Behaviour-based safety (BBS): what it gets right and what it misses

Behaviour-based safety has been one of the most influential ideas in workplace safety for the past four decades. It's also one of the most divisive. Depending on who you ask, BBS is either a proven methodology that has saved countless lives, or an outdated framework that blames workers for systemic failures.

The truth, as usual, sits somewhere in between. BBS gets several important things right, and those things deserve acknowledgment. But it also has structural limitations that become increasingly clear as we gain access to better data and more continuous ways of understanding workplace risk.

Here's an honest assessment of where BBS has earned its place, where it falls short, and what the next evolution of safety thinking looks like.

The origins: Heinrich and the 88%

BBS traces its intellectual roots to Herbert Heinrich, who published "Industrial Accident Prevention: A Scientific Approach" in 1931. Heinrich's widely cited claim was that approximately 88% of industrial accidents were caused by unsafe acts on the part of workers. At the time, the prevailing theory was that some workers were simply "accident prone," which essentially treated workplace injuries as unavoidable. Heinrich's reframing was progressive: if unsafe behaviour causes accidents, then changing behaviour can prevent them. That was a meaningful shift.

From that foundation, researchers and practitioners developed what we now call behaviour-based safety. The core methodology involves structured peer observation (workers watching colleagues perform tasks), recording safe and unsafe behaviours, providing immediate positive feedback, and using the aggregated data to identify patterns that need to be addressed through training, process changes, or environmental adjustments.

The approach gained significant traction from the 1980s onwards. Research published in the Journal of Organizational Behavior Management, drawing on data from 88 international organisations and over 1.3 million observational data points, has shown that well-implemented BBS programmes can achieve meaningful reductions in injuries. A meta-analysis published in Theoretical Issues in Ergonomics Science found a statistically significant reduction in workplace accidents across the studies reviewed, with a pooled reduction of approximately 39%.

What BBS gets right

BBS deserves credit for several contributions that have genuinely improved how organisations approach safety.

It made behaviour observable and measurable. Before BBS, safety management was largely reactive: you investigated after something went wrong. BBS introduced the idea that you could systematically observe and quantify safe and unsafe behaviours before an incident occurred. That shift from reactive to proactive was significant, and it laid the groundwork for the leading indicator thinking that the best safety programmes now rely on.

It put frontline workers at the centre of the process. Traditional safety management was top-down: policies written in an office, procedures handed to workers, compliance audited from above. BBS flipped this by making peer observation and feedback a core activity. When implemented well, this creates genuine ownership. Workers aren't just following rules. They're actively participating in identifying and reducing risk. That participatory model remains one of the most powerful levers for building a strong safety culture.

It emphasised positive reinforcement over punishment. The best BBS programmes are built on recognising safe behaviour, not penalising unsafe behaviour. This matters more than it might seem. Punishment-based approaches drive unsafe behaviours underground. Workers stop reporting near misses because they're afraid of consequences. Positive reinforcement encourages transparency, which is exactly the cultural condition where safety improves. This principle aligns directly with a coaching-first approach to safety: using data to start constructive conversations, not to catch people out.

It introduced data-driven decision-making to safety. BBS observation checklists generate data. That data, when analysed properly, reveals patterns: which behaviours are most frequently at risk, which areas have the highest concentration of unsafe acts, and which interventions are working. This was an early form of safety analytics, and it paved the way for more sophisticated data-driven safety approaches.

Where BBS falls short

For everything BBS gets right, it has structural limitations that become harder to ignore as workplaces become more complex and better tools become available.

It relies on human observation, which can't scale. A BBS programme can only observe what a human observer happens to see, during the time they happen to be watching. In a busy warehouse or distribution centre operating across multiple shifts, the observation window is a tiny fraction of the total operational hours. The near miss that happens at 3am when no observer is present, the exclusion zone breach that occurs during a five-minute gap between observation rounds: these events are invisible to a BBS programme. They happen, they create risk, and nobody ever knows.

This is the fundamental constraint. BBS was designed for a world where human observation was the best available tool. It's no longer the best available tool.

It struggles with systemic causes. BBS focuses on individual behaviour, which means the interventions it generates tend to be behavioural: retraining, feedback, procedural reminders. But many safety events aren't primarily caused by individual behaviour. They're caused by how the operation is designed: the layout of the facility, the timing of deliveries, the positioning of pedestrian walkways relative to vehicle traffic, the staffing levels during each shift.

When a forklift and a pedestrian have a near miss at an intersection, a BBS observation might record it as an unsafe act by the operator. But the root cause might be that the intersection forces blind-reverse manoeuvres because of how the racking is configured. No amount of behavioural observation will fix a design problem. You need data that shows the pattern (dozens of near misses at the same location, concentrated at the same time of day) to make the case for redesigning the operation. That's what a safety heatmap provides, and it's data that BBS was never designed to generate.

It can inadvertently create a blame culture. This is the criticism that BBS practitioners hate to hear, but it's well-documented. When safety is framed as a function of individual behaviour, the implication (intended or not) is that incidents are caused by individuals making poor choices. A 2008 report from the US House Education and Labor Committee raised concerns that BBS programmes tied to incentive schemes were fuelling underreporting of injuries, because workers feared losing rewards or being blamed for incidents.

The best BBS programmes explicitly avoid blame and focus on positive reinforcement. But the structural framing of the methodology (observe individual behaviour, provide feedback on individual behaviour, measure changes in individual behaviour) makes it difficult to completely separate from the perception that safety failures are personal failures.

The data it generates is limited. BBS observation data is valuable, but it's inherently sampled: a human observer watching a specific area for a specific period. It can't capture events that occur outside the observation window, events that happen too quickly for a human to notice, or the spatial and temporal patterns that only emerge when you aggregate thousands of events over weeks and months. The Cambridge University study of DEKRA's BBS data acknowledged that observation frequency matters, but even the most diligent observation programme captures a fraction of what actually occurs.

The observation gap

The limitations of BBS all trace back to the same root constraint: the dependence on human observation.

Humans are excellent observers in many contexts. They understand nuance, they can interpret intent, and they can have a conversation with the person they've just observed. These are genuine advantages that technology doesn't replicate.

But humans can't observe continuously. They can't be in multiple locations simultaneously. They can't quantify risk density across an entire facility over a rolling 30-day period. They can't detect the near miss that happens behind them while they're watching something else. They can't filter events by time of day, shift pattern, or event type to reveal the systemic patterns that drive risk.

Computer vision AI can do all of these things. It processes video feeds from every connected camera, around the clock, classifying and recording every safety event with a location, a timestamp, and a category. It doesn't get fatigued, it doesn't forget to fill in the observation form, and it doesn't have a shift end. The result is a data set that is orders of magnitude more complete than anything a BBS programme can generate through human observation alone.

This isn't about replacing the human element. It's about closing the gap between what humans can observe and what actually happens.

What comes next: BBS meets continuous data

The most productive way to think about BBS and computer vision AI is not as competitors but as complementary layers in a mature safety programme.

BBS contributes the cultural foundation: peer engagement, positive reinforcement, the principle that safety is everyone's responsibility. These are cultural values that no technology can install. They require leadership, trust, and consistent practice.

Computer vision AI contributes the data foundation: continuous, facility-wide visibility into safety events that no human observation programme can match. It captures the full picture (every near miss, every exclusion zone breach, every speed violation) and presents it in a format that enables both coaching conversations and systemic design improvements.

The intersection of these two approaches is where the real value lies. A supervisor who receives a coaching clip of a near miss from yesterday's night shift can use that clip in a toolbox talk the next morning, applying the same positive, constructive feedback principles that BBS pioneered, but now grounded in objective, timestamped evidence rather than a hand-written observation note from two weeks ago.

An EHS manager who sees a heatmap showing that 70% of pedestrian-vehicle interactions cluster at a single intersection during the 6-7am delivery window can make the case for an operational redesign, addressing the systemic root cause that BBS observation alone might never have surfaced.

A site leader who can compare leading indicator trends across multiple facilities can identify which sites are improving fastest and share the practices that are driving that improvement, scaling the cultural elements of BBS with the analytical power of continuous data.

Giving BBS what it always needed

BBS was built on a sound insight: that understanding and influencing behaviour is central to preventing workplace injuries. The research supports this. The challenge was always that the methodology depended on a data collection method (human observation) that couldn't keep pace with the complexity and scale of modern operations.

Computer vision AI doesn't invalidate BBS. It gives BBS what it always needed: complete, continuous, objective data about what's actually happening across your facility, 24 hours a day. It closes the observation gap, surfaces the systemic patterns, and provides the evidence that makes coaching conversations specific, timely, and grounded in reality.

The organisations that will lead safety performance in the coming years won't be choosing between BBS and technology. They'll be combining the cultural strengths of behaviour-based approaches with the data capabilities of computer vision AI, and in doing so, achieving results that neither approach could deliver alone.

If you're ready to see what continuous safety data looks like alongside your existing safety programme, book a demo and we'll show you.

Frequently Asked Questions

What is behaviour-based safety (BBS)?

Behaviour-based safety is a methodology that focuses on observing, measuring, and influencing worker behaviour to prevent workplace injuries. It typically involves structured peer observations, recording safe and unsafe behaviours against a checklist, providing immediate positive feedback, and analysing the aggregated data to identify patterns that need to be addressed. BBS has been widely adopted since the 1980s and is rooted in the work of Herbert Heinrich and the ABC (Antecedent, Behaviour, Consequence) model of behavioural psychology.

Does BBS actually work?

Research shows that well-implemented BBS programmes can achieve statistically significant reductions in workplace injuries. A meta-analysis published in Theoretical Issues in Ergonomics Science found a pooled reduction of approximately 39% across the studies reviewed. A 2022 Cambridge University study of 88 organisations and over 1.3 million data points confirmed that BBS can produce meaningful injury reductions and positive cultural change. However, results depend heavily on implementation quality, and poorly executed programmes can lead to underreporting and blame culture.

What are the main criticisms of behaviour-based safety?

The most common criticisms are that BBS relies on human observation (which can only capture a small fraction of what actually happens), that it focuses on individual behaviour rather than systemic causes (such as facility layout or scheduling problems), that it can inadvertently create a blame culture (even when that's not the intent), and that incentive-based BBS programmes can fuel underreporting of injuries and near misses.

Can BBS and AI safety technology work together?

Yes, and the combination is more effective than either approach alone. BBS provides the cultural foundation: peer engagement, positive reinforcement, and the principle that safety is everyone's responsibility. Computer vision AI provides the data foundation: continuous, facility-wide detection of safety events that human observation can't capture. Together, they enable coaching conversations grounded in objective evidence, identification of systemic design issues, and measurement of intervention effectiveness across entire facilities.

What is the observation gap in BBS?

The observation gap refers to the fundamental limitation that human observers can only watch a small fraction of a facility's operations at any given time. Events that occur outside observation windows, during unstaffed shifts, or in areas an observer isn't currently watching go unrecorded. Computer vision AI closes this gap by processing video feeds from every connected camera continuously, capturing every detectable safety event with a timestamp and location, regardless of whether a human observer is present.