I’ll say something unpopular: “we saved N hours” is a weak metric for an AI project.
I’ve seen this play out repeatedly when reviewing implementations, not just in theory. Saved hours are easy to calculate and just as easy to devalue. An employee now spends an hour less on a task — but what do they do with that hour? If the process around the task hasn’t changed, the hour simply dissolves. They write an email faster — but approval doesn’t speed up. They summarize a contract faster — but Legal still checks it manually from start to finish. They draft a commercial proposal faster — but it doesn’t reflect the current pricing policy, so it gets sent back for rework anyway.
This is a classic trap: automating a single step without rethinking the process around it. We speed up one link in the chain, but the actual bottleneck stays exactly where it was — it’s just more visible now.
So when people ask me how to evaluate the effect of deploying AI agents, my answer is: don’t look at hours, look at process metrics.
- Approval turnaround time for a document, and the share of documents sent back for rework.
- The number of overdue tasks, and response SLAs.
- How fast a commercial proposal moves from request to delivery.
- Time to close a financial period.
- The number of manual touches per document.
- Management visibility: whether a manager can see where a process actually stands, not where it’s supposed to be according to the regulation.
The difference between these and “hours” is simple. Hours describe how fast a person completes their step. These metrics describe how the process as a whole behaves.
The real effect shows up when the process itself changes, not just the speed of one action inside it. Not “the employee registers an incoming letter faster,” but “the incoming letter automatically becomes a tracked task, and the manager sees execution discipline in real time, without waiting for a weekly report.” That’s a different quality of management, not a different typing speed.
Saved hours are a side effect of a properly redesigned process, not a goal in themselves. If you only count hours, you can end up with a nice number in a slide deck and zero actual change in how the company operates.
My next step is simple: don’t take any hours-based metric at face value. Instead, take one specific process and pull it apart completely — figure out what actually changes in it, not how many minutes get shaved off a single step. Take the process baseline before and after, and you’ll see the thing a CFO will actually sign off on. If you want a rough sense of the economics before that kind of audit, start with the ROI worksheet: /en/roi-worksheet. Treat it as a starting point, not a substitute for a real process review.
