Zero harm: the honest conversation about safety goals

I've sat in boardrooms where the CEO proudly declared a zero harm target, and I've stood on warehouse floors where workers rolled their eyes at the very same words on a poster. Both reactions tell you something important about the state of this conversation.

Zero harm is one of the most divisive ideas in workplace safety. For some, it's an essential moral commitment: no worker should be harmed, and anything less than zero as a target is an acceptance of injury. For others, it's a well-intentioned slogan that has done real damage to safety culture, honest reporting, and the ability of organisations to learn from what goes wrong.

This piece isn't about declaring which side is right. It's about having the honest conversation that most organisations avoid.

The case for zero

The philosophical argument for zero harm is hard to argue against. Every worker does deserve to go home in the same condition they arrived. No organisation should set a target that implies an acceptable number of injuries. And as an aspirational commitment, zero harm has motivated genuine investment in safety infrastructure, training, and culture across industries that previously treated injuries as a cost of doing business.

Many organisations have achieved remarkable results under zero harm banners. Bodycote, a global engineering company, launched a comprehensive "House of Safety" framework in 2024 built on zero harm principles, with strategic priorities around leadership, risk awareness, and standardisation. The intent is clearly genuine, and the structure is thoughtful.

The zero harm philosophy has also pushed safety up the corporate governance agenda. In Australia and New Zealand, where officers have personal liability under WHS and HSWA legislation, the language of zero harm has helped board directors take their safety governance obligations seriously. It's easier to get executive attention when the goal is unambiguous.

At its best, zero harm represents a refusal to accept complacency. It says: we will never stop trying to improve. That sentiment is valuable.

The case against

The problems emerge when zero harm moves from aspiration to metric.

Professor Sidney Dekker, one of the most influential voices in safety science, has been one of the sharpest critics. His central argument is that when organisations hold people accountable for achieving zero (an outcome that is statistically improbable in complex, high-risk environments), the predictable result is not fewer injuries but less honest reporting.

Dekker's observation is pointed: when you keep people accountable for low numbers of negatives, that is what they will give you. Organisations find ways to underreport, reclassify, and manage the numbers rather than manage the risk. The famous Deepwater Horizon had a celebrated safety record right up until an explosion killed 11 people in 2010. The numbers looked excellent. The reality was catastrophic.

At the 2025 NSCA National Safety Conference in Sydney, Dekker challenged the audience directly. He described an offshore rig that went seven years without a recorded injury. 126 people on a floating platform. He called it "statistical obliteration" and "theatre," arguing that the absence of recorded incidents doesn't mean the absence of risk.

Research published in the journal Safety Science examined the relationship between zero harm commitments and actual safety outcomes. The findings were equivocal: firms using zero had slightly fewer serious injury and fatality (SIF) incidents overall, but when the data were examined statistically, no significant difference in SIF outcomes between zero and non-zero companies was found. The study concluded that the empirical data does not evidence any clear success from the use of zero in practice.

The AHRQ's PSNet analysis notes that sceptics of zero harm goals have flagged the risk of lowered morale and incentivised underreporting. When the goal feels unattainable, workers can disengage, and reporting systems degrade.

These aren't fringe criticisms. They come from the most respected researchers and institutions in workplace safety.

The real problem: what zero harm does to learning

This is the dimension of the debate that matters most, and the one that gets talked about least.

Safety improvement depends on learning. Learning depends on data. Data depends on honest reporting. And honest reporting depends on a culture where people feel safe admitting that things went wrong.

A zero harm target, taken literally, implies that every incident is a failure. If every incident is a failure, then reporting an incident is reporting a failure. And if the organisation celebrates streaks of zero-incident days (as many do, with signs in reception counting days since the last recordable), then the person who breaks the streak faces social pressure, not just from management but from their own colleagues.

The result is what Dekker calls the erosion of safety culture in the drive for zero. Near misses go unreported. Minor injuries get reclassified. Workers with genuine concerns stay quiet because the system incentivises silence over honesty.

This is the opposite of what a proactive safety culture needs. Proactive cultures depend on a rich flow of near-miss data, honest reporting of events at every severity level, and a coaching response that treats events as learning opportunities rather than failures. If your zero harm target is suppressing that flow, your target is making you less safe, not more.

What actually drives improvement

The organisations that achieve the best safety outcomes, in our experience, don't frame their programmes around the absence of incidents. They frame them around the presence of specific behaviours and capabilities.

They measure leading indicators: near-miss detection rates, coaching session frequency, time to corrective action, worker participation in safety discussions. These metrics tell you whether your system is working before anyone gets hurt.

They invest in coaching, not policing. Research consistently shows that organisations with high employee engagement (built through coaching, trust, and genuine worker involvement) experience dramatically fewer safety incidents than those relying on rules and enforcement alone.

They celebrate learning, not streaks. Instead of counting days since the last incident, they count near misses reported, hazards identified, and coaching sessions completed. These are positive indicators of a healthy safety system, not the absence of failure.

They acknowledge that risk is inherent in complex work. A warehouse with forklifts, pedestrians, loading docks, and shift changeovers will always contain risk. The goal is not to pretend the risk doesn't exist but to manage it intelligently, reduce it continuously, and respond quickly when something goes wrong.

This is closer to Dekker's "Safety Differently" framework, which argues that workers are not the problem to be controlled but the resource to be harnessed. When you trust workers, engage them in solving safety challenges, and give them the data and tools to make better decisions, they become the strongest part of your safety system rather than its weakest link.

Where technology fits in this conversation

Computer vision AI enters this debate at an interesting point. It can serve either philosophy, depending on how it's implemented.

If the technology is deployed as a surveillance system that counts violations and feeds a zero-incident scorecard, it will amplify every problem the critics of zero harm have identified. Workers will see the cameras as tools of punishment. The system will generate numbers that managers chase, and the culture will erode.

If the technology is deployed as a coaching tool that generates learning data, the dynamic is completely different. Near misses detected by AI don't carry the stigma of self-reporting a failure, because the system captures them automatically and anonymously. Faces are blurred. The data isn't about who made a mistake. It's about where risk concentrates and what the team can do about it.

This is inviol's design philosophy. The platform is built around coaching, not policing. When the AI detects a safety event, the response is a team conversation, not a disciplinary process. The heatmap data shows patterns, not perpetrators. The coaching platform logs sessions and tracks whether interventions are working over time.

The technology doesn't need a zero harm target to be effective. It needs a learning culture. And it can help build that culture by providing the continuous, objective, anonymised data that makes honest safety conversations possible.

A more honest goal

If not zero harm, then what?

I'd suggest something like this: we will continuously improve our ability to identify, understand, and reduce risk, and we will measure our progress by the strength of our prevention systems rather than the absence of incidents.

It's less catchy. It doesn't fit on a poster as neatly. But it's honest about the nature of risk in complex operations, it doesn't incentivise underreporting, and it focuses attention on the systems and behaviours that actually prevent harm rather than on a number that may or may not reflect reality.

The intent behind zero harm is the right one. Every worker deserves a safe workplace. No injury is acceptable. The question is whether the language of "zero" helps or hinders the systems that deliver on that intent.

In my experience, the organisations that get closest to zero are the ones that don't talk about it. They talk about coaching, learning, data, and continuous improvement. They build systems that catch near misses before they become incidents. They invest in the leading indicators that predict risk rather than the lagging indicators that count failure. And they create cultures where the most valued behaviour is not a perfect safety record but the willingness to speak up when something isn't right.

That's the honest conversation. And it's one worth having.

Want to see what a learning-focused safety system looks like in practice? Book a demo and we'll show you how inviol's coaching-first platform turns continuous safety data into the team conversations that drive genuine, sustained improvement.

Frequently Asked Questions

What is zero harm in workplace safety?

Zero harm is a philosophy that commits an organisation to the goal of no worker being injured or harmed during the course of their work. It has been widely adopted as an aspirational target across industries, particularly in construction, mining, logistics, and manufacturing. Supporters view it as a moral commitment, while critics argue it can incentivise underreporting and stifle learning.

Does zero harm actually reduce workplace injuries?

The evidence is mixed. Research published in Safety Science found no statistically significant difference in serious injury and fatality outcomes between companies that adopted zero harm commitments and those that didn't. While zero harm has motivated investment in safety infrastructure, critics including Professor Sidney Dekker argue it can suppress honest reporting and erode the learning culture that prevents incidents.

What is Safety Differently?

Safety Differently is a framework developed by Professor Sidney Dekker at Griffith University. It challenges traditional safety thinking by arguing that workers should be seen as a resource to harness rather than a problem to control. The approach emphasises learning from how work actually happens (rather than how it's prescribed), building trust, and creating cultures where people feel safe reporting problems.

What should organisations measure instead of zero incidents?

Leading indicators are more useful for driving improvement than lagging indicators like incident counts. Effective metrics include near-miss detection rates, time from hazard identification to corrective action, coaching session frequency and completion, worker participation in safety activities, and trend data showing whether interventions are reducing risk over time.

How does computer vision AI support a learning-focused safety culture?

When implemented with a coaching-first design (as inviol is), AI captures safety events automatically and anonymously, removing the stigma of self-reporting. Faces are blurred, data shows patterns rather than individuals, and detected events feed into team coaching conversations focused on learning and improvement. This generates the honest, continuous data that learning cultures depend on.