While the construction industry made a great deal of progress reducing the industry fatality rate in the 1990s and early 2000s, the fatality rate has unfortunately remained largely flat since 2008. In fact, in 2019, both the total number of private construction industry fatalities rose to 1,061, the most since 2007, and the fatality rate per 100,000 also increased, according to the U.S. Bureau of Labor Statistics. Construction accounted for just over one in five of all on-the-job deaths in the United States in 2019.


Fatality Rate

Construction hasn’t failed at lowering the fatality rate for lack of trying. The industry takes safety very seriously, and, even though this is a highly competitive industry, companies and contractors eagerly share their safety practices with one another. Unfortunately, despite trying a number of different methodologies, the safety improvement firms are looking for rarely materialize.

But while progress in improving jobsite safety has remained largely flat, technology has made tremendous advances. After all, in 2008, the iPhone was just a year old, and who would have thought back then that today we’d be regularly streaming movies on these tiny devices? In enterprise technology, one of the most exciting developments has been the emergence of predictive analytics, but the construction industry has yet to take full advantage of it.

This is not an untested technology. We encounter predictive analytics almost every day, usually without even realizing it. Netflix’s programming suggestions analyze your past viewing habits to predict what shows you’d most likely enjoy next. Mapping apps like Waze analyze the speed and location of thousands of drivers to predict the fastest route to a driver’s destination. Even the ads we see everywhere online via Google are served up based on our internet browsing habits.

It’s time for construction to apply predictive analytics to jobsite safety.

A Primer on Predictive-Based Safety

Predictive analytics applies statistical and numerical models along with machine learning to identify patterns in data to make probabilistic predictions about what will occur in the future.

Predictive-based safety (PBS) is the application of the results of a predictive model trained by safety professionals and project management teams to identify safety risks and actively prevent incidents before they occur.

Certainly, construction organizations have no shortage of safety data. They have been collecting it for decades to meet compliance requirements and to power traditional behavior-based safety programs. And that’s a good thing, because the more data PBS can access, the better the predictions will be. Typically, this data includes information from safety observations and incidents, with observations generally made by trained safety professionals using a form or checklist.

PBS, however, takes a more expansive approach to data than traditional behavior-based programs. Specifically, to drive PBS, observations should:

  1. Be performed by a wider swath of the firm’s employees and include operations personnel;
  2. Engage craft workers and crews in frequent safety conversations; and
  3. Collect risk rating information about the observation.

With more observers in the field, organizations will gather far more data, which makes it easier to achieve the critical mass of data required to make reliable predictions. But it’s important to collect these observations in a manner that ensures good data. Don’t use observations to nail people for not following safety protocol, because that will cause observers to be less likely to report safety hazards that they witness.

Additionally, it’s helpful to use a two-part scoring system that ranks the frequency and the severity of the hazard. Our research has shown a strong correlation between high observation risk ratings and impending safety incidents. When the safety team sees a high-risk rating on their report, the site in question is more likely to experience a safety incident in the following week.

And don’t ignore good safety practice! Rewarding people for doing the right thing is a far more effective means of encouraging good safety. Go for the carrot, not the stick.

Observational data, however, is far from the only data source available to construction organizations. What about the thousands of site photos that you already have on hand?

Artificial intelligence can rapidly analyze these photos to identify safety hazards, transforming them into observational data that predictive analytics can crunch to make even more accurate predictions.

Finally, all of the project data that your firm has on hand, from contract data and weather to staffing information, can serve additional fuel for the PBS engine.

Predictive-Based Safety in Practice

Predictive-Based Safety in Practice

So, how would PBS look in the real world? Let’s say a contractor has four active work sites and they’ve been using PBS for four months, feeding it observational data, AI-generated observations from site images and additional project information. One morning, when the project manager looks over her safety dashboard, she notes that one project shows a very high risk of an incident within the next week.

After drilling down deeper into the data, she sees that the rate of observations and the ratio of supervision on the project are both out of whack. The PBS platform recommends sending additional resources to the site, so she engages her team to immediately begin performing additional observations. She also works with the project superintendent to hire another foreman to balance the project supervision ratio. Two weeks later after no recordable incidents, the risk rating declines. Accident prevented.

The beauty of predictive analytics is that, given enough data, it can identify risk factors that human beings would rarely notice. For example, if a jobsite experienced a recordable incident last week, conventional wisdom would hold that it’s probably at lower risk this week, because people on site would be more on guard and would follow safety procedures much more closely. But in fact, the data shows the opposite is true. A jobsite that’s already had an accident is 2.5 to 5.0 times more likely to experience an accident the following week.

Predictive-Based Safety

The results of using PBS in the field have been tremendous. Studies show that PBS can accurately predict 70% of incidents on the top 20% of projects, and its predictive power continues to improve over time. Indeed, early adopters of PBS have seen their incident rates drop by more than 50% over a 12-month period.

There’s no reason the construction industry should accept the current rate of jobsite fatalities and recordable incidents. By incorporating PBS, contractors can sharply reduce their recordable incident rate in a short span of time, ensuring that more colleagues and coworkers go home safely every night. That’s an outcome we can all get behind.

[See How Suffolk is Using Predictive to Prevent Incidents]