Artificial Intelligence (AI) and Machine Learning (ML) are critical tools in the modern airline competitive toolbox, but they can be clunky. They are often overwhelming projects and can sometimes initially yield underwhelming results. But the promise of AI to build more meaningful and efficient connections with staff and customers gives airlines good reason to embrace this technology, even in its awkward infancy.
When informing AI systems, the quality of the data supplied can impact the result. The models used to process that data can shape what the neural network makes of it. Computers don’t think like people because people are still figuring out how to make computers feel. Besides, people haven’t quite figured out how they process information either. That’s why it’s essential to have standards for the structure and the use of data that will inform AI applications.
That was one of the salient points made during a fascinating discussion on AI systems at the Aviation Festival, London. The panel led by Alan Talbot, CEO and Founder, Bridge Solutions, Ltd, included Ben Dias, Data Science and Analytics Director, easyJet, Justin Bundick, Senior Director, Data Science & Automation, Southwest Airlines, Oz Eliav, GM, Cockpit Innovation, ELAL, and Alex Mans, CEO, FLYR Labs.
They tackled the question: How can airlines find new data sources to create a more complete single view, operations and support real-time and agile decision making?
“How can airlines find new data sources to create a more complete single view, operations and support real-time and agile decision “making?
Justin Bundick, Senior Director, Data Science & Automation, Southwest Airlines, spoke to the importance of data set integrity, saying: “We need to be staffing and doing a lot more accuracy monitoring. A lot of companies call it ‘drift monitoring’…You have to spend a lot of time and due diligence on that. In the past, you might have been able to deploy a model and check in on it every month or every 60 or 90 days. You need to do it daily or weekly now because of the volatility. You must ensure that you have the right staff to go in and do the deep-dive analysis, and make sure that dataset ‘A’ doesn’t need to switch to data set ‘C.’”
“I think there are a lot of companies pushing very strongly in this space, both from a data and analytics perspective,” Bundick added. “Whatever company you’re in right now, you have to look at your business strategy, translate that into a digital AI strategy that then converts into a data strategy, which then converts into a tech stack strategy, to be able to host and utilize that data. And involved in that, there will also be an IoT strategy. Especially in an industry like ours, where not all the data you need is created by transactional systems, it’s interactions that you need to capture. So, all of that must be tied together. It’s multiple layers of strategy that you have to deploy across multiple different parts of the business.”
“Until the business strategy and data strategies align, you don’t get anywhere.”
Ben Dias, Data Science and Analytics Director, easyJet said: “Until the business strategy and data strategies align, you don’t get anywhere. People processes are important. Even if you have all the business strategies aligned to the data strategy and all the data in one place, if you’re asking the wrong questions of the data, you won’t get the right answers in your business roles. So you have to also train the data literacy skills across the company. You might start with the data scientists and data analysts, but you have to eventually get out to the business as well—having essential, excellent training for people.”
Dias added: “I think that there are two key priorities for me to make it happen. The first one is the literacy level across the company because even if you made the data available, if you’re not able to use it, it won’t help. And it is also creating that platform that is easy to use and available and has the data in it. That’s, that’s hard when you are looking at a company that has been here for a while. The data has built up over time, and the data sets are all over the place. Bringing them together, and making them available, is a challenge, but it’s not insurmountable. You need to bring the data and make it accessible, maybe not in one place—just making the data available and upskilling the people across the company to use it. Those are the two things, I think, that will accelerate the process.”
Oz Eliav, GM, Cockpit Innovation, ELAL spoke to the role of automation in the data gathering process. “Automation is also contributing to the accuracy and objectivity of data,” he said. “If you have the objective data, then you can probably make actionable insights actual intelligence based on this automated data with no human intervention.”
“It takes a lot more investment from your technology organization. It takes a lot more skillset from an overall platform and data engineering perspective. But it’s really powerful.
Speaking to the agile application of AI, Southwest’s Bundick suggested: “It takes a lot more investment from your technology organization. It takes a lot more skillset from an overall platform and data engineering perspective. But it’s really powerful. Because by building those types of platforms for your AI, you’re able to deploy it. But not only that, you’re able to monitor those AI that you’re deploying and be able to make adjustments to them when there’s volatility happening. You can change them out without having to bring down a system updated in the system itself.”
Alex Mans, CEO of FLYR Labs, suggested airlines have underutilized their data. “Inform yourself with the broader data you have access to…Find ways to extract signal get past the noise get past the data sparsity collect more data and focus on making maximum use of that before you look too far out,” he said. “But equally important is structuring the data so that it can be processed efficiently. “Data sources change over time….Most airlines that we work with, we build our own canonical model on top of wherever data sets they have—because it’s never perfect.
“Once you get that out of the way, things will move a lot faster in the future.”
“Most importantly, we cannot afford as an enterprise SAS company to go and custom hook-up every data source we need wherever and however it sits. We need our software to read from a predictable common format. So we always install our own canonical data model because it creates a consistent system boundary between different airline systems and our solution. That enables us as a technology vendor to move much faster on deploying new capabilities. Because every airline we work with, the data we’re looking for, regardless of the airline, is structured the same. We still go through the steps of converting the airline’s data into our respective format, just because once you get that out of the way, things will move a lot faster in the future.”
While legal compliance to data regulations is an essential requirement, the ethical use of data is also a concern, with AI systems guiding decisions that directly impact people, Justin Bundick pointed out. “At Southwest, we are starting to spend a lot of time on governance processes around the ethical use of data and having the right touchpoints in place to understand the features we are using. Do we agree with those features? Should we be using them? What policies do we put in place around that? Even more than that, as we start to monitor the efficacy of our algorithms, we also monitor how the features that we are using are influencing potential decisions made. Are those decisions driving unethical behaviour? We don’t have the magic bullet yet. I don’t know that many companies do, but it’s something that we’re very focused on. It’s one of our priorities, as we move into 2022, to establish that. We’ve got a rough framework right now; we want to make that a robust framework by the end of 2022.”
by Marisa Garcia
