Leading vs lagging safety indicators: how AI changes the game

If you've spent any time in EHS, you've heard the terms "leading indicators" and "lagging indicators" more times than you can count. They come up in every safety conference, every board report, and every conversation about how to measure whether your safety programme is actually working.

But here's the thing: most organisations still rely almost entirely on lagging indicators — even though everyone agrees that leading indicators are more valuable. And the reason isn't a lack of desire. It's a lack of practical data.

That's changing. Computer vision AI is solving the leading indicator problem in a way that wasn't possible even a few years ago. Let's start with the basics and work our way to why that matters.

What are lagging indicators?

Lagging indicators measure things that have already happened. In safety, the most common examples are injury rates (like TRIR — Total Recordable Incident Rate), lost-time injuries (LTIs), workers' compensation claims, and days away from work due to injury.

These metrics are essential for compliance and reporting. Regulators want to see them. Insurers use them to set premiums. Leadership teams use them to benchmark against industry averages. They're a necessary part of any safety programme.

But lagging indicators have a fundamental limitation: they only tell you something went wrong after someone got hurt. As OSHA puts it, lagging indicators measure "the occurrence and frequency of events that occurred in the past." They're a rear-view mirror — useful for understanding where you've been, but not for seeing what's ahead.

There's also a perverse incentive problem. When organisations are measured primarily on injury rates, there's pressure (conscious or not) to underreport. A low TRIR might mean your site is genuinely safe — or it might mean people aren't reporting injuries. It's very difficult to tell the difference from the number alone.

What are leading indicators?

Leading indicators are proactive measures that predict future safety performance. Instead of counting injuries after they happen, they track the activities, conditions, and behaviours that either prevent or contribute to incidents.

OSHA defines leading indicators as "proactive, preventive, and predictive measures that provide information about the effective performance of your health and safety activities." The Campbell Institute at the National Safety Council builds on this, defining them as measures that "can drive the identification and elimination or control of risks in the workplace."

Traditional examples of leading indicators include the number of safety training sessions completed, the frequency of safety audits or inspections, near-miss reports submitted, safety observation participation rates, and the time taken to close corrective actions.

The evidence for their value is compelling. Research cited by the Campbell Institute found that organisations with established leading indicator programmes saw an average 77% reduction in incident rates over three to twelve years. That's a remarkable result — and it makes intuitive sense. If you measure and improve prevention activities, you get fewer injuries.

So why doesn't everyone use them?

If leading indicators are so effective, why aren't they the primary metric in every safety programme? The answer is practical: leading indicators have traditionally been hard to collect, hard to standardise, and hard to trust.

Consider near-miss reporting — widely regarded as one of the most valuable leading indicators. Near misses sit at the base of the safety pyramid and represent the early warning signals that, if addressed, prevent serious injuries. But as we've explored in a previous post, near-miss reporting is deeply unreliable. Workers underreport for all kinds of reasons: fear of blame, unclear definitions, inconvenient reporting processes, or simply not recognising the event as significant.

Other leading indicators suffer from similar problems. Safety observation counts can be gamed (more observations logged doesn't mean better observations). Audit completion rates tell you an audit happened, not whether it was any good. Training records confirm attendance, not competence.

The result is that many leading indicator programmes become compliance exercises rather than genuine safety tools. The data exists, but it doesn't reliably predict anything — and EHS teams know it.

How computer vision AI changes the leading indicator equation

This is where the game changes. Computer vision AI generates leading indicator data automatically, continuously, and objectively — without relying on human reporting or manual observation.

When a platform like inviol analyses your CCTV feeds, it captures safety events that are, by definition, leading indicators: near misses between people and vehicles, exclusion zone breaches, speed violations, and risky interactions. These events haven't resulted in injury (yet). They're the early warning signals that tell you where your next incident is most likely to come from.

The difference from traditional leading indicators is threefold:

Scale. Instead of capturing the handful of near misses that get manually reported each month, computer vision AI captures every detectable event across every camera, every shift, every day. The dataset isn't a sample — it's close to a census of your site's safety-relevant interactions.

Objectivity. The AI applies the same criteria to every event, every time. There's no variability based on who's observing, how busy the shift is, or whether someone decides the event was "worth reporting." This consistency makes the data genuinely comparable across time, shifts, zones, and sites.

Automation. Nobody has to fill out a form, remember to submit a report, or walk a specific area at a specific time. The data is generated passively from cameras that are already running. This eliminates the single biggest barrier to leading indicator adoption: the burden of manual data collection.

What this looks like in practice

Let's say you're an EHS manager at a distribution centre with 40 CCTV cameras. Before computer vision AI, your leading indicator data might look like this: 12 near-miss reports submitted last month (mostly from the same three diligent reporters), 8 safety observations completed, and 4 audits conducted. You know there's more happening on the floor, but you can't prove it, and you can't quantify it.

With computer vision AI, your data looks fundamentally different — and you don't even need to connect every camera. inviol typically only needs to be implemented on a selection of your cameras, focused on the highest-risk areas: forklift traffic lanes, pedestrian intersections, loading docks, and exclusion zones. From just 10–15 strategically chosen cameras, the system might detect 347 safety events last month: 182 pedestrian-vehicle interactions, 89 exclusion zone breaches, 76 speed events. You can see that Zone C near the loading dock has 3x more events than any other area. You can see that the Thursday night shift has consistently higher risk than other shifts. You can see that events in Aisle 7 have been trending upward for six weeks.

Now you have something actionable. You can direct coaching resources to the areas and shifts where they'll have the most impact. You can target your safety walks to the zones that need attention. And you can track whether your interventions are actually working, because the same system that detected the original events will show you whether they decrease after you act.

This is the promise of leading indicators finally being delivered at scale.

The best programmes use both

To be clear, this isn't about abandoning lagging indicators. TRIR, LTIs, and workers' compensation data still matter. They're the ultimate outcome measures — the proof of whether your programme is keeping people safe.

But lagging indicators alone are like checking your bank balance without looking at your spending. They tell you the outcome but not the process. Leading indicators — especially the kind that computer vision AI generates — give you the process data you need to steer your programme proactively.

The ideal safety measurement framework uses lagging indicators to confirm outcomes and leading indicators to drive daily action. Computer vision AI makes the "daily action" side of that equation reliable for the first time.

Across inviol's customer base, this combination delivers measurable results: an average 67% reduction in safety risk and a 42% reduction in incidents over three years. The Warehouse Group cut incidents by 60% in two months — driven not by lagging indicator targets, but by leading indicator data that showed their team exactly where to focus their coaching.

The shift has already started

The conversation in the EHS world is changing. More organisations are recognising that measuring injuries alone isn't enough — you need to measure the conditions and behaviours that cause them. OSHA, the Campbell Institute, and leading EHS practitioners all advocate for a balanced approach that gives leading indicators the weight they deserve.

Computer vision AI is what makes that practical. It turns the leading indicator aspiration — "we should measure prevention, not just outcomes" — into a daily reality backed by continuous, objective, scalable data.

Your safety programme already measures what went wrong. It's time to start measuring what's about to.

Ready to see leading indicators in action? Book a demo and we'll show you how inviol generates real-time leading indicator data from your existing cameras — giving your team the visibility to prevent incidents, not just report them.

Frequently Asked Questions

What is the difference between leading and lagging safety indicators?

Lagging indicators measure past outcomes — things like injury rates, lost-time incidents, and workers' compensation claims. They tell you what already went wrong. Leading indicators measure proactive prevention activities and conditions — like near-miss frequency, safety observation rates, and risk trends — that predict future safety performance. A strong safety programme uses both.

Why are leading indicators better for preventing workplace injuries?

Because they measure risk *before* an injury occurs. Leading indicators like near-miss data, exclusion zone breaches, and unsafe interaction trends give EHS teams the information they need to intervene proactively. Research from the Campbell Institute found that organisations with established leading indicator programmes saw an average 77% reduction in incident rates.

How does computer vision AI generate leading indicator data?

Computer vision AI analyses video feeds from existing CCTV cameras to detect safety-relevant events — near misses, speed violations, exclusion zone breaches, and risky interactions — automatically and continuously. These events are leading indicators by definition: they represent risk that hasn't yet resulted in injury. The data is generated at scale, objectively, and without relying on manual reporting.

What are examples of leading indicators in workplace safety?

Traditional examples include safety training completion rates, audit frequency, near-miss reports, and corrective action closure times. Computer vision AI adds a new category: automatically detected safety events such as pedestrian-vehicle near misses, exclusion zone breaches, and vehicle speed violations — captured continuously across every camera on your site.

Can leading indicators really predict where the next injury will happen?

They can't predict individual incidents with certainty, but they reliably highlight where risk is concentrated. If a particular zone consistently shows high rates of near misses or exclusion zone breaches, the data strongly suggests that zone is more likely to produce an injury if conditions don't change. This allows safety teams to focus resources where they'll have the greatest impact.