Background and objectives

Our client is a leading UK water & waste water company serving roughly 2.5m customers. It has recognised the potential value of using predictive analytics / machine learning techniques to a) predict the likelihood of collecting debt from customers and; b) predict the Next Best Action to avoid customers from defaulting and collect the debt. Although it was uncertain whether the prediction of the debt provision would support the argument to reduce the debt provision, it was generally believed that the current provision was several £m too high. swhere was selected to conduct a proof of concept for its extensive experience in the water & waste water sector and its track record in predicting mass customer value drivers such as debt and churn.

 

Our approach

Together with our client, we decided to first predict the provision based on 5 years of customer interaction data incl. billing and payment data. The data sets we used are very common to all water & waste utilities and which allows us to hit the ground running and develop a baseline model relatively quickly.

 

Results

In extracting the relevant features, our approach identified a relevant ratio which can be used for the extrapolation of collection performance beyind 3-5 years: the ratio of remaining balance after X years over the balance after X-1 years. Using these ratios and predictions for the first 3-5 years, our methodology allowed our client to reduce the bad debt provision by multiple millions. Our client continues to use our model for the Provision prediction as a SaaS to sustain the competitive debt provision level.