Work
Enabling a travel agent to deliver personalised customer experiences
Operating in the highly competitive online travel sector, a travel agency was looking to build brand loyalty through delivering exceptional customer booking experiences.
We built an automated and predictive model to accurately predict the likelihood of an individual customer making additional purchases, giving them a clearer picture of their customers’ preferences and more confidence in experimenting with new dynamic, personalised strategies.
Customer
An online travel agent specialising in flights with affiliates in over 30 global markets.
Problem
Operating in the highly competitive online travel sector with thin margins, the company is striving to become a market leader by building brand loyalty through delivering exceptional customer booking experiences. To do this the company offers personalised product and pricing offers based on their customers’ search criteria and past purchases.
To deliver even more compelling personalised experiences, the company wanted a much more accurate picture of their customers’ preferences. Currently, it only analyses customer preferences at an aggregate level in a highly manual, repeatable piece of work. The results of this analysis were not always accurate because it could not capture the complexity of individual customer behaviour.
The company enlisted Faculty to help it to go beyond simply grouping customers into segments and move to an automated and more effective predictive model that can understand and accurately predict the likelihood of an individual customer making additional purchases.
Solution
To better predict what products customers are likely to purchase, we built a cross sell propensity model. The propensity model consisted of three Light GBM boosted decision tree models, which used the search query for each customer and assigned them a probability of purchasing the additional products. To do this, we used both the search criteria and the customer’s historical buying behaviour from their login details.
We then deployed the model on the company’s own Google Cloud infrastructure, complete with an API endpoint for future development. We also built a dashboard using Looker, to give users (such as product managers and pricing analysts) full insight into what the model’s predictions were, and what the most important features were to the model when making its predictions.
In addition to providing more accurate results, we used Faculty’s explainability tooling to give transparency to the model’s decision-making. This gave product managers peace-of-mind that the model was behaving as expected, while also deepening their understanding of what factors encouraged customers to purchase the products.
Impact
The cross sell propensity model removes 40% of the error that existed in the manual approach used before by targeting customers based on their individual behaviour and preferences.
The marketing team now has much more confidence in experimenting with new dynamic, personalised pricing and online content strategies as they have a clearer picture of the preferences of their customers.
As the model is also fully explainable, this unearths previously unseen and useful insights for product managers that helps them to understand sales drivers.