Six inputs, and the arithmetic on what a workflow costs now, what it would cost automated, and how long the build takes to pay for itself. Including the share a person still has to do, which most calculators quietly set to zero.
Based on 46 working weeks, allowing for holiday.
The build pays for itself in 8.6 months.
That frees 0.38 full time equivalents at 37.5 hours a week, and returns 3.2x on total cost over three years.
Freed hours are only money if the time is actually reallocated. A team that keeps the same headcount and absorbs more work has gained capacity, which is usually the better outcome but is not a line in the accounts. Treat the saving as the value of what those hours do next, not as a salary you stop paying.
Because automation almost never removes all of a task. Exceptions, approvals and checking the output remain. Most calculators hide that at zero and produce a flattering number. Here it is an input, floored at five percent, because a workflow needing no human involvement afterwards is not one worth modelling.
Fully loaded: salary plus employer costs, divided by hours actually worked. Using the headline salary understates it by a third or more and makes the case look weaker than it is.
Only if the time is reallocated. A team that keeps the same headcount and absorbs more work has gained capacity, which is usually the better outcome, but it is not a line in the accounts. Treat the figure as the value of what those hours do next.
Then either the task is too small to be worth automating or the residual share is too high for automation to help much. Both are worth knowing before commissioning anything, which is why the tool says it plainly rather than hiding it.
Different question, and the wrong answer is expensive. A deterministic workflow is cheaper, more predictable and easier to debug. An agent earns its cost when the input varies too much to enumerate. The AI agent readiness tool works through it.