Analyzing the value of bus charter requests
- Sponsored by: FlixCharter
- Project Leader: Dr. Ricardo Acevedo Cabra
- Scientific Lead: Dr. Berit Johannes
- Term: Winter semester 2019
At FlixCharter we rent out whole buses from our partners to our customers. When customers ask on our website for a price, they give us information about their trips, such as departure time and location, arrival location, and number of passengers. Based on this information (plus other factors such as seasonality etc.) we calculate a price and provide an offer instantly.
When a customer then reserves or books with us, we collect many more data, such as customer type (school, sports club, company ,…), when was the request reserved or booked by the customer, whether it has changed, and whether it has been cancelled or declined, we have a cancelation rate of >20%. Between selling and sourcing the requests, we combine some of the trips to tours and assign these tours to buses that have been rented very far in advance (called rental orders) to reduce idle time and empty kilometers (and thus to create value and to increase margin and supply).
Therefore, trips of a customer request can either become part of a tour or be matched directly with a bus partner (also called “sourced 1:1”). The earlier the trips are sourced 1:1, the lower the sourcing cost will be. On the other hand, the longer the trips stay in the pool of trips available for optimization, the greater the optimization potential will be. Additionally, changes by the customer and cancelations often destroy the created tours and result in inefficiencies.
We would like to evaluate customer requests and trips with respect to their value for optimization, that is, we want to know when and for how long we want to put trips aside to be used for the optimization team to build efficient tours and to fill the rental orders, and when to release them to be sourced 1:1.
Attributes and data that could be computed and considered, are, for example, time between booking and sourcing, cancellation and change probability, uniqueness of trip, location of rental orders, departure date, pre-booking time, customer type, price, number of passengers, and distance traveled.
The ML/data analysis model could be run alongside with the optimization algorithm, to decide every day, which trips to include in the 1:1 sourcing pool versus the bundling pool.