In May, I participated (virtually, of course) in the CURT (Construction Users Roundtable) Safety Conference 2020 on how artificial intelligence (AI) is changing construction risk management. You can see my presentation here, but feedback on the presentation was so positive, I thought it would be worthwhile to share some of what I spoke about here on the Newmetrix (formerly Smartvid.io) blog.
The main issue I wanted to address in my talk was this: The construction industry’s total recordable incident rate and its on-the-job fatality rate have both remained largely the same since 2010, according to the U.S. Bureau of Labor Statistics. That’s a problem, because while just 6% of the US labor force is in construction today, our industry accounts for just under 1 in 5 of all workplace fatalities.
Over that same period of time, however, technology has made enormous advances. Think about it -- ten years ago, the iPhone finally began selling more devices than Blackberry, the first iPad had shipped and you were still watching movies at home using DVDs.
The reason that progress on reducing incident and fatality rates has stagnated isn’t because the industry hasn’t been working hard enough. We've seen the adoption of new tools aimed at increasing hazard awareness, the implementation of leading indicators as measures of safety performance, and the wide spread adoption of behavior-base safety techniques to involve the worker in preventing incidents.
I contend that the problem is data. More specifically, how we collect and connect data.
Data collection and connection
In construction, we often approach data collection in the same way we approach the rain: a costly and time-consuming nuisance that slows productivity and, once it’s finished, is never discussed again. The way we connect data is no better. In almost every organization, data is siloed and segregated so that different systems can’t share information.
Unless we sort out our data issues, construction won’t be able to apply the power of machine learning (ML) and AI to safety, and we won’t achieve the next level of risk reduction: Predictive-Based Safety (PBS). With PBS, organizations focus on worker behavior, just as they do with Behavior-Based Safety. But PBS is much more powerful, because it also harnesses the power of data to identify predictive indicators and unearth which sites are at the greatest risk of an incident in the coming days. As a result, safety managers can take action to prevent incidents before they occur.
How powerful is PBS? In our experience, organizations that leverage PBS have been able to reduce their recordable incident rate by as much as 60%.
But PBS is no cinch to enable. Descriptive analytics is what we’re most accustomed to, thanks to the backwards-looking weekly safety reports we see. Here, we’re looking into the past to explain what happened. If we want to significantly reduce our incident rates, however, we need to look to the future, to determine what will happen so we can take action. And to do that, we need to have not only a lot of data that’s ready for AI to analyze, but we need many different kinds of data as well. So it’s important not just to collect good data, but also to ensure that we can connect to as many different kinds of data as possible.
There are many common data issues, but two deal breakers when it comes to applying predictive analytics involve:
- Missing data, where data is trapped in a system where it can’t be accessed or, worse, isn’t connected in the first place, and
- Master data issues, where stand-alone systems are not tied to master data, such as project or employee IDs, which makes it essentially impossible to connect data tables to other company systems.
To fix these issues, there’s no time like the present, and Newmetrix can help. When we designed our platform, we designed it so that it would be easy for our AI, nicknamed “Vinnie,” to ingest data from a large array of sources, starting with images.
Vinnie has been trained to identify people in images and video — even those who are partially concealed behind a lift of a column — and then determine whether they are:
- Wearing PPE
- Working at height
- Near an exposed edge
- Within six feet of another person or standing in a group larger than 10 (social distancing safety during the COVID-19 pandemic)
- And many more
Vinnie isn’t picky about where these images come from. He’s happy to analyze all the site photos you’re regularly taking, and integrating Newmetrix with Box, Structionsite, Procore, BIM 360 and many more systems is as easy as a few mouse clicks.
These analyses alone can go a long way toward identifying trends and predicting future incidents. But image analysis isn’t the only trick Vinnie has up his sleeve. We also provide a platform that makes it easy for anyone on a worksite to make a safety observation. And while Newmetrix can use this observational data to enable Behavior-Based Safety, it becomes even more powerful when Vinnie uses it along with image data to enable PBS.
When Vinnie is used to his full capacity, the AI can identify which sites are at the greatest risk and provide safety managers with recommendations for lowering that risk.
There’s much more in my presentation. If you’ve got 30 minutes, you can watch it here. Additionally, if you’d like to see whether Vinnie can help your organization lower its recordable incident rate, take a look at our free predictive analytics assessment.