Definition
The proportion of the economically usable action window that AI detection actually delivers, measured per confirmed detection as lead time provided versus lead time required to take the cheapest available intervention path.
The Measurement Gap
There is no metric that tells you whether the lead time your AI detections provide is long enough to change the intervention economics, which means a program can report high detection rates while nearly every detection still resolves as unplanned work.
Formula
Plain language: For each confirmed detection, record the lead time delivered (detection to functional failure or forced intervention) and the lead time required to reach each intervention tier: emergency, expedited, next scheduled window, deferred-to-turnaround. Score each detection by the cheapest tier its lead time made reachable, as a fraction of the value of the cheapest tier theoretically available for that failure mode. TSDV aggregates these fractions.
Notation:
TSDV = Σ_d (V_achieved_d / V_best_d) / N_detections × 100
Variables:
- V_achieved_d = avoided-cost value of the cheapest intervention tier actually reachable with the delivered lead time for detection d (currency, DERIVED from intervention cost model)
- V_best_d = avoided-cost value of the cheapest tier theoretically reachable for that failure mode with ideal lead time (currency, DERIVED)
- N_detections = confirmed detections in the period (count, INTEGRATION from CM alerts joined to work orders)
Proposed formula, pending validation pilots.
At a Glance
Decision Supported
Our detections are "early", but early enough for what? Did the lead time let us plan the repair into a scheduled window, or did we still end up in an emergency, just a predicted one?