Observations, past incidents and external information can help prevent the next injury
An electrical contractor in Alberta decided about four years ago to introduce predictive analytics software to improve its safety performance. Over a three-year period, the company identified a direct correlation between increased employee participation and reduced recordable incidents. With the new program, employees were identifying 10,000 to 15,000 hazards annually, up from just a few hundred per year.
“The company built their own leading indicator, which tells them how many safety activities employees are doing per the number of hours they work,” says Josh LeBrun, president and chief operating officer at eCompliance in Toronto. “It’s almost the opposite of a lagging indicator. It’s based on all the activities a person is doing.”
Predictive analytics can help an organization significantly reduce incidents, and that effectiveness is encouraging the technology’s drive in new directions, from smart personal protective equipment (PPE) to machine maintenance.
In collecting data for a predictive model, the safety team should look at a broad range of data, says Andrew McHardy, senior manager, infrastructure and capital projects practice, at Toronto-based Deloitte Canada. While incident records form a core source of information — when an incident occurred, who was involved, severity, etc. — the team also needs to look at other company systems that contain relevant data points, such as training, operations, maintenance, overtime, shift schedules and nature of the task. Valuable information may also come from outside of the organization.
“There’s a lot of open-source information that tells us about demographics, the environment, weather, seasons, industry,” he says. “It’s some of those, from a safety point of view, non-traditional data sources that can often be the most valuable... To drive and build out statistical models that are meaningful, we need to think about data in a big way.”
One purpose of the eCompliance software is to collect observation data. Employees are encouraged to identify all hazards they see, providing information on leading indicators. Workers use a mobile app to report hazards when they see them. In addition to their description, they can include a photo of the hazardous area in their message to the safety manager. Information collected is more timely and more accurate than that acquired by having a small safety team walk around and observe people working, LeBrun says. Moreover, much more data is collected.
“We have customers who have 200, 300, 400 employees in their organization that are identifying 15,000 to 20,000 hazards a year. It allows for some really interesting and impactful data,” he says.
The application also collects information on actual incidents.
“We can start to correlate the leading indicator data with the events and provide information to companies on what is likely to precipitate an event. Or, looking at other companies within their industry, if workers are doing certain things, do they have a higher probability or lower probability of incident?” LeBrun says. “We’re seeing a ton of hazards that could theoretically result in an incident, so we’re going to get ahead of that.”
The next step is to analyze the data with a view to identifying accurately the factors affecting safety performance.
“Safety is obviously about human behaviour. It’s about culture. And there’s a certain element of bias that creeps into decision-making. Analytics gives us an objective lens on that to be able to challenge that and reinforce a risk-based approach where we’re really trying to understand the factors that affect performance and quantify them where possible,” McHardy says.
Identifying these factors and their significance allows the safety team to build a hierarchy of factors and start to be predictive.
“Then you focus not just on lagging indicators but look at leading indicators. You can then project forward and anticipate what mitigations or interventions in your initiatives in your safety program could make the greatest impact, whether that’s more of what you’re already doing or making refinements to your safety program,” McHardy says.
Once decisions are made, it’s important to share the predictive data with employees so they understand the reasons for the decisions. This way, a safety team can increase worker buy-in and the likelihood of better results.
With safety analytics models, the value is in the interpretability and being able to use them as tools for culture change, understand root causes and what’s driving these models, says Alik Sokolov, manager, financial advisory, at Deloitte Canada.
“If all you can tell is that someone’s at risk, but you don’t know why, how do you action that in the field? You can’t just prevent the top 20 per cent of your riskiest employees from working. You need to be able to tackle the root cause. And what makes those models actionable is understanding the root cause.”
Another type of predictive software works by analyzing personality and is based on the idea of a direct connection between certain behavioural types and safety. For example, The Predictive Index software, sold by Whitby, Ont.-based Predictive Success, assesses employees for five types of behaviour that affect a person’s tendency to act safely: dominance, extraversion, patience, formality and judgment.
“The software measures your five behaviours on a subconscious level — how your subconscious feels you should act. This is your makeup. You’re hard-wired this way, so it’s hard to cheat. It also measures how you consciously think you need to modify those behaviours,” says Eric Irwin, managing principal.
Employees answer questions based on an ideal behavioural model — the behaviours best suited to the task. The software draws on data collected through a number of studies, Irwin adds.
People high on “dominance,” defined as assertive, independent, self-confident and driven to win, tend to be more focused on themselves and more likely to take risks in order to win. A truck dispatcher, for example, may tell a driver to get a delivery to its destination in five hours when he knows it will likely take eight.
In contrast, people low on dominance are unselfish and focus on harmony, collaboration and protecting their team against risk.
“That’s who they are. They are naturally driven to protect their team more than they are to achieve any particular results. They’ll put the safety of their team ahead of results. From a safety point of view, you want to be lower on dominance,” says Irwin.
Another significant behaviour is formality, the degree to which someone follows rules. People high on formality need to follow rules and structure. They hate making mistakes.
Behavioural assessments are designed to help organizations decide which employees should be promoted into supervisor roles and what behaviours need to be modified. They help identify which worker will make the kind of supervisor who will oversee a team that works safely, as well as which workers and supervisors need coaching on how to work more safely.
“Whether it’s a dispatcher or high-risk work supervisor, their behaviour is critical to predicting a safe environment for their workers. If they’re low on dominance and high on formality, they’re going to naturally protect their teams and follow the rules exactly as written. That’s a great mix to ensure you will improve safety results,” Irwin says.
PPE can now be used to carry sensors that collect and send data between workers and safety managers to improve safety at a work site. For example, the Smart Helmet Clip, developed by Deloitte and Toronto-based Cortex Design and expected to be available later this year, is designed to increase situational awareness of surface operations for mining staff underground.
The device is battery-operated and clips into a docking mount under the bill of the miner’s helmet. The sensing device conducts continuous local environmental and activity monitoring, and the information detected is passed up to a dashboard so surface operators have data on what everyone underground is doing at any given time, says Dylan Horvath, president of Cortex Design.
“The position of the helmet itself uses accelerometers to monitor activity tracking. Based on the movement of the helmet, you can tell quite a bit about what the miner is doing: If they’re walking, on a vehicle, operating equipment or if they’ve taken their helmet off and put it on a seat in a vehicle. It will also detect if a miner is down. All that vibration and behavioural data is captured by this device just from moving the helmet,” he says.
While the initial purpose of the helmet clip will be safety monitoring, the device will likely be used in the future to provide predictive data. Over time, Horvath says, the device can build up analytical data for individual workers. As a result, when any one of them begins to act in a way not normal for them, those changes in behaviour can be detected and used to predict possible safety risks, such as worker fatigue.
“If you can imagine a worker who is normally very alert coming in one day and feeling sick or tired or diminished in some capacity, their head movements would be different,” he says.
“We haven’t implemented this yet, but the ability to collect this type of data is the intent of the device. So if you have a worker who is compromised or at risk of not being able to operate heavy equipment, then surface operations can make some informed decisions.”
Predictive analytics technology is also being used to alert managers to potential problems with equipment, which cause injuries due to faulty machinery. For example, in the case of Avantis PRiSM, made by Schneider Electric, sensors in the machinery collect data that is input into the predictive software. The program is able to compare historical operating data on any machine with real-time operating data to identify slight deviations in that machine’s behaviour. The program then provides early-warning notification to managers or controllers, who have time to analyze and fix the problem before the equipment fails.
There is constant technological advancement in the predictive analytics space, LeBrun says, primarily coming from machine learning. His company has seen some interesting ways to integrate big data information that is part of public knowledge. For example, they have been able to measure the weather for incidents recorded with eCompliance.
“We correlate the level of incident with the weather patterns, and we can see where there is a high level of risk of incident based on the weather. Then we can send out push notifications to employees to say, it’s going to be rainy; here are the things you have to watch out for. So we can start to be more predictive and proactive in reducing risk.”
When selecting predictive software, McHardy says a company should have a good understanding of its aspirations and objectives. It should also consider the size and scale of its operations, as well as the size and scale of its data and its relative quality and fit for purpose.
“What is their existing architecture? What tools do they have in-house? Then they can make an informed decision on what technology they might need,” he says.
Make sure the analytics software is easy to use, advises LeBrun. If it is not, employees won’t use it no matter however good it is.
“You won’t have any data, and you won’t really be able to predict anything,” he says.
It’s also important for safety managers to assess their data management system.
“They need a robust analytics platform that will allow the company to analyze all the data being generated by employees and, ultimately, to make really strong decisions to reduce risk,” McHardy says.
With the data from its predictive safety software, the electrical contractor in Alberta was able to produce a kind of participation score, LeBrun says. With that information, every employee can understand their level of involvement in safety and a company can reward employees who have a higher level of safety participation.
“And it’s a good way,” he adds, “for safety teams to identify who the real safety champions in the company are.”
Linda Johnson is a Toronto-based freelance journalist who has been writing for COS for seven years.
This article originally appeared in the April/May 2018 issue of COS.