A car auction business with one of the largest databases of vehicles sold in New Zealand was seeking better insights into the buying and selling of vehicles. It wanted to create a competitive trading advantage through research and considered that this might be achieved through the development of a statistical model. While they typically relied on industry experts for valuations, they wanted a way of spreading expertise across all retail sites and to validate human-made estimates.

KPMG worked with the client to understand the economics and dynamics of the used vehicle market in New Zealand. We analysed vehicle sales data and conducted feature engineering to create a set of over 400 attributes relating to the value of a used vehicle. We developed machine learning models to create predict valuations using a different combinations of known vehicle attributes.

KPMG worked alongside with the client to compare predicted results with expert valuations so the model could be tuned and improved.

KPMG’s models improved the accuracy of valuations by 30% overall and by over 50% for key vehicle categories.

KPMG delivered workable code, documentation and a quick-access, intuitive static lookup table to allow the client to continue to run and develop the model.