Identifying financial vulnerability with unsupervised machine learning by Rebecca Vickery
As we enter a period of rising prices and stretched household budgets, it has never been more important for energy suppliers to provide practical support for customers who are financially vulnerable. At EDF, we have always demonstrated our care and support for customers with any vulnerability. However, this has relied on customers self-reporting or manual scans of customer records.
This year we used data science to introduce a systematic method to analyse Smart meter Pay As You Go (PAYG) customers for signs of financial distress. Our solution uses unsupervised clustering techniques to automatically classify customers so that we can provide personalised recommendations and signposting of support. Alongside developing the model we also partnered with our data engineering and dev-ops team to expand our bespoke big data platform to handle the deployment of machine learning models at scale.
In this talk I will highlight the technical methodology we used to develop a machine learning model to detect customers experiencing financial difficulties. I will cover how this model was deployed through a custom machine learning platform and how the model has been utilised to help some of our most vulnerable customers here at EDF.
Supported by
