Benchmarking safety performance across multiple sites

If you run safety across multiple sites, you've almost certainly sat in a meeting where someone compared Lost Time Injury Frequency Rate (LTIFR) across locations and drew conclusions about which site is "doing well" and which one "needs attention." On the surface, it seems like a reasonable exercise. In practice, it's one of the most misleading things you can do with safety data.

The problem isn't that benchmarking is a bad idea. Benchmarking is essential for multi-site organisations. It helps you allocate resources, identify where to focus improvement efforts, share best practices, and hold site leadership accountable. The problem is that most organisations benchmark using the wrong metrics, and then make confident decisions based on incomplete information.

Here's how to benchmark safety performance across sites in a way that actually tells you what's happening and why.

Why lagging indicators mislead in multi-site comparisons

Consider a common scenario. Site A has 1,000 workers and an LTIFR of 1.6. Site B has 250 workers and an LTIFR of 2.3. At the quarterly review, leadership concludes that Site A has a stronger safety programme.

But that conclusion might be completely wrong. With a smaller workforce, Site B's rate is far more volatile. A single additional injury could swing the number dramatically. Site A might have a higher absolute number of near misses that simply haven't (yet) resulted in recordable injuries. Site A might also have stronger reporting suppression, where workers are less willing to report injuries because the culture ties bonuses or performance reviews to incident rates.

Lagging indicators like LTIFR and TRIFR have three specific problems when used for site-to-site comparison. They're statistically unreliable at individual site level (small sample sizes create meaningless volatility). They're backward-looking (by the time a rate changes, the cultural and operational conditions that caused it have been in place for months). And they can be gamed (consciously or unconsciously, through reporting suppression or reclassification of injuries).

None of this means you should stop tracking lagging indicators. They have regulatory, insurance, and historical value. But they should not be the primary basis for benchmarking safety performance across sites.

The metrics that actually enable comparison

Meaningful multi-site benchmarking requires leading indicators that are continuous, objective, and comparable across locations regardless of workforce size. Here's what to measure.

Event density per monitored zone. When computer vision AI is deployed across multiple sites, every site generates the same type of data: safety events (near misses, exclusion zone breaches, speed violations, pedestrian-vehicle interactions) detected by connected cameras, tagged with timestamps and locations. Normalising this data by monitored zone gives you a directly comparable metric: events per zone per week, or events per camera per day. This metric isn't influenced by workforce size, reporting culture, or definitional inconsistencies between sites.

Event density trend (rate of improvement). Even more useful than the absolute event count is the trend. Is Site A's event density declining faster than Site B's? A site with higher absolute event density but a steeper improvement trajectory may actually have a stronger safety programme than a site with lower density that has plateaued. The trend tells you which sites are actively improving and which are standing still.

Heatmap concentration ratio. Within each site, what percentage of total events are concentrated in the top three hotspot zones? A high concentration ratio means risk is localised and addressable. A dispersed pattern (events spread evenly across many zones) may indicate more systemic issues. Comparing this ratio across sites helps you understand the nature of each site's risk profile, not just its volume.

Coaching frequency and response rate. How often are supervisors at each site running coaching conversations using real event data? And is the coaching translating into measurable changes in event density in the coached zones? Sites with high coaching frequency and declining event density are executing well. Sites with high coaching frequency but flat event density may have a quality problem in their coaching conversations. Sites with low coaching frequency have a cultural or leadership problem that no amount of data will fix on its own.

Time to corrective action. When a safety issue is identified (whether through a detected event, a heatmap hotspot, or a coaching conversation), how long does it take each site to implement a corrective action? This metric reveals operational responsiveness and how seriously safety is prioritised when it competes with production pressure. Compare the average time from event identification to completed corrective action across sites, and the differences in cultural maturity become visible quickly.

Intervention effectiveness. When a site makes an operational change (moves a barrier, adjusts a traffic flow, changes a delivery window), does the event density in the affected zone actually decrease? Tracking intervention effectiveness across sites tells you which locations are making data-driven changes that work and which are making changes that don't move the needle.

Building a cross-site benchmarking framework

The practical challenge of multi-site benchmarking isn't choosing the right metrics. It's ensuring the data is consistent and comparable. Here's how to set that up.

Standardise what you detect. All sites should be monitoring the same core event types: pedestrian-vehicle proximity, exclusion zone breaches, vehicle speed violations, and any other detections relevant to your operation. With computer vision AI, the detection algorithms are consistent across every connected camera, regardless of location. This eliminates the definitional inconsistencies that plague manual reporting systems (where one site's "near miss" is another site's "non-event").

Normalise for scale. Comparing raw event counts between a 50-camera site and a 12-camera site is meaningless. Normalise by the number of monitored zones, cameras, or monitored square metres. This gives you a rate that's comparable regardless of facility size.

Use the same reporting cadence. If one site reviews data weekly and another reviews monthly, their response times and improvement trajectories will look different for reasons that have nothing to do with safety performance. Standardise the review rhythm: daily event triage, weekly coaching reviews, monthly heatmap analysis, quarterly trend reviews. The cadence itself becomes a benchmarkable metric.

Create a shared dashboard. A single platform where regional and national safety leaders can view all sites side by side, filtering by event type, time period, and zone, is the infrastructure that makes benchmarking operationally useful rather than a quarterly reporting exercise. The data should be accessible in near real time, not aggregated into a slide deck that arrives three weeks after the period ends.

Turning benchmarks into action

Benchmarking only creates value if it drives decisions. Here are the three most productive ways to use cross-site comparisons.

Prioritise resource allocation. If Site C has twice the event density of any other site and a flat improvement trend, that's where your next investment in coaching support, operational redesign, or leadership development should go. The benchmarking data makes the case objectively, removing the politics and subjectivity that often drive resource allocation in multi-site organisations.

Identify and share best practices. If Site D achieved a 40% reduction in pedestrian-vehicle events after redesigning its traffic flow at a specific intersection, the heatmap data tells you exactly what changed and what the impact was. That's a specific, replicable practice that can be adapted for similar zones at other sites. The benchmarking framework surfaces these wins and makes them transferable, rather than leaving them trapped in one site manager's experience.

Create constructive accountability. When every site leader sees the same dashboard and knows their peers can see it too, benchmarking creates a natural accountability loop. This works best when the emphasis is on improvement trends (how fast are you getting better?) rather than absolute rankings (who has the lowest number?). Sites that are improving quickly should be recognised and studied. Sites that are plateauing should receive support and attention, not just scrutiny.

The common trap: ranking without context

The biggest risk in multi-site benchmarking is creating a league table that rewards low numbers and punishes high ones without understanding why the numbers differ.

Two sites can have very different event densities for reasons that have nothing to do with their safety programmes. A site with a complex layout, high traffic volumes, and mixed vehicle-pedestrian flows will naturally generate more events than a site with a simple layout and separated traffic. A recently deployed site will have a higher baseline than a site that's been running the platform for 12 months.

Context matters. The most useful benchmarking frameworks compare each site primarily against its own trajectory (is it improving?), then secondarily against similar sites (how does it compare to peers with similar operational profiles?), and only then against the portfolio average. This layered approach prevents the simplistic "best to worst" ranking that drives gaming, discouragement, and reporting suppression.

What your board actually needs to see

Multi-site benchmarking data feeds directly into the board-level safety reporting that leadership teams need. But boards don't need to see 47 metrics across 15 sites. They need a clear, high-level view that answers three questions: are we improving overall, where are the risks, and are our investments working?

A quarterly board report built on leading indicator benchmarking might include the portfolio-wide event density trend (the big-picture trajectory), the top three improving sites and what's driving their improvement, the sites that need attention and the planned interventions, and the ROI of safety investments measured through risk reduction percentages and operational efficiency gains.

This is a fundamentally different conversation from "our LTIFR was 1.4 this quarter, down from 1.6." It's specific, forward-looking, and connected to the actions that are driving improvement rather than the outcomes that have already happened.

The advantage of consistent data

The organisations that benchmark most effectively are the ones with a consistent data platform across all their sites. When every facility generates the same type of event data through the same computer vision AI system, the comparability problem disappears. You're not reconciling different reporting definitions, different observation methodologies, or different levels of reporting willingness. The data is objective, continuous, and structurally identical from site to site.

For organisations with 10, 20, or 50+ sites, this consistency is the difference between a benchmarking exercise that produces insights and one that produces noise. Many of inviol's 120+ customers operate across far more sites than they currently monitor, and the expansion from one site to many is often driven precisely by the demand for comparable, cross-site safety data.

If you're ready to see what cross-site benchmarking looks like with continuous, consistent leading indicator data, book a demo and we'll walk you through it.

Frequently Asked Questions

Why is LTIFR a poor metric for comparing safety across sites?

LTIFR (Lost Time Injury Frequency Rate) is statistically unreliable at individual site level because small workforces produce volatile rates where a single injury can swing the number dramatically. It's also backward-looking (reflecting outcomes that have already occurred), and it can be influenced by reporting culture rather than actual safety performance. Sites with strong reporting suppression may appear safer on LTIFR while having higher underlying risk.

What leading indicators should I use for multi-site benchmarking?

The most effective leading indicators for cross-site comparison include event density per monitored zone (normalised for scale), event density trends (rate of improvement over time), heatmap concentration ratio (how localised the risk is), coaching frequency and response rate, time to corrective action, and intervention effectiveness (whether operational changes actually reduced event density in the target zone).

How do I make safety data comparable across different-sized sites?

Normalise metrics by the number of monitored zones, connected cameras, or monitored area rather than by workforce size. Computer vision AI provides consistent detection across every site, eliminating the definitional inconsistencies that plague manual reporting. Standardise the review cadence (daily, weekly, monthly, quarterly) so that differences in responsiveness reflect cultural maturity rather than reporting rhythm.

How do I avoid creating a league table that drives gaming?

Compare each site primarily against its own trajectory (is it improving?), then against similar sites with comparable operational profiles, and only then against the portfolio average. Emphasise improvement trends rather than absolute rankings. Recognise sites that are improving quickly and provide support to sites that are plateauing, rather than simply rewarding low numbers and punishing high ones.

What should a board-level safety benchmarking report include?

A quarterly board report should answer three questions: are we improving overall (portfolio-wide event density trend), where are the risks (sites needing attention and planned interventions), and are our investments working (risk reduction percentages and operational efficiency gains). This provides a forward-looking, action-oriented view rather than a retrospective summary of injury statistics.