The number every automation estimate omits

Nearly every automation business case runs the same arithmetic: take the hours saved, apply a loaded hourly rate, then subtract what it costs to keep the agent running. Whether that holds up is another matter: simple to verify, simple to present, and wrong in a specific, predictable direction.

Where the estimate breaks down

No agent clears everything, and that is not an engineering shortfall, it is simply the nature of the work. The real question is what the remainder costs, not just that it exists. A failed attempt costs more than an untouched task, because someone still has to read what the agent produced, judge it wrong, and then complete the work regardless.

At five hundred tasks a month and ninety-two percent accuracy, that works out to forty escalations. If each one takes twelve minutes rather than the original six, exception handling alone absorbs eight hours a month, which takes what looked like a fourteen-thousand headline figure for year one down to something nearer eleven.

Why the honest number holds up better

The naive number has usually already crossed the board’s desk, more than once. Volunteering the exception line up front does two things: it lends credibility to the rest of the estimate, and it holds up in the meeting where finance tries to take it apart. A defensible smaller figure outlasts a larger one that does not survive scrutiny.