“Hours saved” is the first metric every AI project reaches for, because it is the easiest one to claim. It is also the easiest one to be wrong about, and 2026's data on AI ROI shows exactly why.
The productivity paradox
Workday's January 2026 research, a survey of 3,200 leaders and employees at large organisations, found that nearly 40% of the time people saved with AI was lost again to rework: correcting errors, rewriting low-quality generated content, verifying what the model produced. That is close to four hours back out for every ten saved. The “hours saved” number was real. It just was not the number that mattered.
- 56% of CEOs say they have realised neither higher revenue nor lower costs from AI over the last 12 months (PwC Global CEO Survey, January 2026).
- 12% report both: lower costs and higher revenue. Overall, 26% saw costs fall and 30% saw revenue rise (same survey).
- 22% of senior finance leaders say they can tie AI spend to business results today (CloudZero's own survey of 260 finance leaders, June 2026; CloudZero sells cloud and AI cost tooling).
What a hard-dollar number looks like
A hard number has a before and an after, measured on something the business already counted. Unplanned downtime events in a quarter, and what each one cost in labour, parts and lost output. Days between job completion and cash in the bank. Jobs that had to be revisited. Put a dollar value on the before, change one thing, and count again. That is the standard “hours saved” rarely clears, because hours saved is an estimate of time that might otherwise have been spent on something valuable, not a figure anyone can audit.
The numbers worth tracking instead
Quote turnaround. Error and rework rate. Cash days. On-time delivery. Fuel and energy where relevant. These move in a direction you can point to, and they are numbers the business was already keeping an eye on before AI entered the picture, which makes them much harder to dress up.
Where to go from here
Every AI Transformation Strategy ends with a baselined measurement plan: quote turnaround, rework rate, cash days and on-time delivery, recorded before anything ships. For why the order of the work decides what you can measure, read Automate, Connect, Delegate. For why so many projects never get far enough to measure anything, read why most AI projects stall.