How do we measure whether AI is actually working?

Measure outcomes your leadership team already tracks—margin or cost per unit of work, cycle time, error or rework rates, revenue per head, or customer SLA performance—not model demos or “hours saved” anecdotes. Establish a baseline before change, define a small set of leading indicators (quality, throughput, exceptions), and review them on a fixed cadence with the same rigor as financial reporting. If the numbers do not move on business KPIs within the agreed window, treat it as a scope, workflow, or ownership problem—not a prompt tweak.