The dashboard tells on us
Walk into any board safety review and the headline number is almost always a recordable rate. There’s a reason: TRIR and LTIFR are easy to define, easy to benchmark, and easy to put a target on. Insurers ask for them. Procurement portals ask for them. Annual reports print them next to revenue. So they win the slot on slide one by default, even when everyone in the room knows they’re looking in the rearview mirror.
And the rearview mirror is genuinely a bad place to drive from. For most sites, recordables are rare enough that one bad month can swing the rate by half. A small contractor with two injuries in a year can look worse than a peer with twelve, just because the denominator is smaller. Researchers have been pointing this out for years — a single year of TRIR data is closer to noise than signal for predicting what happens next.
Why leading indicators stay theoretical
Ask the same safety leader to list their leading indicators and you’ll usually get a confident answer: near‑miss reports, observations, training completion, audit closure, hazard hunts. Good list. The problem is what happens when you try to make a decision off any of those numbers.
Take observations. Most behavior‑based programs end up flooded with the easy stuff — missing hard hats, untidy walkways, a cable nobody coiled. Those are real, but they’re not what kills people. The serious exposures — energy isolation, working at height, mobile equipment, confined space — show up far less often in the data, because they’re harder to see and harder to write up in 30 seconds on a phone. So the dataset skews toward low‑energy, low‑consequence stuff, and the trend line looks healthy right up until something catastrophic happens that the program never had a chance of catching.
Near‑miss reporting has its own version of the same problem. The moment you set a target — “every crew submits one near miss a week” — you don’t get more learning, you get more paperwork. People file the closest thing to a near miss they can find on a Friday afternoon. The number goes up. The quality goes down. Whatever signal was buried in the original reports gets diluted by the ones that exist purely to hit the KPI.
The classification mess underneath
Here’s an underrated reason leading indicators wobble: two safety pros looking at the same event often don’t classify it the same way. Studies on this put inter‑rater agreement somewhere around 65 percent without a strong taxonomy — meaning a third of the time, two reasonable people disagree on what bucket an incident belongs in. If that’s your raw data, your “trends” are partly real and partly an artifact of who happened to triage the report that week.
This is the part executives almost never see. The chart looks crisp. The categories look stable. But underneath, a slow turnover in who’s coding events can shift the numbers as much as a real change in operations would. Imagine running finance off ledgers where every accountant defined “revenue” slightly differently — nobody would tolerate it. We tolerate it in safety because the alternative feels like more work for no obvious win.
What a credible leading-indicator program actually looks like
The teams I’ve seen make this work share a few habits, and none of them are about buying a new dashboard.
First, they pick a small number of indicators that map to their actual top risks. Not eighteen. Three or four. If your worst exposures are line‑of‑fire and energy isolation, then your leading indicators are about line‑of‑fire and energy isolation — control verification rates on those specific activities, not a generic observation count. The test is simple: if this number moved, would I change what I do tomorrow? If not, it’s decoration.
Second, they measure whether controls are present and working, not whether activities happened. “Permits issued” tells you nothing. “Permits where the energy isolation was independently verified by someone other than the issuer” tells you something. Same for training: completion rates are vanity; competency checks on the floor are the real indicator.
Third, they treat near misses as a quality problem, not a volume problem. One detailed, well‑investigated near miss involving a high‑energy hazard is worth more than 50 reports of someone almost tripping. So they audit a sample of reports for depth and follow‑through, and they retire the “reports per crew per week” KPI before it rots the data.
Fourth, they keep the lagging numbers in the deck — you can’t not — but they demote them. Recordables stay as accountability metrics, not steering metrics. The conversation in the room is about what the leading indicators say, with the lagging numbers as a reality check rather than the agenda.
The honest version of the pitch
None of this is glamorous. Leading indicators are messier than recordables, harder to benchmark across companies, and easier to game if you’re not paying attention. That’s exactly why so many programs slide back to TRIR‑on‑slide‑one: it’s a clean number, and clean numbers win arguments even when they don’t reflect risk.
The shift worth making isn’t “replace lagging with leading.” It’s narrower than that. Pick the two or three exposures that could actually hurt or kill someone at your sites. Build a small set of leading indicators that tell you whether the controls for those specific exposures are real and verified this week, not last quarter. Hold the line on quality over quantity, even when the volume chart looks less impressive. And put those numbers in front of the lagging ones in the deck — not because the lagging ones don’t matter, but because by the time they move, the decision you needed to make was already three months ago.