Travis Perkins plc: Using Data Science to Improve Employee Safety

In the bustling world of construction, where heavy materials and bulky equipment are the norm, safety is paramount. Travis Perkins plc, the U.K.’s largest distributor of building materials, knows this all too well. From cement to bricks and plasterboard, their trade-focused businesses handle materials that could pose significant risks to worker safety.

From Travis Perkins plc’s perspective, the challenge was clear: How could they proactively improve health and safety risks in their branches? With their core value of “everyone going home safe and well, every single day,” the company needed a solution that would help to maintain and improve their approach to incident prevention management. 

Plus, Travis Perkins plc knew they wanted an end-to-end platform to improve cross-team collaboration and find visual ways to present data-driven solutions to business teams.

The Spark of Innovation: Incident Correlation

The Health and Safety (H&S) team at Travis Perkins plc had been conducting regular branch inspections to reduce workplace injury risks. However, with limited resources and time constraints, they need to determine which branches should be prioritized for inspection.

The data science team was called upon to create a proof of concept. Their mission was to determine if they could find correlations to incidents and rank branches on a scale of 1 to 10 based on their risk level.

Building the Safety Beacon With Dataiku

Working closely with the H&S team, the data science team began to construct what would become known as the Safety Beacon project (named after flashing beacon lights on trucks). They carefully selected historic health and safety audit data and combined it with transactional information to paint a comprehensive picture of each branch’s operations.

Using Dataiku’s powerful data preparation tools, they cleaned and organized the large volumes of data, one example being a product hierarchy for identifying potentially dangerous items like timber based on weight and handling difficulty. Branch characteristics such as location, urban or rural setting, and the presence of a yard were all factored into the equation.

The team employed percentiles and normalization techniques to develop an algorithm that could accurately assess branch risk. The result was a dashboard in Looker that provided a clear risk score out of 10 for each branch, along with other relevant information.

As the Safety Beacon project went into production, its impact became increasingly apparent. The dashboard evolved into a dynamic tool, allowing teams to track changes in risk scores over time. Monthly emails alert H&S managers and regional directors to increases in risk, enabling proactive measures to enhance safety.

For the H&S team, the Safety Beacon became an invaluable incident prevention tool. It provided new insights to help prioritize resources and identify branches that need additional support. Branch managers and directors gained unprecedented visibility into the safety of their operations, empowering them to take swift action when necessary.

We see Dataiku as an accelerator for our data science and AI journey. Robert Barbour Group and Insights Director, Travis Perkins plc

The Dataiku Difference

The success of the Safety Beacon project was significantly bolstered by Dataiku capabilities, notably:

1. Dataiku made it easy for subject matter experts from the H&S teams to collaborate with and explain their processes to the data science team. This helped to facilitate the cleaning and preparation of data, notably combining historic health and safety audit data with transactional information available in the data lake.
2. Seamless deployment of the risk score dashboard in Looker ensured easy adoption by business end users.
3. Data had not been previously used in this manner by the H&S team. Advanced AI/ML capabilities introduced them to new possibilities in data-driven decision making.
4. The Dataiku platform is responsible for the increase in trust in the models built by the internal data science team. Visualization of the entire end-to-end project (from data prep to MLOps) via the Dataiku flow has also aided model understanding and explainability.

In addition to the health and safety use case, there are other benefits that Travis Perkins plc sees from the Dataiku partnership at large. Previously, with solutions developed by a third party, the team lacked the visibility and control needed to effectively address questions and potential issues. With Dataiku, the data science team is able to monitor their models at different stages and understand proactively if anything is wrong before negatively impacting items downstream. 

Additionally, the integration with Google Cloud Platform (GCP) and ability to host external models were key factors in choosing the platform. With GCP, the team has been able to easily access relevant data in a timely manner. Finally, with Dataiku, the entire process to run their models and have them be visible and accessible takes five minutes, a change ushered in with automation and scenarios in Dataiku. Previously, they had to send their data to a third party, load it, try again if it failed, and so on. Now, the team has strong governance over the data that goes into their data science models.

Keeping people safe and well is the Group’s number one priority, so we are continuously looking for new ways to improve our approach to safety and reduce incidents. This new solution allows us to bring a unique data-driven approach based on years of knowledge to the forefront of safety, to drive positive and transformative change, not only for our group but potentially also for the wider construction industry.

-Lauren Cramp, Travis Perkins Health and Safety Director

The Safety Beacon project enabled data-driven insights to enhanceTravis Perkins plc’s’ approach to workplace safety. By harnessing the power of data and the Dataiku platform, they’ve created a proactive, intelligent system for identifying and mitigating risks. The team is excited to continue to drive augmented decision making in their other use cases with Dataiku, including fraud prevention, churn, and customer segmentation.