Data on workplace incidents—and more importantly, their absence—can tell contractors a great deal about the effectiveness of their safety programs. But can that data also help predict when and where an injury may occur and—with the right intervention—prevent it from happening at all?

Article Index:

• Putting PA to Work
• Behind the Numbers

A recent study from Predictive Solutions Corp., a Pittsburgh-based safety software developer, suggests the answer may be yes.

The study cites research done with Carnegie Mellon University that finds predictive analytics—which use computer algorithms to predict future events based on patterns gleaned from analysis of historical and current data—can predict workplace injuries with accuracy rates as high as 97%. In addition, the study claims that an unnamed “top 20 construction company” used predictive analytics to achieve a zero lost-time incident rate at 90% of its worksites.

Nick Bernini, Predictive Solutions data scientist, declined to name specific clients in an interview with ENR but said that demand for the company’s services from multiple industries “is definitely going up” and that “a majority of our work is construction.”

For predictive analytics to work, contractors need data—a lot of data—which typically implies that a well-established and well-managed safety program already exists. What predictive analytics offers, says Bernini, is a way for companies to take their limited resources “and focus them in the right areas.”

Bernini says that in performing analyses for Predictive Solutions’ clients, the two most important pieces of information are “the injury incidents that we’re trying to predict and safety observation data points performed by managers, foremen or even the workforce.”

The analysis also looks at staff hours worked on a specific site, personnel training, production schedules, video monitoring and surveillance, images and any sensor information, such as from forklifts or wearable gear. Personnel data may also be used, albeit “on an anonymous basis, to look at trends,” Bernini says, rather than collecting information on specific individuals.

All that client information is fed into Predictive Solutions’ models, often in real time. Should the algorithm identify a risk at a particular location, clients are notified with an estimated time frame of the incident occurring, along with the likely cause and a suggested mitigation strategy.

While this may sound futuristic, Rick Zellen, a senior risk engineer in the Denver office of insurer Zurich North America, says that many contractors are already using forms of predictive analytics to help prevent problems of the past from occurring again. There are also training benefits, he adds, particularly for newcomers to the construction industry.

“The information from the analytics can be reinforced by the personal experiences of longtime employees,” Zellen explains.

This valuable information, however, comes at a cost. “Getting the most out of predictive analytics requires collecting a lot of data and continually maintaining it,” he adds.

Some insiders question whether predictive analytics can be applied to a highly dynamic environment such as construction. 

“To drill down into the data, you have to be pretty specific,” observes John Hymel, a principal with Sentinel Safety Consultants, Herriman, Utah. “For example, construction employees are often doing different tasks at different parts of the jobsite from one day to the next.”

Another concern, Hymel says, is accurately defining equipment types and how they are used. For example, there are almost as many types of construction ladders as there are ways to be unsafe while using them.

Then there’s the matter of where all that data originates. “If you’re relying on field observations, you can’t assume workers will have a depth of knowledge to make those distinctions,” Hymel says.

Calling reliance on predictive analytics a “crystal-ball mentality,” Hymel argues that constant vigilance is the best preventative strategy, particularly since construction injuries can occur with no warning. “Rather than taking time to analyze data, it’s far better to take immediate action, such as getting an employee trained up if you see something wrong on a walkthrough.”

Then there’s the variable that not even the most sophisticated algorithm may be able to predict—human behavior.

“Workers are subject to human frailties—forgetfulness, emotional states—that can’t be predicted,” Hymel says. “Any number of things may result in an accident, and none of them can be predicted.”

Putting PA to Work

Despite the costs and potential limitations, several major construction firms are incorporating predictive analytics into their safety management programs.

Eric Zuhlke, national safety director for Kansas City, Mo.-based  JE Dunn Construction Group, said via email that his company’s blend of software tools and techniques “provides a high level of accuracy used to drive optimal safety results,” allowing project teams to achieve more with fewer resources.

“Whether we’re trying to select the best partner for a certain scope of work [or] are interested in setting effective leading activities to drive a different safety outcome or pulling visual reports from one of our data collection points, we believe in the practice,” Zuhlke added. 

A key to his company’s success with predictive analytics, Zuhlke noted, is having the internal IT resources to support those platforms.

“The significant investment we’ve made in growing our IT platform has developed into a competitive advantage for us,” he said. And even with the demands of keeping the data up to date, IT gives JE Dunn the ability “to adequately maintain a robust predictive analytics program in today’s environment.”

It’s not all about computers, however. Zuhlke noted that many departments—safety staff, risk management, project management, integrated project services and operations—are involved with verifying the accuracy of the data.

“Depending on the predictive analytics platform, the complexity for interpretation varies,” he added.

For the past year, Hensel Phelps Construction, Greeley, Colo., has been using predictive analytics to enhance implementation of the company’s safety program by its project leadership team.

Relying on in-house resources and Zurich North America to develop the system, the company scores managers on how well they apply more than 20 elements from its published safety and health practices, such as planning procedures and performing the required hazard analysis.

Jerry Shupe, corporate director of safety and health at Hensel Phelps, says the system is designed to go beyond “checking the box” about whether a manager is doing something or not.

“We wanted to dig into what exactly the managers were doing,” he says, “and make sure they were applying these practices directly to a definable activity.”

Assessments are made every six months, with a focus on specific activities determined from recent incidents. Areas where managers overall are performing well are removed or modified, such as the recent addition of new confined-space requirements to the criteria for emergency planning.

An interesting outcome, Shupe says, was how the managers reacted to being evaluated, not unlike the principle of the Hawthorne effect, where awareness of being evaluated spurs better performance. Only in this case, Hensel Phelps’ managers have willingly looked for ways to make those improvements.

“Nobody wanted to be ‘average’ when it came to safety,” he says. “This forced them to get back into the safety and health manual, understand what it’s about and what is expected of them, and then find the best management practices to achieve and exceed those goals.”

As a result, Shupe adds, the scores have started to creep up. He’s now looking at additional ways to enhance the project to get real-time data at the project, district and corporate levels.

Behind the Numbers

Hymel doesn’t dispute the value that more data can bring to a construction safety program. Rather, it’s more a matter of what contractors do with that information.

“I’ve found that sometimes companies with higher incident rates have better safety programs than those with lower rates,” he says “Going behind the numbers takes time and effort, and not all companies are doing a good job of analyzing data or doing anything with it.”

Hymel likens overreliance on data to “microwave solutions to crock pot problems” that aren’t easily fixed.

“They take time, energy and understanding,” he says. “It’s not something you can rush.”

Which is why, he says, the real-time data-gathering approach that will benefit contractors the most is the practice of day-to-day safety observations and interventions they already use.

“Just being more effective in observing over the course of the day and seeing how work is being performed will tell you a great deal,” he says. “The best contractors are the ones doing this already."

Lessons Learned

• Safety experts claim that predictive analytics can foretell workplace injuries with accuracy rates as high as 97%.
• PA uses real-time data and algorithms to build models that can identify risks at a particular location.