A UK supermarket chain with stores nationwide.
The supermarket operates a loyalty scheme, a critical part of its business to attract and retain customers. Its marketing department concentrated much of its efforts on increasing the number of customers signed up to the loyalty scheme. However, simply counting the number of people attracted to sign up to the scheme through marketing campaigns did not provide valuable information, because each customer spends different amounts.
The company typically waited months, even years, to assess the effectiveness of a campaign to recruit high spenders, as analysis relied on sales data. It asked Faculty to build a model that would differentiate between new customers signing up, according to how much they would be likely to spend.
First, we analysed existing data, which showed that half of those holding loyalty cards are low spenders, 30% are medium spenders and one-fifth are high spenders. But this last group accounted for about two-thirds of total revenue. We developed a predictive model based on the data available when a new member signs up, such as year of birth and postcode.
By knowing the address of a new member it was possible to calculate how far they lived from the nearest store, the shop’s size and weekly turnover, the member’s distance to a competitor and the competitor’s size. Once the new member visited a store, the data relating to total value and number of items of their first shop was also collected.
We then incorporated a Support Vector Machine model capable of predicting whether the member would be a high spender.
Results showed that, after a new member’s first visit to one of the client’s stores, the classification accuracy (whether the person was a high, medium or low spender) was 55%. With seven days of data, the model could identify two-thirds of high spenders.
Reducing the waiting time for feedback on a campaign allowed the supermarket to optimise its marketing efforts and to assess and identify the right channels to reach high spenders.