Working on an airline has been a step back down to Earth for Virgin Atlantic's data and AI leader. Richard Masters’ career in the grounded world of data science and artificial intelligence started by studying the cosmos as an astrophysicist at the University of Oxford, a stellar experience that has given him a unique perspective on his current role.
“I'm in two minds about the term data-driven,” he tells The Stack. “Most data is noise. Astrophysics involves gathering data and imaging galaxies. But the first thing you see is static and you have to spend half your time reducing it in order to see the signal. Cutting through that noise is key - and it can be a hard thing to do.”
As VP of Data and AI, Masters’ remit covers all enterprise data at Virgin Atlantic, ranging from the information generated by its commercial activity all the way to the data from engineering or maintenance teams. He's responsible for forging the links between data and AI, focusing on “how we build the engines and algorithms”.
He’s been in his current role for roughly 11 months after spending two and a half years at EY - which was actually a short break from Virgin, where he previously held the position of Head of Data Science and Engineering. Today, he’s focused on unlocking the benefits of Generative AI (GenAI). Masters reports to the CFO and says the Virgin management team “really gets the opportunities and the snake oil out there when it comes to generative AI”, as well as the need to build an ecosystem to support AI-driven customer experiences.
“We’re focused on getting the data right - which means making sure our governance is correct,” he says.
Generating AI innovation
Last year, Virgin Atlantic recorded total revenue of £3.1 billion, up £265 million from 2022. Its strategy focuses on "creating meaningful customer and people journeys that differentiate us whilst extending our reach and relevance through data, digitalization and brand power growth."
The airline is now on a GenAI journey which began with relatively basic use cases such as summarisation and categorisation of messages from email, web forms or other sources, which is then combined and processed in combination with its knowledge policies and procedures, as well as details of the tone of voice and other brand-related information to makes the responses of an AI model more consistent, convincing and human in tone. It was the first airline to adopt fully automated generative AI pricing technology, which is now live and being used to price selected routes.
Customer service chatbots demonstrate the power of GenAI, Masters says, which offers transformative benefits for both passengers and airline staff.
"When the customer calls up the call centre because they've lost their bag or need some special assistance, parsing all the information and then generating an answer is a really great Generative AI use case,” he says. “Then you build on top of that.”
Once all the airline’s data is in place correctly, it can then be used to get the information required to respond to a prompt. As more data is collected, stored and made available for processing, it has a cumulative compounding effect on the accuracy and usefulness of the GenAI models. “Creating consistency gets people to the right answer,” Masters says.
Skyhigh data and AI ambitions
To fuel its AI journey, Masters’ team uses a variety of Databricks solutions, including AI/BI Genie, a conversational tool which lets teams engage with their data through natural language prompts - letting Virgin “talk to our data more fluently”. By conversing with its databases, Virgin Atlantic is now able to gain a consistent view of customers across all touchpoints by joining up its data sources to break down silos. It can also zoom in on the history of customers’ relationship with the airline to offer tailored incentives and rewards.
Databricks creates a collaborative environment for data science, allowing exploration and sharing through notebooks. Initially, Virgin Atlantic focused on isolated data models, but more recently, the company has shifted towards using Unity Catalog to centralise data domains like flights, customers, and pricing, enabling analysts to explore relevant data independently.
“The balance lies in having a platform that offers the right tools and ease of use for your technical teams - engineers and scientists - to collaborate and create effectively,” Masters reflects.
“At the same time, you need to support the rest of the business by enabling self-service for building dashboards, reports, and conducting deep-dive analytics. For example, if a report raises a question, you need the ability to explore and understand why. There are point solutions that address some aspects, but nothing that fully covers the entire ecosystem yet.
“Databricks is moving in that direction. It’s starting to bridge that gap by helping teams collaborate on data more effectively. The consultative approach and strong communication are important, but so is having the technology stack that allows quick iteration on data.”
READ MORE: Homomorphic encryption pioneer on Apple, DARPA and cracking the code for mainstream adoption
Virgin Atlantic has also adopted machine learning operations (ML ops) with tools like MLflow to manage models and democratise their use. Databricks acts as an orchestrator for large language models (LLMs) and machine learning, with a focus on creating composite models. This ecosystem helps deliver AI-driven services, such as chatbots and predictive models for flight load factors, which can be shared with both internal teams and external partners. The platform’s guardrails and management tools allow flexible and secure use of models, whether hosted internally or externally while balancing cost and performance. Databricks also lets Virgin use a variety of open-source models, including Llama, Claude and GPT.
“Tools like Genie give a glimpse of how processes can be made easier,” Masters continues. "Ultimately, being able to translate what the business needs into something the data team can work with is crucial. Tools are just tools - talented people can adapt to them. But the more you can get different teams working together on a common platform or using a shared language, the easier it becomes to deliver real value.”
Learning from the jet set
The airline industry has been at the cutting edge of data science since the dawn of aviation. It was an early pioneer of databases, with IBM’s SABRE system first used by American Airlines in the 1960s to manage its reservation data in an era when the Apollo program used comparable databases during its mission to the moon.
Airline booking systems, originating from those early relational databases, are complex due to the interconnected nature of the airline industry. Many journeys are not simple point-to-point travel from A to B, involving multi-leg journeys, cancellations, and changes in itineraries, making ticket and passenger tracking intricate. The underlying systems are still based on an approach originally developed for a very different world.
“It's still based on the logic of physical coupon books, and that limitation is still present in those underlying systems,” Masters reveals. “So cracking tickets and bookings is very complicated versus other industries.”
Airlines face unique challenges in defining key terms such as “passenger” and “booking.” For example, “passenger” could refer to a seat or an individual, and “booking” could represent an entire itinerary that may evolve over time. This creates ambiguity in data models, complicating how airlines handle bookings and manage passengers. Developing accurate models is, therefore, crucial - but not always easy.
Newer technologies, such as knowledge graphs, are being explored to help standardise definitions across different departments, ensuring that marketing, operations, and commercial teams all understand terms like “passenger” in the same way. This alignment is still a work in progress but is a key focus for improvement in airline operations.
“A good chunk of the volume has this really complicated setup of tracking the ticket and the passenger,” Masters continues. “So getting that modelling right is a really unique challenge for airlines. I've worked in pharma, banking, and all over the place. All those industries have their challenges and complications, but the airline booking setup is quite unique.”
READ MORE: Databricks' new open data type wins praise from Goldman's CDO
Piloting change at a global organisation
Getting transformation projects off the ground is tricky for all businesses. For Masters, the key is "knowing when to engage the right leaders and ensuring the team has both technical expertise and a consultative approach throughout.”
“When trying to enact change, you need to think about the ultimate problem and who you need to involve at the right time, which can be tricky,” he reflects. “Timing is crucial, as some people may have too much on their plate, so you need to find the right leaders with the capacity to support you. Picking your moment is important, but so is demonstrating expertise. Your team needs to constantly show they understand the underlying technology and how it adds value without being overly technical."
Team culture should focus on delivering value and fostering expertise in the technology which creates that value, he says. “Balancing those two is critical, but it’s also difficult,” Masters adds. “Building a consultative mindset within these functions becomes key to transformation."
Your chance to explore Virgin territory
As our conversation closes, Masters reveals something which may be important to readers of The Stack: he’s hiring more than a dozen people and aims to grow the core data and AI team to between 40 and 50 people.
“Finding real data scientists who can handle noisy, unstructured business data is a challenge. We’re also focused on the concept of data products, and communicating that clearly to the business. Whether we’re building dashboards, data sets, or models, we need to promote and market them like any product.
“I often use the example of Apple product managers presenting new devices during keynotes. That’s how we want data to be presented. We can build great solutions, but if the business doesn’t know how to use them or extract value beyond the original use case, it’s not as impactful. So, finding people who understand both data science and data product management is another key focus for us.”
Check the Virgin Atlantic jobs site for the latest opportunities.
