
Investment requests name tools instead of calculated effects.
As CFO you decide on investments that have to pay off, and AI initiatives often arrive without the figures that carry a decision. What gets named are tools and percentages from studies, not the case volumes, handling times and running costs of your own company.
The second difficulty is operations. Classic software has licence and maintenance costs that can be planned. AI operations scale with volume, and whoever does not model that ends up with a variable cost position without a cap.
And finally you need reliable numbers for your own steering. When the forecast is off at the end of the quarter and nobody can name the cause, that is not a sales problem, it is a data problem.

Investment requests name tools instead of calculated effects.

The running costs of AI operations are not modelled.

The monthly close needs manual work in spreadsheets.

The forecast deviates without a nameable cause.

Two departments calculate the same metric differently.
01
Case volume, time per case and running costs calculated before anything is invested.
02
Model costs per case, monitoring and a cap: a plannable position instead of an open one.
03
Reports come out of the system reproducibly, independent of one person and their spreadsheet.
04
Defined stage criteria make deviations explainable, and that makes them steerable.
Four figures per initiative matter for approval: cost of the build, running cost per case, saved handling time and shortened cycle time. The last one is overlooked most often although it usually has the bigger effect: tied-up capital and lost enquiries arise in the waiting, not in the processing.
In operations a fifth one joins: cost per completed case over time. If it rises without growing volume, something is drifting, which is why monitoring is part of operations and not an optional extra.
Services from the Agentic Growth Stack, picked for this agenda.