Make the AI business case survive real operating costs

The team can quote a model bill or a time-saving estimate, but finance cannot tell whether the complete workflow creates value. Volume, exceptions and support effort are uncertain. Cheap responses can still be expensive completed tasks when people redo the work. Freed staff time is not automatically cash savings, and average usage can hide peak-capacity and exception costs.

We build a cost model around the work that actually gets accepted, not the cost of a single response.

• We measure cost per accepted task, using task volume, current completion effort and observed rework or clearly marked estimates.
• Normal and exception paths are mapped, including implementation, review, retries, storage and support.
• Buy, custom cloud and private operation are compared under low, central and high workload assumptions. Private comparisons include utilization, redundancy, licensing, capacity and operating labor. No deployment is presumed cheaper.
• Missing data appears as ranges and named gaps. If no tasks are accepted, the model shows that cost per task cannot be calculated. Billed usage is reconciled separately from forecasts, and the last approved version is kept.
• Financial models and staffing assumptions stay in a restricted planning workspace, open only to the people responsible for review.

• An assumption register and sensitivity view. It shows which variable changes the recommendation, instead of one apparently precise ROI score.
• Cost per accepted task. The denominator is explicit, and exception, review and support costs are included.
• A decision memo. It sets out how benefits will be measured and the conditions under which the work should stop.

• Finance signs off the cost categories and how staff capacity is treated.
• Operations approves the baseline task definition.
• The sponsor decides which benefits are achievable and how they will be measured.

If volumes are low and ordinary software handles the process reliably, a simpler product or process change may have a stronger business case.

• The full cost of AI implementation: a planning guide
• AI investment evaluation: compare automation, analytics and assistants
• What customer conversations are worth: a business case finance can audit