market companies Lever surveyed have no single person accountable for how people and AI work together, and where no one owns it, leaders are far less confident they know how to divide the work. When a decision belongs to everyone, it tends to get made at the team or individual level, made inconsistently, or not made at all.
Timing matters as much as ownership. Companies that bring their people leaders into AI decisions before tools are selected report much stronger results than those that bring them in afterward. I interpret that as evidence that in starting with a choice of a tool, rather than leadership making decisions about how work will change, the features of the tool end up shaping a lot of those decisions by default.
Both decisions are also collective. A stop rule needs the sponsor, the finance lead and the technology lead to agree before the results come in. A division of labor needs the people who run the work to agree on where AI’s output ends and their judgment begins. Neither one can be bought from a vendor, and neither holds up if only one leader believes in it.
Starting with a pilot doesn’t negate planning
The strongest counterargument is the familiar advice to skip the planning and start with a pilot. There’s real research behind it from studies of staged investment under uncertainty which found that committing in stages with an explicit option to abandon outperforms committing everything up front. That’s a good argument for pilots, and it’s also an argument for stop rules. A pilot is valuable because you can walk away from it. Without an agreed reason to walk away, a pilot stops being an option and becomes the first step towards a production commitment.
The mid-market read
Lever’s data is unusual because it’s about the mid-market, companies with 100 to 999 employees, most of them between $10 million and $500 million in annual revenue. Much of the research leaders see describes large enterprises with governance teams and transformation offices that mid-market firms don’t have.
The understanding that spinning up new initiatives doesn’t automatically translate to more value isn’t scarce anymore. What does seem scarce is leadership teams that have actually answered the two questions. In a mid-market company, the people who need to make that agreement can often fit in one conference room (or one screen of a Zoom gallery), which makes these two decisions more readily achievable.
Two lines per initiative
A practical starting point doesn’t require new technology. Pick your three largest AI initiatives, and for each one, ask members of the leadership team to write two lines separately:
- What result, by what date, would make us stop this?
- Who checks what it produces, and who owns the error when it’s wrong?
Then compare the answers. Where they match, you’ve already made both decisions. Where they’re blank or different, you’ve found the conversation to have before the next dollar goes in.
So, for the initiatives you’re running right now, could your leadership team answer the two questions?
Neither decision needs a new platform or a bigger budget. Both need a leadership team willing to decide together, before the next initiative starts.
Sources
In order of appearance.
- Gartner (October 19, 2026) — “Gartner Survey Finds Only 13% of CIOs Say Their Enterprise Has Realized Significant or Transformational Value from AI”. Gartner 2027 CIO and Technology Executive Agenda survey (n=2,002, May to July 2026); source of the 13%, the one-in-five success odds, the stopping-rule and role-redesign findings, and the Alicia Mullery quotes.
- Lever Talent (October 2026) — The Co-Intelligent Workforce Report: 2026 Baseline Edition. Survey of 484 Director-and-above leaders at US and Canadian companies with 100 to 999 employees (August 17 to September 14, 2026); source of the mid-market adoption, design-clarity, ownership and HR-timing findings and the respondent quotes.
- Ipsmiller, E., Brouthers, K. D., and Dikova, D. (2019) — “25 Years of Real Option Empirical Research in Management,” European Management Review 16(1), 55–68. Review of the staged-investment research behind the pilot counterargument.