Predictive-based safety: Reshaping the landscape of insuring construction

Authored by AXA XL Strategic Operations Manager for Risk Engineering, for North America Construction Rose Hoyle and VP of Industry Strategy at Smartvid.io Tim Gattie

As more and more industry firms adopt construction technology, business leaders are exploring creative ways to harness AI and leverage big data for more than just “interesting” insights. In particular, we are looking for ways to have greater influence on the entire construction project lifecycle, including insurance. AXA XL and Smartvid.io are teaming up to explore what’s possible for the future of construction insurance when we wield the immense power of predictive analytics in the field.

What is Predictive-Based Safety?
At its core, Predictive-Based Safety (PBS) is about using data to change behaviors. While the behavioral-based safety systems currently in wide-spread use on construction projects today have brought recordable incident rates down significantly, it is by applying technology and data together to drive behaviors that will take contractors to the next level of safety success.

There are 3 elements of a PBS program: Data Collection, Data Transparency, and Prediction Deployment. Each of these elements requires training related to the application of appropriate behaviors by those involved.

Data Collection
The predictive models that enable PBS require data from several sources including safety observations, safety incidents and other project data. Collecting data on interactions between observers and employees in the work environment by observing safety behavior is essential if PBS is to be implemented correctly. These observation interactions must be purposeful and have safety at the center. The observers should be trained in guiding the interaction, hazard recognition and collecting meaningful observations.

There are many tools available to collect observation data, however those purpose-built for applying artificial intelligence (AI) and predictive analytics collect the data differently. They collect both direct user input and background data that predictive models use to rate and rank each observation. The data are collected and stored in a way that allows them to be understood and integrated by predictive models and business intelligence software.

The second source of data is from incidents that take place on each project. Incident data collection should include every single incident that takes place on all projects including near misses/ hits, property damage, equipment damage, 3rd party incidents, and of course, injuries of all types, regardless of how minor. The data from these incidents serve as really powerful indicators of safety risk. For many contractors, reporting at this level of detail requires a change in both process and procedures. They may even need to use new tools to collect the data digitally. Contractors who build a culture of reporting all incidents are often surprised at how many incidents went unreported before having the culture, tools and processes in place.

Additional enterprise data are necessary to round out predictive models. To achieve this, it is important that a contractor’s IT department works to integrate data across their organizations. The insights that can be gained from combining and overlaying data from different sources can be staggering and redefine the rules-of-thumb the organization has employed for years or decades.

Data Transparency
Providing full data transparency is key to implementing a successful PBS program – but it comes with a catch. Once a meaningful dataset is assembled, the next challenge becomes managing reactions to the data. For many organizations, having full transparency into a near-real-time dataset is something they have never experienced. PBS includes a set of expectations with regard to viewing and interpreting the data. Reacting poorly to the data results in a hesitancy on the part of observers to report what they are actually observing, skewing the data driving predictive models or worse, it could lead to the refusal by many observers to use the data collection tools rendering PBS ineffective.

Prediction Deployment
The outputs that come from a predictive model take two forms: predictions and prescriptions. Predictions answer the question “Which job is most at risk of an incident?” Prescriptions are actions that project teams can take to intervene and reduce the risk of the prediction actually occurring. Prescriptions answer the question “What should I do?” Teams need to be trained on the use of the insights provided by the model and on how to interpret them.

With predictions and prescriptions in hand, project teams are able to focus in on the details of their safety program that require attention rather than use the shotgun approach that is so common today. Their actions are purposeful and deliberate with the goal of “proving the model wrong”. At the executive level, predictions help narrow the list of projects to help operations and safety executives focus support resources where they are most needed. Effective deployments of PBS have seen reductions in contractor-recordable incident rates as high as 60% in the first year.

Potential Insurance Implications
Predictive-Based Safety has the capability to not only change the way we build, but also the way we insure the way we build. Traditional worker’s compensation premiums are developed starting with NCCI rates, then factoring in credits or debits based on losses, EMR, and the contractor’s safety practices. These elements are standard underwriting data points for determining reliable indicators of estimated future results. While the aforementioned methodologies have served well thus far, what if there was a tool available that goes beyond “standard”? A tool that could add additional credible value to reduce losses and better predict future premium costs.

Smartvid.io’s AI engine touts an impressive prediction accuracy rate as high as 86%, and commensurate reductions in the occurrence of recordable incidents as high as 60%. With this level of confidence, data and AI may soon be able to influence worker’s compensation premiums, much like your personal auto insurance telematic plug-in could score you a “safe-driver discount” after observing your actual driving habits for a period of time, rather than rating you based on your past history and that of other drivers “like you.”

Both for safety and other construction risk metrics, using real time data to assess actual risk versus past risk may allow insurers to better predict future results in their underwriting, and provide more personalized customer service. It’s conceivable that with enough data, the commercial world could mirror the emerging innovations we are seeing in the consumer market, including customizable policies, supported by personalized risk engineering services, and more streamlined claims processes, all courtesy of big data and AI. What if you could save lives and reduce your premium in real-time? With predictive-based safety, this may soon be a reality.