How Fetcherr uses Generative AI to drive hyper-efficient airline pricing

Fetcherr
Fetcherr

Whether you believe that Artificial Intelligence (AI) is set to take over the world by this time tomorrow or that we will have to wait a little longer, one thing appears to be certain: this world-changing technology has already started to disrupt those industries that rely on fast-paced dynamic pricing. 

The example of what happened in the financial markets looms large here. 

In a matter of a few years, algorithmic trading took the financial markets by storm. And that was even before ‘AI’ became the buzzword on everyone’s lips. 

Some quantitative finance pioneers struck gold by entrusting decision making to ‘black box’ smart models, capable of placing trades autonomously after being fed huge amounts of data. 

Most importantly, though, they brought about such a radical transformation of the way of doing business that nowadays, the majority of trades in most financial markets are executed by machines with no human intervention whatsoever. 

Could the same thing happen to airline pricing? 

Air fares do indeed have some similarities to securities pricing.  

In addition to being dynamic, price formation is the result of a complex optimization of multiple variables. Any model trying to get to a profit-maximizing fare must take into consideration a huge amount of dynamic data, from seat inventory and competitors’ actions to political, business and weather events affecting traffic flows. 

To the credit of the airline industry, it has always believed in the science of revenue optimization, adopting dynamic pricing quite early on.  

However, the architecture of the present-day query-based systems on which, at present, most of the industry relies was developed many decades ago, and this shows.  

Most legacy systems operate using a defined set of rules that project prices based on past events. This is an approach that enables airlines to update their fares several times per day. It is far from being effective when circumstances don’t match past behaviour, though, and further still from the pricing efficiency of other markets, for example the Nasdaq. 

This is suboptimal for the airlines, and it is not for lack of technology options. 

But what if you could train a system with enough data so that it always gave you the optimal fare for each specific moment in time, in a continuous, non-discrete manner? 

It is at this point that Generative AI enters the picture, in the form of Large Market Models (LMM). 

You may be familiar with AI-powered generative AI tools such as ChatGPT or DALL-E, which are able to generate content and data (hence their name) in response to a user-entered prompt.  

There is usually not one single correct answer for a prompt asking to generate a text or an image. When you ask the system to provide a specific, quantifiable answer to a complex business optimization problem, though, things get interesting.  

This is what airline tech startup Fetcherr is doing for a handful of airlines, including Virgin Atlantic, Azul and Royal Air Maroc, to name just a few. 

It is possible to feed large amounts of data into a model that takes into account the many variables that influence pricing, in order to derive from it some predictions and actionable recommendations. 

Fetcherr combines millions of data points collected from multiple sources, internal and external, including many that are not obvious. 

Let’s say there’s a major football game at one of the cities served by the airline. Well, the system would take this into account too, as this is a factor likely to push prices up on certain dates. 

The system takes a holistic view of all factors that can affect fare levels across its network. 

The system then applies enormous computational power to simulate a broad range of scenarios, comes up with optimal pricing and pushes it to the airline’s other systems. It then integrates with all major Passenger Service Systems (PSS), as well as fare databases such as ATPCO. This process takes place on a continuous basis and it is fully automated. 

What’s more, the system adapts and keeps learning, understands where the competition is and what the consumer is willing to pay and generates prices accordingly. 

Nasdaq-style pricing for airlines

Speaking with AeroTime, Dr. Uri Yerushalmi, Fetcherr’s Chief AI Officer and Co-founder, noted the similarities with high-frequency trading. 

“We are bringing speed and price efficiency to the airline industry. All our infrastructure is geared towards this vision that someday, buying airline tickets will be just like trading a stock.” he explained. 

Another concept on which Yerushalmi insists is that of transparency. 

Yes, Fetcherr’s Generative AI may be a ‘black box’, a system capable of taking decisions on its own after taking into consideration many more parameters that a human mind could comfortably process, but it has also embedded explanatory capabilities. 

That is, if the revenue analyst wants to understand why the system came up with a specific fare, it is possible to ask and to receive an answer. 

What’s more, the whole interface has been designed in such a way that facilitates this human-machine interaction. 

Fetcherr’s founders are well aware of the fact that many airlines are reluctant to give away control of something as essential as their fare policy. This is why a number of safeguard features have also been built in, enabling a gradual approach. 

Revenue managers can define the bounds within which the AI is allowed to act unsupervised. A hypothetical example would be an airline that sets an upper boundary of 7% for price raises. If Taylor Swift suddenly schedules a concert somewhere and the system expects demand to that city to surge, proposing a 20% price rise, a human would have to approve it. 

That example presents a fairly simple scenario, but because of the complexity of such a system, some decisions can be quite intricate to explain.  

As a result, Fetcherr is moving away from a traditional way of communication, using an interface with widgets and graphs, to a more flexible chat-based system, which also uses natural language to provide answers. It can also generate graphs and other data representations upon demand. 

Though the Generative Pricing Engine can be operated on its own, Fetcherr has developed another AI tool called Generative Inventory Engine, which makes it possible for airlines to optimize capacity alongside pricing. 

“We don’t intend to mimic humans but to deliver tangible results. Our market model is quantitative, and it is easy to measure the accuracy,” explained Dr. Yerushalmi, before adding extra detail about the results Fetcherr is obtaining with some of its airline customers. 

He explained that, following the implementation of Fetcherr’s Generative Pricing Engine, some airlines have seen double digit growth in Revenue per Available Seat Kilometres (RASK). 

“We see ourselves as a hedge fund, but instead of trading a portfolio for customers, we are trading airline prices and generating a return on this for the airline,” added Yerushalmi, before concluding on an optimist, bullish note on the future of AI-powered applications. 

He highlighted how Fetcherr’s customers have been gradually expanding the scope of tasks and decision-making they entrust to the Large Market Model suggesting that increasingly this is going to become the norm, not just in aviation but across all industries. 

“If the AI is built in the right way from the ground up and in a transparent way, it can gradually take on more and more decisions. You start with micro-decisions and then you progressively move up the macro-level.” Yerushalmi concluded. 

So what’s in this for the human users, then? 

That question will be the subject of an upcoming article in which we will share some insights about the way AI systems can in fact become a valuable tool to empower the airlines’ revenue management teams. 

Leave a Reply

Your email address will not be published. Required fields are marked *

AeroTime is on YouTube

Subscribe to the AeroTime Hub channel for exclusive video content.

Subscribe to AeroTime Hub