How Amadeus is developing AI – sustainably

It’s all about the long view, says CTO

Amadeus CTO Sylvain Roy explains why Amadeus does not view AI as being antithetical to sustainability of its cloud operations.

There’s no getting away from the fact that travel, and aviation is highly carbon intensive. Aviation contributes approximately 2.5% of carbon emissions, which doesn’t sound like a lot, but actually a very low proportion of the global population flies at all so it’s a highly carbon intensive activity.

ICT contributes approximately 3.5-4% of carbon emissions and there is a much greater awareness of the environmental impact of the datacentres that power the cloud where so much computing now happens.

Sylvain Roy, Senior Vice-President of the Global Technology and Cloud platform (GTC) organization and Chief Technology Officer at Amadeus is aware that, as the software that powers the global travel and tourism industry, the company must consider how sustainably it carries out its operations.

“Fuel consumption is not something we do directly,” he explains, “but our customers (many of whom are airlines) do and being such a major actor in this industry we have to help them with that.”

Some of it is about using technology to optimise inputs such as a fuel consumption optimisation module within flight management software.

In terms of what Roy can influence, the focus is on developing in cloud as sustainably as possible.

“Amadeus is committed to sustainability. We want to reduce energy consumption and carbon emissions. Our developers have a carbon emissions calculator so they can assess the impact of their design on carbon emissions.”

Like all proponents of cloud, Roy views it as greener by default.

“As an IT company we have to clean up our own shop and to ensure that our energy consumption is as clean and as limited as possible. The move to the cloud is part of that.”

Cloud services vary in their respective eco-credentials, and all three hyperscalers are not shy to claim that their datacentres can deliver much greater efficiency at scale than the on-premise, smaller datacentres model that it has replaced.

For a customer like Amadeus this is true, and it’s also true that a Microsoft datacentre (Azure is the Amadeus cloud of choice) will have a PUE much closer to 1 than the type that Amadeus would have been using ten years ago.

However, the way that businesses use cloud services also has an impact on their indirect emissions. Amadeus is not alone in viewing FinOps and carbon management as aligned objectives. As Roy notes:

"If you have too much cost you will have too much carbon.”

Amadeus does not rely on the carbon calculation tools that Microsoft provides, choosing instead to develop their own tooling. The extent to which hyperscalers undercount the emissions of their datacentres, and subsequently customers undercount theirs, has been the subject of Computing research. That Amadeus chooses not to rely on integrated tooling is interesting.

“Sustainability starts with design”

“How much computing and how much storage do you incorporate into your solution?” asks Roy. “For example, machine learning like decision trees, even a simple neural network might be more energy intensive when you are training it, but when you execute your algorithm, it costs very little.”

There is no doubt that training LLMs is highly energy intensive. Roy’s approach to the impact of AI on the broader sustainability of Amadeus operations is to take a long view

“Say you want to fly from London to Bangkok. There are thousands of variables in terms of routes, airlines, many fares and schedules. To interrogate the tree of possible solutions and get to the best price, the amount of computing required is crazy. Bringing in a bit of AI at the beginning of the process to determine on what branch of the decision tree you will do the computing, uses much less compute.”

Essentially this boils down to using more energy in the short term to use much less over the longer term.

This is what Roy considers to be the responsible use of AI. Apply where it makes sense and use emissions calculators that can tell you the impact of what you're doing.

GenAI at Amadeus

Like many other technical organisations, Amadeus has been using machine learning for some years, and has put together a framework to decide how generative AI should be applied. Roy says there were two considerations. The first was how to enhance the Amadeus product. The second was how to utilise GenAI to become more efficient internally.

“With the latter we use GitHub Copilot and Office Copilot and we use an LLM to get access to our knowledge base. We’re doing a lot on the internal side already and there are no questions about the benefits.

“On the product side we have 70 initiatives under way. Some are already in production. The first category is using conversational interfaces rather than using a GUI. So for example we have a corporate booking tool where instead of having a webpage you have a conversation with a GenAI in Microsoft Teams. For a corporate traveller it makes booking easier with no complicated interfaces and your company policies already in place.”

Amadeus is also using GenAI to build websites for destinations and using it to personalise fares. The company has also built a digital assistant Roy says.

“Our product can be complemented with a digital assistant which means that as a user there is an interface where you can ask questions. We have a fairly complex product because it solves fairly complex problems.

“It makes things easier and brings real value. The same applies to the conversational booking tool. I haven’t gone any booking on the old UI since I’ve been able to do it via a conversation. I say what I want to do and in a couple of sentences i get where I want to be.”