In 2011 the New York Times reported that the Santa Cruz police department arrested two women in a downtown car park. One for possession of illegal drugs, the other had outstanding warrants, both were peering into car windows when the police arrived.
What was unusual about the incident was the fact that the police were alerted by a computer program that predicted a high likelihood of car burglaries in that area on that day. This predictive policing was based on the previous 8 years of crime data recalibrated on a daily based with new crimes and associated data.
The Santa Cruz police department may have been an early adopter within policing but insurers and other B2C companies were deploying predictive analytics at least 5 years prior to this for fraud detection and customer churn among other things.
Indeed, back at the beginning of 2020 we highlighted how predictive analytics is being used to eradicate polio in the last three countries with outbreaks today (Nigeria, Afghanistan and Pakistan) by predicting where the disease will next occur and dispatching vaccinators to those areas before the outbreak occurs.
Might this ‘last mile’ eradication of polio and the countless other examples of the use of predictive analytics be emulated in health and safety to enable the ‘last mile’ in safety performance – breaking through the statistical barrier that is 0.45 fatalities per 100,000 workers?
There are two components to predictive analytics, data and technology. The technology (algorithms, software programs etc.) have been around for many years and successfully deployed for at least 15 years. The challenge to deploying predictive analytics to improve safety performance is the data. Where does the data come from?
Data could be from incidents, i.e. data you already hold. However, collecting more information about any future incidents will help inform the models. Not just additional information about the cause (from what height, for example) but also what may seem to be unrelated data (such as weather conditions, time of day or shift patterns). Then there are near-misses. The more data that can be collected, the better the predictive models.
Data from safety inspections and safety observations are also great sources of ‘fuel for the model engines’. The great thing about predictive analytics is that the more data you have, whether directly related to incidents, or seemingly unrelated, the better. You never know what combinations of events can be the trigger to predict an incident. From personal experience in the mobile industry, customers calling to ask for a new battery (this was a while ago) was the flag that they were about to leave the service and a trigger for the call-centre agent to make an offer aimed at retaining their custom.
The good news for e-permits customers is that data on buildings, contractors, work requested, approved (or otherwise), and closed out, together with all the related RAMS and a full audit trail is available within the database. Indeed, for non e-permits clients reading this, the availability of contractor data together with a contractor’s likely familiarity with the system (acquired through use with other e-permits clients) is a major driver for new e-permits clients to adopt the system.
There may be an element of ‘model-weariness’ at the present time, but this should not prevent those with a Duty of Care over workers exploring what is a tried and tested solution elsewhere to see if it can be deployed to break through the statistical barrier that is UK fatality rates and improve the safety performance of industry as a whole.If You Like This Post, Please Share It!