How is AI ROI different from an ordinary business case?
The benefit side is familiar. The cost side is not: model run cost grows with use, someone has to review what the system produces, the data usually needs work, and errors carry a cost of their own. A defensible case prices all of them for one outcome, then adds them up as a total cost of ownership, with year one separate from the run rate.
Do hours saved mean money saved?
Only when spend actually goes: a role not backfilled, overtime or contractor hours cut. A task done 40% faster frees capacity. Moving that capacity to other work can be worth more than the saving, but it is value, not cost avoided, and a sound case counts it once, on the value side. Sometimes the right answer is better service, and then the case should say so rather than count payroll it will never see.
Can you put a number on it before anything is built?
An order-of-magnitude number, yes. The Assessment Sprint produces three cost paths from your own systems, with year one and the ongoing run rate separated. Where a decision needs a full financial model, we say so at the readout.
Will you tell us not to do it?
Yes, when one outcome costs more with AI and nothing it adds, such as revenue, lower risk or a capability you did not have, covers the gap, or when another route gets there for less. That answer is cheaper in week two than in month nine.
Is this only about AI?
No. The same questions apply to analytics and to the data platform underneath. AI makes them more urgent, because run cost grows with use.
