Everyone uses it. Nobody measures it.
By Dino Nokic · SEP 2026 · 4 MIN READGartner asked 161 chief audit executives about AI this year and published the answers on Wednesday. 93% of audit leaders say their teams use AI. Fine, that's everyone. Now the rest of the numbers: 60% have no formal AI strategy. 54% haven't started measuring the return. Only 15% have defined, organized use cases. About half describe their AI use as individual and ad hoc — one tool here, another there, whoever found it first.
Think about who answered. Audit is the function whose entire job is asking "how do you know?" It's the most controlled, most documented, most evidence-driven room in the building. And more than half of it adopted AI without a plan and without a baseline. If that's the audit department, I'm not going to guess what dispatch looks like.
Ad hoc is not a phase, it's a state
The comforting version of this story is that ad hoc use is step one and strategy is step two. Everybody starts messy and tidies up. I don't buy it, because I lived the other version. Ad hoc use doesn't mature into a system. It hardens into habits. Ten people, ten tools, ten private prompts, ten different ideas of what "good enough" means, none of it written down. Six months in, the work depends on it and nobody can say where.
That's not an AI problem. That's the same foundations problem I've watched sink every technology before it. No map of the process, so nobody knows which step the tool actually touched. No baseline, so nobody can say whether it helped. No gate, so nobody owns what happens when it's wrong. The tool changed. The failure didn't.
The number that should bother you
Not the 60% without a strategy. Strategies are documents; you can write one in an afternoon and the operation won't notice. The one that matters is the 54% who haven't started measuring. Because "haven't started" means there's no before. And if there's no before, there will never be an after — only a feeling. The team feels faster. The manager feels the team is faster. The budget request says "faster." Then someone asks for the number and the room goes quiet.
I've been in that room. Years ago I bought tools on that feeling and got burned, and the honest lesson wasn't "the tool was bad." It was "I never wrote down what I was trying to change, so I couldn't tell." Gartner's survey is that lesson at industry scale.
What a baseline costs
Less than you think, which is why it's embarrassing that it gets skipped. Pick one flow. Count it for a week before anything changes: how many, how long, how many bounced back. In transportation and logistics, for us, the flow was inbound email and the count was ugly — three-plus days to react, seven to ten escalations a day. That week of counting is the whole reason I can say the reaction time went to about a minute and the escalations went to three to five in six weeks. Without the week, I'd have the feeling everyone else has. With it, I have a number that survives an audit — which, given this survey, might be more than the auditors can say.
The intelligence didn't do that. The map did. The system came after, and it was pointed at one measured seam instead of sprinkled over the whole department.
If you're in the 54%
Don't write the strategy first. Strategy without measurement is a wish with headings. Do this instead: pick the one flow where AI already showed up on its own — there is one, the survey says half of you have it — and go get the before. Ask the people using it what it touches, count that for a week, and write down who checks the output. Three lines. Then you're in the 15%, and everything you decide next has something to stand on.
Ninety-three percent adopted. Fifty-four percent never measured. The gap between those two numbers is not a technology gap. It's a map nobody drew — and the first step is a week of counting, not a strategy deck.