Every automation pitch quotes hours saved. It is the wrong first number. Hours saved is a gross figure. What a team keeps is net of the time it spends checking the machine, fixing what it got wrong and explaining the new process to the people who did the old one.
The dividend is real, later and smaller
In the desks we have watched, the first quarter after an agent lands is usually slower, not faster. The saving shows up once the exceptions are understood and the humans stop double-checking every run. Budget for the dip and the payback arrives on schedule.
Where the saving actually lands
It rarely lands where the pitch said. The clearest gains we have measured are in consistency and turnaround, not headcount. Work that used to wait for one person now moves at any hour, and the variance between a good day and a bad one narrows. That is worth more than the headline hours, and it is easier to defend.
What to measure instead
Track exception rate, time-to-correct and the share of runs a human touches. When all three fall together the dividend is compounding. When exception rate falls but human touch does not, the team has quietly decided it does not trust the machine, and no amount of saved hours will fix that.