Safety performance indicators fail in two directions. Some are unmeasurable: they require data the organisation does not collect and never will. Others are perfectly measurable and completely inert — a number that goes up and down, is dutifully presented, and changes no decision.
The second kind is more damaging, because it looks like it is working.
Three questions that kill a bad indicator early
Before an indicator goes on the list, put it through these. Most candidates fail at least one.
1. Where does the number come from — precisely?
Not "from the safety database". Which field, filled in by whom, at what point in the process? If the answer involves someone remembering to tick a box that has no other purpose, the data will degrade within months. Indicators built on data the organisation collects anyway survive; indicators that require new discipline usually do not.
2. What would you actually do if it moved?
Ask it in both directions. If this indicator got 20% worse next quarter, what decision would that trigger? If you cannot name the decision, you have a statistic, not an indicator. And if it got 20% better — would you relax anything, or would you simply note it approvingly? Both answers are informative.
3. Can it be gamed without improving safety?
Every indicator creates an incentive. "Number of open findings" can be improved by closing findings hastily. "Reports received" can be improved by encouraging trivial reports. This does not disqualify an indicator — but it means you need to know the failure mode before you attach it to anyone's objectives.
An indicator nobody can influence is discouraging. An indicator that is easy to influence without doing anything useful is worse.
Exposure data is the part everyone skips
A count on its own is nearly useless. Fourteen occurrences means nothing until you know whether it was over 400 sectors or 4,000. Yet exposure data — sectors flown, hours, movements, jobs completed — is exactly the data most operators struggle to get in a usable form, because it lives in operational systems that were never designed to feed the safety office.
If you fix one thing about your indicator set, fix the denominator. Rates survive a growing or shrinking operation. Raw counts do not, and a fleet expansion will quietly make your safety performance look worse while nothing has actually changed.
A workable set for a smaller operation
Three indicators, reviewed properly, beat twelve reviewed superficially. A reasonable starting shape:
- One reporting-health indicator — reports per 1,000 sectors, or the proportion of reports that received a response. This tells you whether the rest of your data can be trusted.
- One operational indicator tied to your actual highest risk, whatever your risk assessment says that is. It should differ between an offshore operator and a business jet fleet. If yours could belong to any operator, it is probably not measuring your risk.
- One process indicator — for example, corrective actions closed within their agreed date. Slow-moving, unglamorous, and an early warning that the management system is losing grip.
The review test
Look at the minutes of your last four safety review meetings. For each indicator, find one decision that was made because of it. Indicators with no decision attached across four meetings are candidates for removal — and removing them makes room to look properly at the ones that are left.
Alert levels, honestly
Setting alert and target levels before you have a baseline is guesswork dressed as rigour. Collect data for a few cycles, see what normal looks like for your operation, then set levels that mean something. A threshold chosen from a textbook will either never trigger or trigger constantly, and both outcomes train people to ignore it.
Want help picking a set that fits your operation? Book a demo and we'll walk through it with your own setup.