Frame the value pool, not the use case
Most AI ROI conversations start at the use-case level. They should start at the value-pool level: where in the operation is intellectual work happening, what does it cost, and what fraction of it is bottlenecked behind information retrieval, drafting, or summarisation?
Value-pool framing surfaces the right 6 to 12 candidate use-cases, scored against NPV, time-to-value and operational risk. This is the deliverable of a Moweb assessment.
Unit economics survives audit
Per-task economics is the only AI ROI view that survives a 24-month review. Cost per claim, cost per document, cost per ticket - measured against pre-AI baseline.
We default to publishing per-task economics in every production deployment and revisit them quarterly.
Common ROI traps
Conflating proof-of-concept ROI with at-scale ROI. POC environments are forgiving in ways production is not. Adjust for the production overhead, especially evaluation and governance.
Treating ROI as headcount avoidance. The strongest ROI cases are usually quality and cycle-time improvements, not workforce reduction. Boards respond more constructively to quality framing.
Under-investing in monitoring. ROI realisation depends on continuous evaluation. Underspend here means ROI claims decay.