A Belfast-based Risk & Compliance software provider has been collaborating with the Health and Safety Executive (HSE) and construction giant Costain as part of an ongoing project to unlock artificial intelligence’s (AI) potential in improving the management of risks on worksites.
AuditComply, the winner of c has recently completed a three-month pilot that aims to deliver ‘better evidence-based management of common safety risks’ such as work at height and utility strikes by comparing HSE data collated from returns under the Reporting of Injuries, Diseases and Dangerous Occurrences Regulations 2013 (RIDDOR) with Risk Assessment Method Statements (RAMS) supplied by Costain.
The Risk & Compliance software provider was approached by SafetyTech Accelerator, which runs the SafetyTech Challenge on behalf of Discovering Safety, a partnership between the HSE and Lloyd’s Register Foundation, last year to pitch an AI solutions proposal to resolve the challenge. The theme for 2022 was ‘to explore the risks from taking shortcuts on a worksite and how they can undermine compliance with rules in RAMS’.
As the Discovering Safety website explains, AuditComply applied its AI-based risk management platform to the challenge and imported RAMS and RIDDOR reports using a template feature that could be configured for different document types.
AuditComply’s industry partners then reviewed the data, comparing both the RAMS and RIDDORs and flagged up areas of non-compliance where operatives had failed to follow the RAMS and taken shortcuts that led to different incidents that ranged in severity.
As Kevin Donaghy, AuditComply’s CEO explained, the pilot’s success was down to the industry partners taking the lead and guiding the software provider throughout.
‘Costain and HSE were the experts that we could work with in the field to understand what the real problems are that they see on a day-to-day basis and how we then tailor our capability around that,’ he said.
‘They provided us with consultants to walk us through real scenarios so we could develop end-to-end solutions. How would that work and where were the real risks within it? By identifying where the real risks lay – the hidden risk is what they all talk about – how do we then translate that into something that we screen and look for and confidently predict?’
Before the pilot commenced, the collaborators needed to undertake some preparatory work to ensure the software provider’s AI-based risk management platform could be successfully applied and valuable lessons could be learned and taken into the next phase of the collaboration.
‘There is a huge challenge around this when you look at projects like HS2,’ said Donaghy. ‘You have thousands of contractors submitting different types of documents and these can be from photographic PDFs through to word documents. How do they come up with a consistent way of being able to screen them and look for the risk within that? Typically, the risk is buried deep within these documents somewhere; sometimes it’s a lack of documentation where the risk might reside.’
To overcome this, AuditComply needed to draw on a known dataset, build an AI module to screen this data and then run tests so that they could confidently predict the likely risks within the data. To enable this to happen, the HSE provided 10 years’ worth of anonymised RIDDOR data while Costain contributed the RAMS.
‘We would have a library of RAMS that had to be anonymised from different sources,’ explained Donaghy.
‘What was really difficult about that was, when you had a RAMS from a particular source, that RAMS was kind of useless if it wasn’t connected to an incident or an event. If that RAMS looked good, but we knew it had resulted in a particular accident that was only one part of the challenge.
‘First of all, you had to go one step back and look at your RIDDOR reports and then look at which RAMS are associated with those RIDDOR reports. In other words, where did this accident happen? What did the investigation tell us? What were the RAMS as part of that?’
Donaghy addde that even though they could identify RAMS that had resulted in an incident, it wasn’t possible to say explicitly that the failure to follow those RAMS was the sole contributory factor.
‘However, with a certain degree of confidence, if you collected enough of them, then you could say, “Okay, we know these RAMS are slightly deficient so there’s something in that”, versus the other RAMS that never resulted in accidents.’
Typically, many of the incidents identified involved work at height and a failure to follow the RAMS instruction to wear a harness. Another was ignoring RAMS instructions to use hand-digging near utilities supplies and using machine tools instead which resulted in an electrical strike.
Importantly, the data screening using the AI tool enabled the collaborators to grade the RIDDOR reports and start to identify high-priority incidents, says Donaghy.
As the Discovering Safety website notes, the results of the pilot have identified the need for further work, for example, RAMS may need to be made easier to use and fit for purpose for those working on site to minimise, and ideally avoid, non-compliance.
AuditComply is meeting up with the HSE and Costain to discuss the next steps in their collaboration, which will include identifying the common features of RAMS that are poorly used so companies can better design them and encourage all worksite operatives to comply and avoid taking shortcuts.
‘Most companies when you ask them for their credentials for a particular project will upload their certifications and RAMS,’ said Donaghy.
‘When they are uploaded into a system, if you had a human trying to review all of them in a consistent manner, it’s really difficult. We are trying to screen it and take a first pass with the AI machine to give us a level of consistency and also raise any concerns.’
Reflecting on the technology’s future potential, Donaghy says businesses will most definitely achieve scale if they can employ AI and machine learning because it significantly reduces administration and the many man-hours that are required to screen data.
‘There is a massive amount of opportunity with it, but it has to be bite-sized and it is useless if you don’t marry the technology solution with the industry expertise,’ he says. ‘There will always be that human oversight’.