Ask how many indicators an organisation tracks monthly that measure something happening BEFORE an incident. The usual answer is none, and the usual reaction is surprise — because the reporting pack is thick.
Nearly everything in it counts events that already happened: incidents, lost days, near misses, first aid. Those numbers are real and they were useful. They drove the first three years of improvement, because the causes they exposed were the obvious ones — missing guards, absent permits, untrained operators. Fix those and the curve falls.
Then it flattens, and the organisation does more of what worked: more inspections, more toolbox talks, more counting. The curve does not move, because what remains is not the kind of failure a count reveals. It is the permit issued correctly by someone who did not visit the location. It is the procedure followed exactly, for a condition it was not written for.
A leading indicator measures the condition rather than the outcome: how many permits were audited at the workface rather than at the desk, how long a reported defect waited before it was closed, what proportion of planned critical-task observations actually happened. None of them is a count of harm. Each of them moves before harm does.
The hard part is not choosing indicators — it is accepting that a good one will look worse before it looks better. A reporting-culture measure rises when people start reporting, which reads as deterioration to anybody who only sees the total. An organisation that cannot explain that to its own board will quietly go back to counting.
How many indicators do you track monthly that measure something happening BEFORE an incident?