A British luxury car manufacturer.
As a luxury car maker with a very small addressable market, the company is dependent on customer loyalty to drive repeat sales. In this case, the value of a loyal customer who buys more than one car is around three times that of a customer who buys only one car.
Faculty was charged with building a data-driven model that could identify existing loyal and non-loyal customers; could accurately predict which non-loyal customers were most likely to become loyal in the future; and, for existing loyal customers, could predict which car they were most likely to buy next.
After reviewing the company’s data set, we highlighted the need to incorporate some less obvious features. This included days from the model launch to purchase and, for loyal customers, the time between the purchase of the first and second cars.
We then created a supervised machine learning classifier that distinguished between loyal and non-loyal customers. We tested a number of classification algorithms, before concluding that the best performer was the Support Vector Machine algorithm. This correctly identified four out of five loyal customers and three out of five non-loyal customers in a previously unseen test.
Once we had identified to a greater degree of accuracy the loyal customers, we used a random forest to determine which of four available models they would be most likely to buy next.
Our methodology for predicting new loyal customers has the potential to earn this car company millions in extra sales by allowing it to focus its customer care resources on those customers who are most likely to buy again.
Our model is now being used by the company’s European sales and marketing teams to improve the effectiveness of its future loyalty marketing campaigns, and there are plans to roll it out globally.