Identifying bookings with a high-risk of host rejection to improve customer service and customer satisfaction

Project Details


Fall 2017


Arturo Heyner Cano Bejar, Tonny Kuo, Nick Danks, Kellan Nguyen





AsiaYo! operates in a highly competitive industry serving as intermediary and agent between accommodation providers (hosts) and accommodation seekers (guests). High customer service is a critical business goal for AsiaYo!. Rejection of bookings by hosts can cause customer dissatisfaction and potentially loss of customers. Currently, AsiaYo! experiences a host rejection rate of 15% of orders. Rejection of a booking by the host triggers a reaction from the AsiaYo! customer service team in which they contact the guest, offer alternatives and provide solutions to resolve the rejection and convert it into an alternative booking. This is a re-active process and only occurs after some time when the host has rejected the booking.
The goal of this analysis is to provide a tool for the AsiaYo! customer service team to proactively intervene on bookings with a high risk of rejection at the time of the booking. This will be achieved by providing a probability of rejection for every booking made. High rejection risk bookings can then trigger a customer service team intervention allowing for fast, efficient, and proactive customer service response.
The data used in the predictive task is generated from the booking transaction and includes the number of guests, the number of nights, the number of rooms, the amount paid, the day of week for check-in date, the day of week for booking date, and the days-in-advance period from booking date to check-in date.
The key challenge of this analysis was the correct prediction of rejected orders – that is predicting a transaction as highly likely of rejection and it indeed then resulted in rejection. To this ended we used several classification algorithms and statistical analysis to arrive to a best possible recommendation. Results from our analysis are promising. However, their interpretation must be handled with care. Our performance metrics, sensitivity (0.69) and accuracy (0.53), lead use to best results when using the Naïve Bayes. Recommendations are to implement the predictive algorithm right after a booking is done, update the algorithm periodically to compensate for time, and input more variables that potentially lead to better performance such as behavioral data.

Application Area: