The Woolworths Experiment

The most pressing question for many people who are already ‘sold’ on Safety Differently is ‘how do you do it?’ How do you enable and empower people and their organization to develop and implement Safety Differently? This is what micro-experiments help you do. If you’ve seen Safety Differently, the Movie, you will know about ‘The Woolworths Experiment.’ And if you’ve read the book The Safety Anarchist, you will have come across a more elaborate description of it.

The Woolworths Experiment, the first deliberate, randomized-control, documented micro-experiment with Safety Differently, was born at a Learning Lab for company executives a few years ago. As I was teaching, I took up the Drachten shared space example. I showed a picture of the crossroads. I explained how Monderman and his traffic engineers had taken away all the signs, traffic lights, separators and lines. How they had redesigned the square as one large open space made up of uniform brick. I talked through the results (gridlock a thing of the past, an almost tenfold reduction in accidents, a sense of simultaneous ownership and vulnerability among the people who used the square—make things feel riskier and you get safer and more engaged behavior).

And then I wanted to move on.

“Hang on a minute,” said a Woolworths executive. “What if we were to do that?”

I had a sense of what he might mean, but said, “Do what?”

He replied: “Take everything out. All the safety stuff we have put in our stores. All the top-down rules, the signs, the checklists, the procedures.”

I listened.

“As an experiment,” he continued. “See what happens. See how people create safety when they’re left alone.”

He had certainly aroused my interest. An experiment, with several conditions, testing how safety anarchism could actually work in real-life workplaces under controlled circumstances? That would be very cool.

Fast-forward 18 months

Fast-forward 18 months. That is how long it took to convince not only some of his fellow executives, but also the various regulators who oversee different aspects of their operations. And of course we had to design the experiment, do a pilot trial, get unions on board, and explore a way to randomly assign conditions to groups of comparable stores.

We had to get ethical clearance from the university, because this was a true experiment, with us meddling with people who did work that could potentially hurt them. One of the features I suggested was to make the experiment ‘safe-to-fail.’ We were able to pull the plug on the whole thing at any time, and quickly revert to the old system of top-down safety controls. If ever we got the slightest hunch that risk was going up because of the experiment, or worse, that someone had got hurt in one of the experimental conditions, we’d call it quits immediately.

It wasn’t as if nobody was getting hurt under the old system. People were. In fact, incident and injury rates had flatlined for a while, and were now on the rise. Woolworths, a huge supermarket chain and one of the largest private employers in the country, was becoming concerned that doing ever more of the same was not going to lead them to something different. An almost 100-year old company, it was organized in a classical hierarchical way, with very little decision power lower down. Store managers could not even decide to put a particular product on a shelf other than what they had been told from above.

You can imagine what their safety regime looked like. So called ‘safety packs’ were sent down from head office every month, specifying topics of concern, new legislative requirements, and containing new checklists or procedures that had to be implemented (e.g. team talks, particular ways of cleaning or shelving or unloading or organizing back-of-house work).

What did the frontline say?

What did people actually think about this, the people who worked in the stores? “Our current safety pack is a long task,” the complaints from the frontline went, “and the team loses interest in it. We need something that is not as time-consuming, and that is not just one point with which to keep people interested in, and proactive about, safety.” And as a store manager put it: “filling out the safety pack does not improve our safety results.”

Safety meetings needed to be held by a particular group at specific times. A notice board was to hang in a certain place, and a required menu of things related to safety was to be displayed there. There was no evidence that anybody every looked at what was on it. Tools for people’s work, from mops to knives to machine guards on meat-slicing machines, were all sourced by the head office.

One worker told us: “I don’t think about safety. I just follow the rules and do as I’m told.” Don’t think. Just do as you’re told. Follow the rules. How much more Safety I could you get? Getting on your store’s safety committee was not based on merit or skills or knowledge. It was sometimes seen as punishment or as a welcome (though in content entirely useless) reprieve from, say, stacking shelves.

It was time for a change.

A micro-experiment

But changing everything overnight, in all stores (and into what exactly?) was seen as too bold, or stupid, or dangerous. And where was the evidence that another approach might work better? This is where the micro-experiment comes in.

A micro-experiment is a safe-to-fail, small-scale project, using the company’s own workplaces and workforce. The aim is to explore and test doing safety differently, for example by taking out a procedure, or removing duplicate paperwork. In Woolworths’ case, it involved taking out pretty much everything related to safety The intention was to do this at a small group of stores, under controlled conditions, compared to other, similar stores, where we either did something different or changed nothing at all.

The only things we could not take out were fire exit signs, as they are federally mandated. And there were a few more items like them. The idea of a micro-experiment is that it generates the kind of credible, internally validated data that an organization can use to build some confidence that a different approach to safety might actually work for them.

In the Woolworths experiment, we devised three conditions:

  1. Take everything out. This condition, which we formally called the ‘local ownership condition,’ was the one in which we removed all the safety processes, procedures, checklists and rules that were not specifically required by state or federal law. In this condition, we wanted to create the completely open conditions for grass-roots safety to germinate and grow. We took everything out, made no suggestions about what to do instead, and left the stores with only one rule: ‘don’t hurt anyone.’
  2. Take everything out and retrain according to Safety Differently. This condition, which we formally called the ‘ownership and engagement condition’ was driven by deliberate change management, which included training sessions for store workers and managers. These were modeled on the ideas of Safety II and Safety Differently: see people as a resource to harness, not as a problem to control. Don’t tell people what to do but ask what they need to be successful, and stop counting negatives as a measure of your progress. Instead, identify and support the positive capacities in your people and teams that make things go right. We wanted this condition in there to see whether there were any radical differences between how people organized safety for themselves when left entirely to their own devices, and how they did so when actively instructed or inspired along new lines. In this condition, too, store workers and managers were empowered to take out what they didn’t think was useful.
  3. Control condition. This condition was literally our control. It involved a group of stores that were comparable to the stores in the other two conditions, but we changed nothing in them. They kept doing what they had been doing. Head office stayed in control of safety. It kept sending down its safety packs and expecting compliance in return. Store managers or workers were not given any more leeway.
Three times ten

We found 10 stores to assign to each condition, for a total of 30 stores. This was of course a bit tricky. We needed to avoid ‘picking the winners’ for the first two conditions (which I’ll collectively call the ‘ownership’ conditions). That would have been easy. In conversations with Woolworths managers, we quickly learned that some store managers were known to be willing to try new things, to be naturally more open to new ideas, interested in their employees and accessible for them. It would have been easy to seek those out and assign them to the ownership conditions, as that would surely lead to success. But it would mess up the experiment, because how could we fairly compare across the conditions if we put the presumed winners in the conditions we wanted to win, and the left the more hopeless stores and store managers to the control condition?

So we started with a relatively contained geographic area in which we found 30 stores. Even across this area, there were socio-economic variations, and stores that had had an internal furnishing upgrade versus those that hadn’t yet. There were stores with great managers and stores with so-so managers. There were large stores and smaller stores. There were male and female managers. And a whole bunch of other factors.

We sat down with Woolworths managers and devised three groups of ten stores each which had—to the extent we could control this—as much or as little of all of these factors as the next group. We needed to be sure that there were no a-priori biases toward success or failure in any of the three groups. They had to start from the same place. And so they pretty much did. Then we randomly assigned the three groups of ten stores to the three conditions. The experiment started the day we took everything out of the stores in the first condition and started training people from the second condition. It finished a year later. There was no loss of data during the year of the experiment, as all stores stayed with us throughout.

Bigger worries

Of course there were some concerns beyond the sheer design of the experiment. If there is collective representation, for instance, then what do unions say when you start ‘experimenting’ with worker safety? Interestingly, our experiences show that the responses are quite diverse, or even ambivalent. On the one hand, unions are rightly concerned when you announce you are going to take away the reasonable employer-provided protections that seem to keep their workers safe. And what about the organization’s lawyers, how do they look at this? Again, our experience was that there is no substitute for sitting down with stakeholders, including lawyers, and being open-minded about their concerns. We rationally went through all the pros and cons of changing these things about work. With reasonable safeguards in place, and a limited scope that specifically aims to improve how an organization does its business and protects its workers, there really are few obstacles. This went for regulators as well. Organizations such as Woolworths have a number of regulators watching over their operations. We found that the ones who were most closely concerned about workplace health and safety had also begun to understand that doing more of the same was not going to generate different results at Woolworths. They, too, were keen to hear new ideas and explore different ways to improve safety results.


When given the opportunity, people gladly throw off the yoke of bureaucracy and compliance. 19 out of 20 stores (a full 95%) from the two ownership conditions immediately ceased compliance activities mandated by the monthly safety pack. They all agreed that these things added no value, and didn’t impact safety outcomes. A store manager commented: “I think that removing the administrative tasks has inspired the team to be driven to look at safety in a different light. Instead of a chore, it is now more enjoyable: they look, observe and engage in what really matters, day to day.”

And indeed the store manager’s role changed as well. They no longer performed the role of overseer and auditor. Instead of chasing workers for dates and signatures on meaningless paperwork, they found that they were spending more time with people—listening to what mattered to them, discovering the daily obstacles and challenges that stood in the way of creating success. Workers, in turn, found managers to be much more responsive to their concerns. Local ownership really meant something. When we surveyed workers on their perceptions of leadership, those in our two ownership conditions rated their store managers higher on the ability to empower individuals and enhance skills and self-sufficiency than anywhere else in Woolworths.

Interestingly, stores and store managers in the ownership conditions also became more assertive in requesting help from the head office. Now that they had more ownership for safety, and more engagement locally, they didn’t hesitate to make their needs and demands known to those who were tasked with supporting or supplying them. Some were bemused that it took an experiment run by a university to restore or invigorate their internal organizational links and relationships. And stores in the ownership conditions saw more initiative across the board. In one instance, box cutters supplied by head office had long been considered a hazard, so store workers now sourced better cutters on their accord.

Freedom in a frame

These are not complex interventions, of course. but the results can be amazing. In the second ownership condition, there was a reduction in the number of lost-time injuries (if we still wanted to see that as a relevant measure: many people did). It was interesting for us to see that the number and diversity of initiatives (like bringing in or adopting new tools to perform backstore tasks) were greater in the second ownership condition. Apparently, only setting people free was not enough: people need some inspiration of what can be done, of what they can potentially achieve, they need some knowledge, and active empowerment through examples of what others have achieved in similar circumstances.

Decluttering compliance and bureaucracy is a good start. But the second ownership condition showed that engaging people actively in a different way of doing safety, and giving them the freedom and autonomy to pick and choose and develop what they want, is an even more powerful combination. The trap of course, is that any guidance on how to do safety differently can become yet another kind of authority, another kind of top-down intervention, another way of telling people what to do. We avoided this as much as we could, by leaving the actual development of safety work and other interventions inside stores to people themselves.

A whoopie prize

The jewel in the crown of the experiment came toward the end. One of the stores in the second ownership take everything out condition was awarded Woolworths’ annual safety prize. The committee awarding the prize wasn’t aware of the experiment, but must have liked what they saw, and the results it produced. We can’t say for sure that the store won the prize because it was in the take everything out and retraincondition. But we can say for sure that being in that condition didn’t hurt their chances of winning it. That should be reassuring to anyone wanting to try a similar micro-experiment.

But wasn’t this all caused by the Hawthorne effect? The Hawthorne effect refers to organizational research originally conducted during the 1920’s and -30’s at the Hawthorne Works, an electric factory in Illinois. In those experiments, researchers wanted to know whether worker productivity changed with variations in lighting, break times, and working hours. It changed, for sure, but not with any clear correlation to the variations in whatever the researchers were manipulating in the workplace. Productivity went up across the board. In fact, when the researchers packed up and left, productivity slumped again. Researchers concluded that worker productivity goes up simply because you’re paying attention to workers, and because you show interest in their situation. Clearly, a little humanity goes a long way. But it does create a potential confound in studies such as the Woolworths experiment.

The way we dealt with that was to be scrupulous about how much attention we gave to, and how much time we spent with workers and store managers across all conditions. So even the stores in the condition in which nothing was changed, where the old regime was still in place, got as many visits and conversations from us as the other two. In this way, we kept the amount of attention given to workers constant across all three conditions, thereby spreading any Hawthorne effect out over all conditions equally and thus leaving them comparable. This gave us confidence that the change in leadership perceptions and safety results in the two ownership conditions really were related to our safety anarchism changes, and not just because we were there.

Can you do your own micro-experiment?

So what do you need so that you can conduct your own micro-experiment? Before I answer that question, just consider this: In many organizations, it may not be smart to call a micro-experiment an ‘experiment,’ as it invokes fears and uncertainties about ‘experimenting,’ about trying out new ideas, methods or activities that play fast and loose with people’s safety. For risk averse managers or boards, it is probably less problematic to call it a ‘project.’ Organizations always have projects going on. It can then even designate someone to be the ‘project manager.’ This should not, however, detract from the rigorous scientific design of the experiment that runs under the label of ‘project.’ It is this design, after all, and the strict comparability across conditions, that allows leadership to draw valid and reliable conclusions about doing safety differently in its own organization. Here’s what you can do:

  • Find two or more groups (sites, teams, locations) that are comparable because they do similar work and have a similar make-up. To the extent that you can control it, make sure that these groups will remain relatively stable for the duration of the experiment (e.g. no management shakeups, no radical changes of leadership). If there are such changes along the way, you may have a harder time attributing any results to what you did, as opposed to what was done to the group by those other factors.
  • Study what you can change or take out. Is there unnecessary bureaucratic clutter? Is there overlap? A typical case of overlap would be procedures that a contractor uses, which do almost the same as those the lead organization uses, but people working for the contractor (which is working for the lead organization) have to do both. Are there rules that nobody believes in? You can find this out by asking what people consider to be the stupidest thing they have to do every day in order to be allowed to work on a particular site or project. It’s a great question to ask, and you’ll surely get enlightening answers.
  • Do a small pilot. This might involve just talking to people, or testing your idea through a thought experiment, or actually testing it live with a group of people. You can learn a lot from these small pilots (e.g. you might learn that the thing you wanted to take out is not at all what frustrates people the most).
  • Reserve the time to let the change(s) take effect. Don’t think you can do a micro-experiment inside of a few weeks, though you might see some immediate effects (as we did in Woolworths: the previously mandatory safety packs were abandoned as soon as they were no longer required in the ownership conditions). Other effects will take more time to become visible.
  • Measure the changes. You can do that by using safety indicators and measures you are already using, but you might also want to think about additional measures to take that are more positive than that (e.g. leadership perception, empowerment and locus of control, happiness at work).
  • Collate the findings, celebrate the successes and communicate them to others in the organization, so they are inspired to take your experiences on board. They may even be inspired to do their own micro-experiments so as to innovate and improve an area of their own work.

Remember, a micro-experiment is powerful in part because it involves data generated by your own organization. It’s not just an idea or a belief: it is evidence that another way of working is possible—and very possibly better.


The experimental protocol and study execution of the Woolworths Experiment was conducted largely by Michelle Oberg, a doctoral student in the Safety Science Innovation Lab at the time of the Woolworths Experiment. Readers will likely have come across Michelle’s excellent implementation and application work of Safety Differently since.

Martin O’Neill was the senior leader at Woolworths who championed The Woolworths Experiment. He fields calls every month from colleagues at other organizations, asking him how in the world he did it. Countless men and women at Woolworths were instrumental in making the micro-experiment happen. You can see them in Safety Differently: The Movie.

Finally, this blog contains excerpts from the book The Safety Anarchist (Routledge, 2018).


  1. Matita Tshabalala Reply

    Very well done. This gives many of us a head start in terms of how to go about doing safety differently. Thank you

  2. charlestortise Reply

    It would seem that the greatest barrier to safety in this case, and probably others is not the adaptive capabilities of those doing the work to do things right and be safe but the style of management that prevails? It is indeed a tyranny and a prison that has trapped the people doing the work within a system over which they have no control but which has the greatest influence on what they do. What you are demonstrating under the title of safety differently is management differently if you do not change one you will never have the other as if you do not change the attitudes of those responsible for the system you will never improve the safety of those working within the system. They have to undergo a transformation and relinquish some of their power of control in order to engage the collaboration of those they employ. It is just a shame that they have to believe there is a trade-off, may be in terms of avoiding costs associated with accidents or other trauma before they will listen. Elsewhere Erik Hollnagel describes Safety II as safety synthesis but you should not just wait for a different way to address accident rates to change.

  3. Craig McManus Reply

    Great article, Sidney. It’s important to provide the ‘how’ articles as much as the ‘what’ articles. Safety practitioners will get excited at what they read on this site and in your books, but that can sometimes be difficult to translate into practical application terms when trying to espouse this model to those in positions of influence within their organisation. More of the same, please!

  4. John Culvenor Reply

    This is an interesting experiment. Well worth doing. A contribution that sought to test interventions that are all too often if not virtually always taken as being useful. But there’s a “but” coming of course. I’m a but stumped in most discussions on this topic about just what is safety 1 and safety 2. But what occurs to me here is that the proposition that this experiment tested a “Take everything out” condition was not actually even close to being true. In a business such as the one described, most of the important safety work remains. The stores will still have physical places designed to standards, both government regulated and internally regulated, they will have equipment with the same “safety 1” history if that is the right expression, there will be thought gone into logistics methods including the shape and size of products, how they are delivered, and merchandised, etc. The process of selecting and training staff, sometimes with trade or other qualifications, remains. All that was taken away was the layer at the store. Fair enough though, it is an experiment where this layer seems to be removed. And it is interesting. And a useful contribution taking a counter point view at testing some types of safety interventions that might not be much use. But we should be cautious to take it for what it was and not something much broader. Most of the “old” “safety 1” interventions would have remained common to all three of the experiment’s conditions. I think the processes at the store and safety with it, including the prospect of elevated risk of serious outcomes, would be inevitable if actually all of the “old” “safety 1” was “taken out”.

    1. clinton horn Reply

      You raise some interesting points John Culvenor that, coincidentally, some of which my PhD research is seeking to address/answer. I am a keen advocate for the principles of Resilience Engineering / S-II and what it brings to framing and contextualising the safety conversation and i’m sure that the contemporary thinkers in this field will acknowledge that there is a ton of opportunity to expand and add to this rapidly evolving body of knowledge in both the academic and, more pressingly, the operational space to provide answers to such questions and to the debate. The amount of literature and debate in both these spaces is growing exponentially and is indicative of the level of curiosity from all sectors that, again, is indicative of the sheer amount of research/exploration opportunity (academically and operationally within business) in this field that I think is both needed and welcomed.

      The various sciences around the construct of human decision-making are many and varied and, as an extension, how workers capacity to adjust and adapt their performance at the point of work is contributing to organisational outcomes in the face of constant variability (as experienced by workers) and often independently of the system.
      It is the word ‘capacity’ that grabs my attention and where I believe much of the debate emanates from (in my humble opinion) and where the opportunities for further exploration are endless.

      As a practicing safety practitioner, I think it is important to explore all sciences/knowledge regardless of origin or bias in our quest for finding marginal gains and added resilience (it’s a great word!) to improve both safety and the wider organisational performance. I try not to over-think the S-I or S-II debate but to, instead, look for the useful elements (or what can be adapted) from both that can contribute to those marginal gains and level of resilience in an operational business context and the economic and commercial constraints they operate in and that inform much of senior management’s own decision-making. It is also a much easier way to bridge S-I and S-II (as a concept) when pitching to senior management who are often not as well read when it comes to systems thinking, and who are perhaps favouring a more traditional/orthodox approach to safety. Change is scary for most, especially when seeking absolutes is the order of the day!

  5. Dominic Cooper Reply

    This case study would have so much more validity if you actually published the pre-and post intervention incident rates, otherwise it is all just hearsay. It would be even better if you published the experiment in a peer reviewed scientific journal (e.g. safety science). It is what all the sceptics are waiting for.

  6. Kevin Bush Reply

    I would like to see this type of experiment across the construction sector with all of the legalities that would entail. I can see it working to some extent, but when you un-regulate you must have a workforce that is vastly experienced in the dangers around them to be able to develop their own controls or specify what would help them in terms of equipment.

Leave a Reply

Your email address will not be published. Required fields are marked *