Four reasons why airline commercial departments will struggle with AI
Experience handling previous technical revolutions is unlikely to help in aviation
Think of airlines adopting new technologies and the history is one of success. Planes fly further, engines burn less fuel and passengers slumber in comfy flying beds.
Commercially, airlines have used technology successfully too. Reservations and ticketing were computerised back in the 1960s. Complex fare products were introduced in the 70s. Inventory management helped optimise revenue in the 80s and whizzy revenue management (RM) algorithms came along not soon after in the 90s.
In the dot com revolution, airlines were early adopters. London-based ba.com, british-airways.com at the time, launched in 1995.
So surely when it comes to the technology of the moment, generative artificial intelligence (AI), airline revenue generating teams will be making the most of the technology swiftly? Sadly I think this is unlikely.
At first glance airline revenue generation should be a great candidate for AI. There is oodles of data, all the RM vendors are heavily invested and airlines love centralisation, which is what AI is all about.
Blockchain, a competing technology, also holds lots of potential for the industry (see this article, this article and this article), but airlines have largely rejected the opportunity for now.
Unfortunately there are four reasons why efforts by airlines to implement AI are likely to fail, at least for the foreseeable future.
First up is sheer inertia. While airlines love to spend millions on complex products from technical vendors like PROS (see article), these systems tend to take a long time to deliver what they promise because they are built on the back of existing technology. Some features take a while to be weeded out, even in the new world.
The PROS implementation of “Real Time Dynamic Pricing” (RTDP) at Qatar Airways from 2007 to 2010 was neither real time nor dynamic pricing. It was, however, able to achieve origin-destination revenue management, which at the time was well established but not heavily adopted, very well.
Like RTDP, the vendors’ early generation “AI solutions” will promise much but solve little. It is worth pointing out that airline RM was one of the original forms of artificial intelligence, using large amounts of historical data to discern patterns that are hard for humans to spot and making appropriate commercial recommendations.
To make things even more sticky, teams of airline analysts and managers will stay wedded to the way they do things today. This is one of the reasons blockchain ticket trading has failed – it is simply too different to the zeitgeist. It will take phenomenal effort for airline leaders to migrate their teams from one way of doing things to another.
Second is cost, especially salaries. Airlines have a track record of overpaying for technology but underpaying for staff.
This means that they will find it hard to recruit AI specialists, who will have higher paying options in other industries. According to Lightcast, a research firm, sales reps with AI skills can earn $45,000 (£35,700) more than those without them.
Given that this year Spanish airline Air Europa tried to recruit an experienced sales rep in their London office for only $35,000 (see article), it seems unlikely they will be able to recruit the AI specialists they need. AI skills are in demand. AI-related job ads in the USA have increased by 122% in 2024 so far, compared to 18% in 2023.
It is possible that airlines will source their commercial AI skills through the technology vendors. This will help achieve some goals because the vendors will in time no doubt create great products. But existing problems where airlines rely on outsiders to get what they need done and lose their best boffins to their suppliers will be entrenched.
The third reason that airlines will struggle with AI is messy data. Airlines have plenty of data, but it is not linked. When I accepted BA’s invitation to see all the data they held about me, I got my name, membership number, a list of all the flights I had booked and that was it. There were lots of things not on the list, including:
1. What I had paid for my tickets
2. What seat I had travelled in
3. Who my travelling companions were
4. My shopping behaviour on ba.com
5. My lifetime Avios earning and redemption.
All of these should be available to the airline. That they are not all available in one place or linked shows how siloised each department is.
Airline revenue management generally work with revenue accounting data to show what has come before and advance bookings to see how similar this year will be to ones before. Sales look at their agency performance. Loyalty consider all the mileage earning and redemption partners.
New and emerging types of data, like Delta’s Sky Pulse (see article) which shows what passengers are doing right now on their planes, are in their infancy.
Much airline data is corrupt (see article) in the sense of not representing what is actually sold.
There are billions of fares filed with ATPCO, many of which are never or hardly ever sold. The “Fare by Rule” Category 35 introduces complexities which are hard for even their controllers to keep track of.
Meanwhile airlines inflate their higher end premium cabin fares to accommodate corporate buyers’ demands for larger and larger discounts.
For airlines using AI to try and develop offer and order type products or a commercial marketplace (see article), the jumble of unlinked data will put the AI at great risk of so-called “hallucinations”, i.e. mistakes.
Revenue managers try to sell the right product to the right passenger at the right time. A high risk of hallucinations might make this harder to achieve than it is today.
Finally, airline commercial systems themselves will be a barrier to AI implementation.
Back in the 90s it was comparatively easy for airlines to build Internet Booking Engines because these simply automated all the commands of the old selling platforms. Implementing AI will be an entirely different challenge and nothing to do with old methods.
Airlines already fill up bookings and reservations with complex robots built on top of the system. These help automatically show, for example, whether passengers are entitled to certain seating, lounge or baggage privileges.
When I did a pricing project for South African carrier Comair ten years ago the airline was a partner of BA. So I got to see how BA’s bookings worked first hand. (Sadly Comair is now defunct, COVID that did them in.)
One way of looking at them were as tremendous technical achievements, accomplishing much with only a basic infrastructure.
But another is seeing them as over-complex and high risk. They were absolutely chock full of robots. I suspect that this is a major cause of the airline’s perennial IT problems – when something goes wrong the system collapses under all the weight and cannot restore itself (see article).
It is hard to see AI being implemented in these legacy platforms at all. It is even harder to see what good it could do when airlines are struggling so much to define exactly what they want to do with IATA’s Offer-Order programme save spending lots of money meeting technical standards. On the other hand, it is easy to see it increasing the risk of system collapse and all the associated costly disruption.
Sometimes you have to feel sorry for airline decision makers
On 8-November this year, British Airways cancelled it’s long-established service to Bahrain. A symptom of the engine problems on their 787 fleet means that some flights have to go.
No doubt the network planners considered cancelling Bahrain the best of a bunch of bad options. Standard airline data will have given them many insights into ticket sales, average revenue, forward bookings and the like.
Unfortunately the standard airline data will not have mentioned that just three days later His Majesty Hamad bin Isa Al Khalifa the King of Bahrain was having tea with His Majesty King Charles III, King of the United Kingdom and reigning sovereign over BA’s hubs and headquarters. We can only imagine the conversation but services to Bahrain have since been re-instated with three flights a week.
Data may be the new oil and AI may be the next big thing, but a King is still a King.
oliver AT ransonpricing DOT com
oliver DOT ranson AT inkaviation DOT com