The airline industry is characterized by extremely low profit margins and exceptional sensitivity to political, economical and social factors. Thus, planning is everything. But even if airlines are suffering now like never before due to the pandemic, the situation was not necessarily easy before that and industry leaders were looking towards new technologies.
Predictive analytics powered by deep learning has opened new opportunities for organizational optimization within the airline industry. Although no one could have foreseen the dramatic impact the coronavirus has had on air travel, airline executives are realizing that better planning is not just nice-to-have but absolutely essential for survival.
Predictive Demand Intelligence
Airlines have been forecasting the demand for a long time, so that’s nothing new. What is new, however, is the accuracy. Demand forecasting has evolved significantly the past 20 years or so, but the real development has come with artificial intelligence, particularly deep learning techniques. Using this powerful tool has taken demand forecasting accuracy to a new level and it’s saving organizations millions of euros.
You see, all organizational planning is based on prospective demand: How many flights should be taken during a certain period, size of plane and number of seats expected to be sold, how much luggage is expected and how heavy is it, how much staff is required, onboard food and merchandise etc. That means that forecast accuracy does have major financial implications. Airlines are therefore fighting to improve their forecasts as every percentage in accuracy has a significant impact.
By analyzing historical data using sophisticated deep learning techniques, forecasting models can be built that are far more accurate than the industry’s traditional methods. But it doesn’t stop there as a number of external information can be fed into the model to improve the accuracy even more; international and national holidays, unemployment rates, interests rates, exchange rates, policy changes, and even major events such as trade fairs.
As an example, there are 32.000 exhibitions held annually with over 300 million visitors where most arrive by air. On a European level, the mapping of exhibitions alone on the continent can improve the accuracy significantly. As an example:
The Seafood Expo Global (which has usually been held in Brussels but has been moved to Barcelona) attracts 2,000 exhibiting companies from 89 countries and around 30.000 visitors.
Motek Stuttgart, is the world's leading event in the fields of automation in production and assembly, attracts 1,000 exhibitors from 25 countries and 35,000 visitors from 81 countries.
The leisure, hobby and entertainment industries alone have a total of 548 events across Europe every year.
Events like this are major business for airlines and, obviously, adding this external data to the forecast model will improve accuracy significantly and allow for better planning. During times like these with unparalleled uncertainty, the need for clarity has never been more, and some European airlines are already ahead of the pack saving significant amounts of money using sophisticated predictive demand intelligence.
But the use of predictive analytics can be used within almost every area of an airline, and an untouched area is the customer-sensitive practice of overbooking flights.
Airlines overbook flights to avoid empty seats from passenger no-show or missed connections. This has long been practiced as this allows airlines to ensure planes are at capacity in order to maximize profit, which is quite understandable considering the low-profit operating model.
However, this does come at a price. It frequently happens that more people show up for the flight than the airline anticipated, which means it’s not a seat for everyone. This creates bad customer experience as frustrated passengers are turned away. Likewise, even if flights are overbooked there might still be empty seats. This is the result of over- Vs under-forecasting, and can certainly be improved with more accurate forecast techniques like discussed here above. But by using advanced predictive analytics airlines can predict the rate of no-show, which will allow them to overbook more accurately.
One can imagine that the passenger profile between London and Berlin on a Tuesday morning in February is different from the passenger profile between Copenhagen and Alicante on a Friday in July. People are traveling for different reasons, business or pleasure, with different amounts of luggage, and far different amounts of money spent on the whole trip up-front.
This simple example tells us that the level of no-show can differ greatly between months of the year, day of the week, time of day, destinations etc. Maybe, the level of no-show between London and Berlin on a Tuesday in February is different depending on if the flight is at 7am or 11am. It probably is, and the implementation of Predictive Overbooking mechanism can help airlines overbook more accurately; sell more seats where no-show is predicted to be higher, or overbook less where no-show is predicted to be lower. The accuracy of this technology is staggering as it has a completely different prediction model for each and every flight. This will optimize capacity, increase profits, and certainly reduce painful overbooking where frustrated passengers are turned away.
As much of this technology is rather new, many airlines are far from utilizing all the opportunities predictive analytics provide. But if the pandemic has taught the industry anything then that’s the importance of constant efforts to stay competitive, which is certainly reflected in the surge of AI usage among leading European airlines.