Call Centre Forecasting
Predicting call centre volume by day and within 30min windows
Challenge
Energy client had a team producing 7-day and 21-day forecasts manually
Challenge was to automate and beat the forecasting accuracy
One constraint was that this needed to be a fail-proof stand-alone tool that a single person could run on a virtual machine
Collected the datasets and created data preparation and feature engineering (lags, interactions, etc.) pipelines
Automated feature selection using machine learning techniques (Boruta, RFE, etc.)
Automated model training, evaluation and selection for both 7-day (overall and within 30min windows) and 21-day, pitting multiple models and parameters and picking the best one every time
Created a simple-to-use command line prompt for the end-user to run forecasts themselves
Solution
Value
Reduced time spent preparing the forecast from 2 days down to 10 minutes
Simple end-user steps to produce a new forecast (append new data, version the data and run the tool on the updated dataset)
Trained the data science team on model details for any refinements or future work required
Roles
Data Scientist, Data Architect
Tools
Sector
Energy & EVs