Dynamic Pricing with Reinforcement Learning

Challenge

Pricing in the airline business is one of the key levers for generating monetary advantages due to its dynamic competitive structure. Pricing individual legs or entire routes is a highly complex task, which today is still largely managed by humans or semi-automated logic. Our client, an international airline, faced the challenge of wanting to fully automate the daily pricing of individual segments. An innovative approach was sought to achieve this.

Approach

To address the challenge, we collaborated with the client's revenue management department to develop a reinforcement learning-based approach that can autonomously manage pricing in the time window up to 365 days before the departure date. The AI system was trained in a simulation environment using existing control data to learn the behavior of manual pricing strategies. A wide array of data points was utilized, representing the current market conditions and influencing the system's pricing decisions.

Result

The result of the project was a fully autonomous pricing control system that can independently manage price developments within a segment for continental flights without human intervention. The agent is capable of making independent decisions about increasing or decreasing the bid price based on the current market situation and the booking status. The system anticipates the demand development over time across different booking classes.