Safety KPIs are the metrics a plant uses to measure how safe its operations are and whether they are getting safer. The core set pairs lagging outcome metrics, TRIR, DART, lost-time and severity rates, with leading activity metrics, near-miss frequency, corrective-action closure, and audit scores.

Most plants track exactly one safety number, TRIR, report it up the chain, and manage the whole program by trying to make it smaller. That is a mistake, and not a subtle one. A single lagging number can be pushed down by getting safer or by getting quieter, and from the boardroom the two look identical. This post lays out the core safety KPIs, the formulas behind them, what good actually looks like against national benchmarks, and how to build a balanced scorecard that cannot be gamed by simply hiding injuries. It is educational, not legal advice.

What are the core safety KPIs?

A working set has both kinds: outcomes that count what already happened, and activities that predict what will. Here is the core group, with how each is figured.

KPITypeHow it's figured
TRIR (total recordable incident rate)Lagging(recordable cases × 200,000) ÷ hours worked
DART rateLagging(cases with days away, restricted, or transferred × 200,000) ÷ hours worked
Lost-time case rate (LTIFR)Lagging(lost-time cases × 200,000) ÷ hours worked
Severity rateLagging(lost workdays × 200,000) ÷ hours worked
Near-miss frequencyLeadingnear misses reported per period (or per 100 workers)
Corrective-action closureLeading% of safety actions closed on time
Audit / inspection scoreLeadingfindings per audit, and % closed on time
Training completionLeading% of required safety training current

What is TRIR, and how is it calculated?

TRIR is the number of OSHA-recordable injuries and illnesses per 100 full-time workers per year. The formula multiplies your recordable cases by 200,000 and divides by the hours your people actually worked. The 200,000 is not magic: it is 100 employees working 40 hours a week for 50 weeks, the standard denominator that lets a 30-person shop and a 3,000-person plant compare on the same scale. Our guide to calculating TRIR walks the arithmetic and the edge cases; the thing to remember here is that TRIR is a rate, not a count, so it does not punish you for growing.

How the 200,000-hour incident rate worksEvery OSHA rate shares one shaperecordablecases×200,000the yardstick÷total hours actually workedacross all employees200,000 = 100 workers × 40 hrs × 50 weeksResult = cases per 100 full-time workers per year
TRIR, DART, lost-time, and severity rates all share this shape: count the relevant cases, scale by 200,000, divide by real hours. Only the numerator changes.

How does DART differ from TRIR?

DART counts only the serious cases, so it is always a subset of TRIR. Where TRIR counts every recordable, including cases resolved with medical treatment beyond first aid, DART counts only cases that cost Days Away, Restricted work, or job Transfer. Because every DART case is also a recordable case, a plant's DART rate is always equal to or lower than its TRIR. Watching both tells you something one cannot: if TRIR is flat but DART is climbing, your injuries are getting more severe even though they are not getting more frequent, and that is a different problem with a different fix.

TRIR contains DART contains lost-time casesThe metrics nest inside each otherTRIRall recordablesDARTdays away/restr/transferlosttimeDART is always ≤ TRIRbecause every DART caseis also a recordable caseRising DART, flat TRIR =fewer but worse injuries
DART cases are a subset of recordable cases, and lost-time cases are a subset of DART. Tracking the layers separately tells you whether injuries are getting more frequent, more severe, or both.

What about severity and lost-time rates?

Severity rate measures how bad the injuries are, not just how many. It scales lost workdays, rather than case counts, by the same 200,000 over hours worked, so it captures the difference between a plant with ten one-day strains and a plant with one crushing that costs 300 days. The lost-time case rate (LTIFR) sits between DART and severity, counting cases that resulted in time away from work. Together they answer the question a raw case count cannot: are we getting hurt less often, or just less badly, or neither?

A worked example makes the point. Suppose a 250-person plant runs about 500,000 hours in a year and books 6 recordable cases, 2 of which cost days away or restriction. Its TRIR is 6 times 200,000 divided by 500,000, or 2.4, and its DART rate is 2 times 200,000 divided by 500,000, or 0.8. Next year the same plant records 6 recordables again, but now 4 of them are days-away cases and total lost workdays jump. TRIR is flat at 2.4, DART has doubled to 1.6, and severity has climbed hard. Anyone watching only TRIR would report "no change" and miss that the plant's injuries got materially worse. That gap is the entire argument for tracking the family, not the headline.

Why lagging metrics alone will fool you

Because the fastest way to improve a lagging number is to stop generating it. Every lagging safety KPI, TRIR, DART, severity, shares a fatal weakness as a management tool: it can be driven down by getting safer or by suppressing reports, and the number itself does not say which. A plant that pressures people out of reporting will post a beautiful TRIR right up until the injury nobody could hide. This is exactly why the difference between leading and lagging indicators matters, and why a scorecard built only on outcomes is a scorecard you can cheat without meaning to.

What leading safety KPIs should you track?

The activities that predict fewer injuries, because those you can grow honestly. Four earn their place on almost any plant's scorecard: near-miss frequency (are people surfacing hazards before they bite?), corrective-action closure (are the hazards we find actually getting fixed on time?), audit and inspection results (what are we finding, and are we closing it?), and training completion (is the required training current?). Add proactive analysis such as job safety analysis currency for high-risk jobs, and you have a leading side that moves before anyone gets hurt.

The reason leading metrics are worth the trouble is that they are actionable this week. You cannot do anything directly about last quarter's TRIR, it already happened, but you can raise near-miss reporting, close overdue corrective actions, and finish the training that is behind, and every one of those pushes next quarter's TRIR down. Leading KPIs give a supervisor something to manage on a Monday morning; lagging KPIs only give them a grade at the end of term.

A balanced safety scorecardManage on both sidesLAGGING (outcomes)LEADING (activities)TRIRDART rateseverity ratelost-time case ratewhat already happenednear-miss frequencycorrective-action closureaudit findings + closuretraining completionwhat predicts the outcome
A scorecard that reports only the left panel can be gamed by silence. The right panel is what you actually manage day to day; the left is what it earns you.

How do you build a balanced safety scorecard?

Pick a short set, define each precisely, and put leading and lagging side by side so nobody can hide behind one number.

  1. Start with the two or three lagging rates you will report externally, usually TRIR and DART, because customers, insurers, and prequalification will ask for them regardless.
  2. Add three or four leading metrics you can actually influence this month: near-miss frequency, corrective-action closure, audit findings closed on time, training completion.
  3. Define every metric precisely, what counts, over what hours, on what cadence, so the number means the same thing each quarter and across sites.
  4. Set targets against a benchmark, not against zero alone. "Below our industry's BLS average and falling" beats a target of zero that quietly rewards non-reporting.
  5. Report leading and lagging together. A TRIR shown next to a rising near-miss count tells a story a lone TRIR cannot.
  6. Review on a fixed cadence, ideally through the safety committee and act on the leading metrics between reviews.
  7. Re-check for gaming. If a lagging number improves while reporting drops, treat it as a warning, not a win.

What does good look like?

Benchmark against your own industry, using the primary sources:

A number to distrust: a perfect zero TRIR at a plant whose near-miss log is also empty. Real safety is loud on the leading side and quiet on the lagging side; silence on both usually means the reporting is broken, not the hazards gone. The best-run plants are, counterintuitively, noisy: they generate lots of near-miss reports and lots of closed corrective actions, and their injury rates fall because those hazards get caught upstream. A quiet plant is not necessarily a safe one; it may just be a plant that has stopped listening.

The practical problem with safety KPIs is not the math; it is that the inputs live in five places. Hours come from payroll, recordables from an OSHA log, near misses from a paper box, audit findings from a binder, corrective actions from a spreadsheet, and stitching them into a scorecard is a monthly manual chore that is stale by the time it is finished. Harmony connects machines, software, and paperwork into one operational layer with no rip-and-replace: reports, audits, and corrective actions become structured data captured at the station, so leading and lagging KPIs update live instead of once a month. AI search returns cited answers across the underlying records, so a rising DART rate can be traced to the machines and jobs behind it, part of the everyday shape of connected worker technology. CLS started the same way, replacing paper production logging with real-time visibility. Harmony is not a safety-compliance product, but it keeps the scorecard honest and current.