As an Insight Fellow, I recently completed a consulting project for Valor Water Analytics. The business problem is this: utility companies deal with substantial time lags between dispensing a resource (e.g., water) to customers, and receiving payment. Customers whose payments are up to a month late receive a late fee. If more than a month passes after the due date and the customer still hasn’t paid, the utility typically cuts off the customer’s water supply. By that point, the customer could have continued using water for up to 3 months without paying for it. Thus, cutoffs are not only bad for the customer, they represent up to 3 months of lost revenues and lost water for the utility. A utility would therefore benefit if it could predict impending cutoffs and avoid them, for example by working with the customer to agree on a payment plan.
My project goal was to create a tool that could predict impending cutoffs from the data available to a utility: customer metadata (location, customer type, meter size), customer billing history (including which payments were late), and customer usage history. It would query the utility’s database, clean the data and compute features, fit a model, make a prediction (on a daily basis), and display the results in a dashboard.
The finished product is CutoffPredictor. More details to come…
Ted Bohn's Data Science Blog 