A travel company specialising in coach holidays, with over 450 local pickup points nationwide.
When it came to analysing customer behaviour, the company knew that it had reached the limits of what a traditional ‘Recency, Frequency, Monetary Value’ (RFM) model could do for them. While the model was relatively successful, the company needed a more sophisticated method of identifying a customer’s tendency to rebuy, and the resulting value of targeting them with direct marketing materials. This was particularly difficult to achieve for customers with more than one historical purchase.
After conducting a deep exploration of their customer data, we found that the company could use machine learning in their marketing. With the right tools, they could analyse more complex or unintuitive features of customer behaviour, and then use these insights to understand if a customer was likely to respond to a particular mailing. We were engaged to help build and deploy this machine learning model, with the goal of reducing the cost of mailings and increasing conversion.
We replaced their old Excel-based process with a bespoke, automated piece of software which extracted a much wider range of features from the raw data. These features – including type of trips bought previously, catalogue contents, time since last interaction, and time since last booking and trip starting – were used to train the machine learning model.
Replacing their old processes with machine learning software allowed for a much more nuanced understanding of customer behaviour, accounting for interactions of multiple features and making predictions on the individual customer level instead of across segments.
This allowed for a much more nuanced understanding of customer behaviour, accounting for interactions of multiple features and making predictions on the individual customer level instead of across segments. For example, if two customers purchased the same beach holiday, the model could predict that one customer would be most likely to repurchase only beach holidays, while the other would seek variety and respond best to catalogues promoting new trip types, such as city breaks.
Using this information, the model created a probabilistic score for each customer, based on their likelihood to respond to a mailing with a purchase.
Across all of the company’s print marketing, our model allowed the company to create 4% more revenue with no increase in marketing spend. The model drove even more substantial results on certain marketing campaigns for individual products, in some cases decreasing the cost of filling a seat by up to 50%.
As a result, it’s now significantly easier for the company to fill undersubscribed coaches at below breakeven cost – an important benefit, considering the significant cost of running coaches with large numbers of empty seats. These advancements will allow the client to more confidently explore new product offerings, assured of its ability to market them cost-effectively to the right customers.