← Insights

Safety metrics

Safety KPIs that tell the truth

July 2026 · 7 min read

Walk onto most sites and you will find a sign, or a screen, doing the reassuring: DAYS SINCE LAST ACCIDENT, climbing in big friendly digits. It is the most photographed safety metric in industry, and taken alone it tells you almost nothing about how safe the operation actually is this month. A number that only moves in one direction, and only resets when something goes wrong, rewards silence far more than it rewards safety.

The sign is a symptom of a bigger habit. Most safety spreadsheets are built the way a finance spreadsheet is built - a grid of numbers, a splash of red and green, a total at the bottom - without ever asking the one question that actually matters for safety data: which way is better? Get that wrong and the dashboard does not just fail to help. It actively misleads.

Leading and lagging - and why dashboards default to one

A safety metric is either lagging or leading. Lagging indicators count harm that has already happened - a recordable injury, a lost-time incident, a day off work, a first-aid case. They are precise, easy to audit, and the first thing a regulator or an insurer will ask for. They are also, by definition, too late: by the time a lagging indicator moves, someone has already been hurt.

Leading indicators count the activity meant to prevent that harm before it happens - a near miss reported, a scheduled inspection completed, a toolbox talk delivered, a corrective action closed on time, a training record kept current. They are noisier and easier to argue about, which is probably why so many dashboards quietly drop them: it is much simpler to total up injuries than to decide whether this month's inspection completion rate is good enough.

A dashboard built only on lagging indicators is a rear-view mirror. It tells you, accurately, what already happened, and offers nothing about what is happening now that will determine next month's number. A useful safety dashboard carries both, and treats them as opposite ends of the same story rather than two unrelated pages.

The colouring trap

Here is where most spreadsheets quietly break. Somebody builds ten KPI cells, picks one colour rule - green if the number is high, red if it is low, or the reverse - and copies it across the whole row. It looks consistent. It is also wrong for at least half the KPIs on the sheet.

Take near-miss reporting. A site that logs more near misses this month than last is, in reporting-culture terms, doing better: people are noticing hazards and telling someone, instead of walking past. Colour that rise red, because the formula just checks whether the number went up, and you have told the workforce, in the clearest language a spreadsheet can manage, that reporting is a mistake. The next month's near-miss count duly falls, the cell turns green, and the site has not become safer. It has become quieter.

A rising near-miss count is the sound of a reporting culture waking up. Colour it red often enough and the workforce learns to keep it asleep.

The fix is not complicated, it is just a step most templates skip: every KPI needs a declared direction. Higher-is-better metrics - near misses reported, inspections completed, toolbox talks delivered, actions closed on time, training compliance - should go green when they rise. Lower-is-better metrics - recordable incidents, lost-time incidents, days lost, the incident rate itself - should go green when they fall. The colour must always mean the same thing: better. Which direction "better" points is a property of the metric, not an accident of the formula.

One rate, one common base

The incident frequency rate is the number most safety reports lead with, and it earns the position because it does something a raw count cannot: it lets you compare a site of forty people with a site of four hundred, or this year with last year after headcount changed. The calculation is simple in principle - take the recordable incidents for the period and scale them against the hours actually worked, against a standard base.

The base is where people disagree, and mostly it does not matter which one you pick, only that everyone in the comparison uses the same one. Per 100,000 hours is common where you want the rate to move visibly on modest incident counts. Per 200,000 hours, roughly the annual hours of a hundred full-time employees, is a familiar convention. Per 1,000,000 hours suits larger or lower-risk operations where a smaller base would make every rate round to zero. None of the three is correct on its own; what matters is that the rate is calculated from your actual recordables and your actual hours worked every period, not kept as a manual side calculation that someone forgets to update the month hours dip because of a shutdown.

Why sums lie about percentages

Year-to-date roll-ups are where a second, quieter error creeps in. Some of your ten KPIs are counts - near misses reported, incidents, inspections completed - and a year-to-date total is exactly what you want: add up twelve months of counts and you have the year's count. But some KPIs are rates or percentages - the incident frequency rate itself, training compliance, actions closed on time - and summing twelve monthly percentages does not give you a meaningful annual figure. It gives you a number with no real-world interpretation, usually well north of 100 percent, that someone will still read out in a meeting as if it means something.

A dashboard has to know, KPI by KPI, whether year-to-date means add them up or average them, and that choice, like the colour direction, is a property of the metric, set once, not a formula copied blindly down a column.

A starter set a small site can actually keep up

The temptation with any KPI dashboard is to track everything measurable. Resist it. A metric nobody updates by the third month is worse than no metric at all, because it sits there looking authoritative while quietly lying. Ten is about the ceiling for a site running this by hand:

• Recordable incidents - the core lagging count, whatever your local definition of recordable is.
• Lost-time incidents - the subset serious enough to keep someone off work.
• Days lost - severity, not just frequency.
• First-aid cases - the near-injuries below the recordable threshold, worth watching for creep.
• Incident frequency rate - the incidents, scaled to hours worked, on whichever base you have chosen.
• Near misses reported - the leading indicator that most needs its colour the right way round.
• Inspections completed - against the schedule, not against a vague sense of doing regular walk-rounds.
• Toolbox talks delivered - a proxy for whether the safety conversation is actually happening on the floor.
• Actions closed on time - findings that turn into fixes, not findings that turn into a spreadsheet nobody revisits.
• Training compliance - the percentage of required training that is current, not overdue.

Four lagging, six leading - deliberately weighted toward the numbers you can still influence this month. Keep a days-since-last-incident counter if the visual helps, but let it sit alongside the other nine rather than standing in for them, and do not let a good run on that one counter excuse skipping the rest of the dashboard.