Applying proven Machine Learning techniques to optimize airline revenues
Optimizing pricing across sales channels with modern revenue management practices has long been a challenge for airlines. But the industry´s gradual transition to retailing with offers and orders will finally enable greater pricing flexibility.
While dynamic pricing gives us the ability to quickly change price points, overcoming the limitations of the 26 RBDs to optimize core and ancillary product pricing rests with continuous pricing. Many systems, processes, and practices will need to change to reach the desired end-state but how can airlines begin adopting modern pricing techniques to increase revenues along the journey?
The key to optimized pricing is understanding a customer's willingness to pay while also taking dynamic market and business factors into consideration. Historical booking data combined with real-time signals help us strike that balance. But with the abundance of data and sources available, a helping hand is needed to analyze and process data patterns, intelligently. And this is where Machine Learning opens a whole new world of opportunities.
By analyzing swathes of blended data, Machine Learning can help us predict a customer's willingness to pay for a product at different price points. The optimal price can be determined based on predicted demand, available inventory at any point of time, and broader contextual factors.
Demand prediction
Historical booking data, contextual factors at the time of booking (like seasonality or market forces), and trip motivation (like leisure, business, family) provide insights into a customer´s willingness to pay at different price points. Additional attributes that affect demand must also be considered, like booking lead time or round trip versus one-way tickets. The data in each of these subsets is grouped based on the unique combination of these contextual attributes to which a formula is applied to plot demand against different price points with the help of Machine Learning-based prediction techniques, like linear and polynomial regression. Furthermore, external factors such as events, weather, epidemics, fuel price, or competitor prices, can impact real-time demand so adjustments to initial forecasts may be required. By comparing the real time demand against the initial forecast, adjustments can automatically be made to the predicted demand.
Revenue maximization
Once potential demand at each price point is estimated, calculating total potential revenue at each price point is straightforward. But before calculating the total revenue, available inventory must be considered. After considering over-booking allowances, if predicted demand is higher than available inventory, demand is capped.
After we calculate the total revenue for each price point, the fare at which the airline will realize its maximum total revenue is recommended. Minimum and maximum limits for recommended pricing can be set by the revenue manager.
Applying Dynamic and Continuous Pricing to flight tickets
Once we determine the price that can provide maximum revenue for the airline, we need to apply it to the flight ticket. This can be done in various ways.
True continuous pricing
In the direct and NDC channels, optimized pricing can be applied directly to flights. However, significant discrepancies may arise if the same products are sold in the indirect channel based on RBD fares. In the worst of case, fares in the direct/NDC channels could even be higher than fares in the indirect channels, putting them at a disadvantage.
Continuous Pricing within RBDs
To overcome unintended pricing differentiation across channels, Continuous Pricing can be applied within each RBD. Pricing parameters can be set for each RBD, ensuring that maximum price-points are never higher in the direct channel than the indirect channel, for example.
ML-based inventory control
In addition to set RBD pricing controls, ML-based inventory control can also be applied. Based on the price recommended by the Machine Learning engine, RBDs can be opened or closed so the indirect channels can also match ML recommended pricing.
Don´t wait for utopia
New standards, processes, and technologies have historically taken a long time to deploy in the travel industry. But airlines don´t have to wait for the full deployment of retailing with offers and orders to optimize pricing. Machine Learning can already help overcome long-standing barriers to Dynamic and Continuous Pricing to optimize airline revenues across channels.
John has over 20 years of experience in designing technical solutions and frameworks across several product portfolios at IBS Software, playing a leading role in migrating solutions to new generation technologies and microservices architectures.
In his current role as Enterprise Architect for Airline Retailing Solutions, he leverages state of the art technologies, including machine learning and container orchestration, to help airlines improve the way they sell and engage with travellers.