The board meeting opens with the usual slide: 47 percent of tender reviews now route through the new risk model. The CFO ticks the coverage box and moves on. No one asks whether the flagged items changed bid decisions or reduced loss rates on won work.
The metric comes straight from the vendor's usage dashboard. Every time an estimator opens the model pane it counts as an instance. The system logs no downstream effect on the submitted price, the contingency held, or the final margin variance after handover.
Finance still reconciles every project close-out by hand. The AI output sits in a separate tab that gets ignored once real subcontractor invoices arrive. The reported time saving never appears in overtime logs or headcount requests.
Procurement tracks AI-generated supplier risk scores but measures success by the number of scores generated, not by any shift in payment terms or claim frequency. The underlying contract leakage stays constant month after month.
The finance controller tried replacing the widget count with a simple cash metric last quarter: actual variation orders avoided. The slide was rejected because it lacked the green upward arrow the board expects from AI updates.
Melbourne-based peers report the same pattern. Teams optimise the visible adoption number while the P&L line items that matter remain untouched. The measurement system rewards visible activity, not transferred value.
Next time the board asks for AI success, demand the single cash or risk number that moved and the audit trail that proves it. Widget volume is not evidence.