Artificial Intelligence and Machine Learning (AI/ML) is an integral part of everyday life for most of us, whether we realize it or not. If you’ve ever scrolled through social media feeds on Facebook or LinkedIn, streamed movies on Netflix, or bought products from Amazon, you’ve manipulated machine learning algorithms.
AI/ML technologies have been widely used in the aviation industry for decades and have a rich and growing history. Many of the most important sectors of modern airlines use AI/ML technologies at scale. Examples include airline network planning, flight scheduling, pricing and revenue management, operations and crew planning.
AI/ML models have had a major impact on the aviation industry, but early implementations had limitations. The previous architecture and development process made it difficult to integrate different models and required significant manual intervention to keep the models current with the latest data.
Our modern ML-based application uses a modern cloud-based microservices architecture with increased scalability and automation, providing our airline partners with additional revenue generation and improved customer service opportunities. Here are five ways this technology is being used.
1. Dynamic pricing and availability of airfares
Airline dynamic pricing is a term used in many contexts. Given airline revenue management controls and data signals in today’s competitive marketplace, Saber has developed an airfare dynamics model to illustrate the engine that helps airlines provide tailored offers to different customer segments. I am referring to reasonable pricing.
Closely related to dynamic pricing is dynamic availability. Dynamic availability updates an airline’s availability control (by closing or opening a booking class) instead of changing prices directly. The airline’s ability to change prices in real-time using dynamic pricing provides maximum revenue performance, but currently it is limited to the airline’s direct channel and his NDC offers. On the other hand, dynamic availability changes can be applied to all GDS sales channels, resulting in slightly lower revenue performance for each change, but higher total applicable sales base. Both approaches therefore add value to the airline.
Saber’s dynamic pricing and availability products have been shown to deliver up to 3% increased revenue. These profits are incremented by what the airline earns from its revenue management system, resulting in more customized offer determination, continuous price points instead of fare classes (see diagram), and more granular pricing control. arises from the ability to (that is, in a round trip) to level instead of departure date for traditional inventory management only). This fine-grained and ongoing pricing will also benefit travelers as it will give them access to new price tiers that were previously unsupported. For more information, see Saber Air Pricing IQ and Saber Dynamic Availability.
2. Dynamic pricing with air travel
One of the biggest trends in airline marketing over the past decade has been the strong and steady growth of air travel ancillary sales. According to his 2022 report for IdeaWorks, trends in airline ancillary revenue since 2013 show an estimated average annual growth rate of more than 15%. Increased ancillary revenue during this period was driven by both discrete ancillary revenue (a la carte) and the prevalence of airline-branded fare bundles.
Historically, ancillary prices have been a secondary consideration for airline marketing teams, and these prices were typically fixed across airline regions. But as sales (and revenue) shift to more incidental, there are more opportunities for more dynamic pricing using AI/ML technology. The AI/ML model is designed for airlines to use supervised learning (based on estimated price elasticities and factors such as marketing segments, seat types, and in-flight locations) and reinforcement learning techniques (see “Experimental Engine,” below). “section).
In fact, our customers have seen up to 10% increase in ancillary revenue with both dynamic price reduction and ancillary cost increase with the use of AI/ML models. For more information, see Saber Auxiliary IQ.
3. Experimental engine
Our new experimental engine platform based on a form of AI/ML known as Reinforcement Learning presents exciting opportunities for all travel sellers. Traditional AI/ML models build complex predictive models based on historical data. However, the experimental engine does not rely on extensive historical data. They try multiple options (i.e. experiment) and observe the results. As the model learns from those results, the parameters that produce the output are slightly adjusted and the process can be run over and over again.
One use case is Saber Hospitality’s SynXis Retail Studio application for hotels. It has many similarities to airline retail applications. Hotels have a wide variety of accessories and choosing the right one for each type of guest is a complicated matter.
The graph below shows how the Experimental Engine model learns which products to offer to two different customer segments A and B. Initially (see left side of graph), the model offers all three products. Over time, we settle on a clear solution: A prefers breakfast and B prefers higher floors. As the model learns as it gets more data points (from impressions and clicks), it becomes more confident about the best products to show and reduces the amount of “exploration”.
Hotel Extra Offer Optimization Example – Automatically discover the best customized offers for each customer segment.
In fact, our hotel guests have seen up to 30% higher guest click-through rates when using our experimental engine to determine supplemental offer display, compared to a uniform display strategy. For more information, see SynXis Retail Studio.
4. Market size forecast
Two of the most important questions in airline network planning when designing future schedules are understanding “Where do people want to go?” “How many people want to travel between a pair of origin and destination?” Knowing both answers will give you travel demand, so you can optimize travel supply (i.e. how many planes you need, where you fly , determine how to schedule). This means you can build systems for estimating expected revenue and profitability, fine-tuning retail and distribution strategies, planning resources, and more.
Accurately answer “How many people want to travel from city A to city B?” This is important for aircraft manufacturers, airlines, hoteliers, travel agents, local businesses for tourism purposes and the investment community. There are approximately 250,000 city pairs in the world with air service, and the demand for travel between those cities changes every month. Therefore, it is virtually impossible to answer that question with manual market analysis.
This is where Saber’s new network planning tool can help. The Saber Labs team is working on an AI/ML model known as the “Market Potential Forecaster”. The model predicts how many people will move from A to B each month for all 250,000 markets over the next 12 months, and updates such predictions regularly. The underlying AI/ML model is trained on both historical trends and forward-looking air travel low fare shopping shopping data. Saber’s Cuneyd Kaya goes into more detail in this recent article.
AI/ML-based market potential forecasting models use a combination of historical data, analysis of segment behavior, and forward-looking shopping data to increase the average forecast accuracy of customers using our solution. It can go up to 90% or more.
5. Suspension of rescheduling passenger flights
Flight cancellations and delays are among the most common disruptions faced by airline passengers, but other types of disruptions exist. Saber’s AI/ML models for realignment during irregular operations (called IROPS), such as unplanned flight schedule changes and cancellations, can be applied to global seat availability, passenger prioritization, airline Providing a new itinerary that takes into account customs helps in such scenarios.
These AI/ML models can also be used by airline analysts as a “what-if” tool to assess the impact of certain disruptions (for example, changing aircraft to smaller-seat planes). Based on the resolution, you can decide if this is an action you can take. Once the tool has been used and the results applied, the rebooking process will take care of this so passengers are automatically guaranteed tickets on the new itinerary.
Each time the IROPS application is run and the passenger is notified of a new itinerary, the passenger must decide whether the new trip is still one they would like to participate in, reschedule or cancel altogether.
A businesswoman who has an important meeting on Monday morning would cancel rather than choose another itinerary that arrives Monday noon. In Italy she would accept rebookings for those taking a week’s vacation to arrive half a day late. An experienced analyst can recognize some of these patterns and only offer re-accommodation options that are likely to satisfy travelers. To further assist aviation analysts, Sabre has developed a machine learning model to help assess whether alternative itineraries are acceptable.
In fact, some airline customers have seen significant improvements in traveler satisfaction and up to 25% reduction in call volume. For more information, see Saber IROPS Re-Stays.
Choosing the Right AI/ML
Not all AI/ML is created equal. That’s why Saber Travel AI’s cloud-native, microservices-based machine learning application is powered by Google’s Vertex AI, the best-in-class integrated AI platform for offers and orders. In the airline industry, making the right decisions in real time can mean the difference between a profitable offer or not. Advanced experiments with Vertex AI are empowering airlines to create personalized and contextual offers for travelers, and those offers deliver and deliver value to both travelers and airlines. more likely to generate.
