30 Apr SAS Daniel Engberg Head of AI and IT Strategies On Getting New Technologies To Work For Organizational Change
Written by Marisa Garcia – Flightchic.com
We spoke with Daniel Engberg, Head of AI and IT Strategies at SAS about the potential of AI to transform airline operations and customer relationships, the challenges of building an AI competency centre from the ground up, and how to ensure that machine learning works at an organisational level.
The SAS AI team was formed in August of last year as an independent team focused on practical applications of data analysis and machine learning. Engberg currently works with a team of two full time staff members and around ten consultants on carefully targeted projects that range from CRM to operations.
What are the challenges of finding data analysts and other technology professionals to push innovations in artificial intelligence at the airline? Is it better to recruit from outside of the airline industry?
My view is that you need is a combination of skills. We’re looking both externally to source competencies among data scientists and statistical people and we need to combine them with people internally who actually know our specific data and our specific business challenges.
To succeed, I think it’s a combination of different competencies. It’s not that we need to source only one skill. For each initiative, we need a combination of resources. First we need to understand how we solve this concrete business problem, what’s actually valuable to bring out, combine that with the knowledge of our own data, how that looks, and combine that with specialists in data science.
It’s more of a mix.
Do you find it hard to compete for the pool of talent in the market, especially given the strength of Sweden’s high-tech sector?
There’s definitely a difficulty to recruit the right skills. That’s also why we try to find long-term partners that can provide us with certain niche skills that we might have a difficulty to acquire. We need to think of what level of competency we need internally, and what get externally, and how can we do that in an efficient manner.
Can we discuss some of SAS’s recent automation programs which rely on AI?
We’ve done some proof of concept for stand-by generation for pilots—how many we need on standby each day. We also set up automation on the chatbot side, for the travel assistant on Facebook Messenger. Right now, we’re looking at revenue forecasting through AI. There’s a lot of discussions going on, and we’re in startup mode on revenue management. We’ve previously worked on scoring models for customers, to target our customers better.
We’re quite heavily discussing how we will do operations as well.
We see a lot of potential of using AI, both in automation on the IROP side, to make that process less resource intensive, to automate more of the decision-making processes around that.
There’s a great opportunity for scenario analysis, and closer to day-of-departure for the operational control center; to have decision tools based on data, to understand the case, ‘If this scenario happens how can we do that.’ It’s discussions that we are having.
In general, I do think the application of AI to personalization and the customer side is very important, but I feel very strongly that we can do a lot for our employees in terms of planning—how can we do resource planning more efficiently.
For example, looking at how we can use technologies like text mining to provide better information to our crew and to our technical people to have the right information at the right time from the vast amount of documentation and procedures that exist. How can we use AI to actually bring that knowledge to our people when they need it and where they are?
What are some of the challenges of filtering through the large data sets produced in aviation, especially with the rise of connected aircraft? Are there specific targets for data gathering and analysis that you’ve found particularly helpful?
Definitely, I think that there are challenges there. My view is that the most important part is to understand how we get started. We need to start doing things to understand what data is valuable and what isn’t.
We can try to predict today exactly what data is going to be valuable, but we can’t really do that.
We need to look at it from the perspective of the problems we’re trying to solve. What is the data required for that? The earlier we get started, the sooner we can say, ‘We can’t build it now because we can’t access the data, because we haven’t saved that data historically. So we can’t actually build the machine learning model around that.’
The earlier we find that out, the better suited the data gathering and modelling will be for the future.
How do you build confidence in the reliability of the model?
I think, in general, it’s about making sure that you have the right stakeholders in place when you do something. Have the people who are working with this on a daily basis—whatever problem you are trying to solve—take part in developing it, and ensure that you fit their needs.
Then, we can start to build trust in the models, through ownership of the model. That’s also the key.
I could have an organization of 20 data scientists developing models, but if nobody uses them they are worthless. It’s a waste.
I think it’s about getting business ownership and getting that last mile that you can deliver the value.
The way we’re looking at it is having a phased approach; where we have a hypothesis, and we try to validate that hypothesis with smaller proof of concept. If we succeed—when we can prove this hypothesis was correct—then we move into the piloting phase, where we put it out into a limited amount of the business. We probably have it run in parallel with the current ways of working.
That builds confidence. If that works, then we can see the real value from it, then scale it.
How do you address visualising the data, making it something that the user can process and relate to?
It’s a difficult question to answer, in general, because its quite use-case specific.
As you say, it’s about visualising it, not just presenting the hard numbers, but putting it in a context where they can use it.
Take group planning, it’s about how the model predicts versus the old ways of working, so that they can see how the model performs.
As a historic analysis, then it’s about how can we provide something that they can use side-by-side, whether it’s Excel or a webpage, or whatever the format that they actually use.
That’s what’s so important about working closely with the business areas. The outcome, the visualization, should be more dependent on the ways of working of the people in the business; so that it’s actually something that they can use in their daily lives. So they say, ‘When I get this, it’s actionable for me.’
A key to all of that is the closeness to the people who are actually going to use that data.
Where do you see advancements in AI at airlines, compared to other industries? Are airlines keeping pace with other industrials or service companies?
I think we could do a lot more in the industry. I don’t think we have started seeing what the full potential is. On the other hand, I don’t think that other industries have either.
I think most of the things we see, and a lot of the things people are talking about, when you really look into it, are still in the proof of concept stage.
The interesting part is how they actually get the real value from it, and we haven’t explored that enough in the industry. I think we could focus a lot more of it.
There are things that are needed quite early, like digitalising a lot of operations so that we don’t have any physical tickets. Everything is electronic, so there’s a lot of data and a lot of potential. But I think we are lacking a bit on the maturity side to understand the data.
There’s a lot that needs to happen on the business side as well. We need to be more tech savvy across the whole business.
I haven’t been discussing this that much with any companies in the airline industry. I’ve collaborated more with companies outside of the airline industry, to get a sense of what they’re doing and get their insights. Also, there are a lot of pitches from consultants.
My peer to peer discussions are more with other Nordic players in other industries. A lot of Industries are focusing a lot more on AI and machine learning than we are—take banking, for instance, they are focusing more on that.
When you talk about outside resources, in-house resources, and consulting, what makes for a good partnership?
I don’t have a specific model around that that’s defined. My view is about long term commitment, ensuring that we can access a pool resources. Especially, when you talk about AI and data scientists, it takes a while to understand the data. So, the more that we can utilize the same resources and the same partners, then the more value that we’re going to get from the partnership.
Where are AI and new technologies headed? What’s the paradigm shift for IT operations?
In general, my perception of this whole area—the shift that we’re seeing with AI and technology in general—is that we are moving to a world where were where we are using technology to completely change our business models—using a multitude of different technologies—to ensure that we can more quickly adapt our business to our changing landscape.
I think we’ve only seeing the start of it. Everything that we’ve been discussing now, with the different pilots, has been about small changes to a specific process that we are doing more and more continuously.
That’s quite different from the traditional IT, when we would get a new system that takes two years to implement and have a stable process running for ten years. That’s one of the master challenges that we’re trying to convey, in this area. It’s not about a specific technology, but about how we can build a toolbox to ensure that, when we look at a process, we ensure that we pick the right tools for the right tasks.
How do you ensure that? How do you make the organisation comfortable stepping away from the traditional airline process of a prolonged RFP process?
We’re handling it by ensuring that we are clear that we have a hypothesis to try out. We have a short investment of, say, eight weeks. Then we try the product for that time or we develop a machine learning model ourselves—or together with partners.
After eight weeks, with the evaluation of the results, we should we move on and try to narrow down the scope of the problem. We should not try to solve everything from the beginning. Otherwise, we get into the large scope projects; there will be large discussions and long time periods.
Rather, we say, ‘We can do this. We have the value proposition and we can gradually grow it over time.’
That’s also a way to ensure that we have the ability innovate faster, and that we don’t hesitate to close things down when they are not working.
If you have a minimal investment—a six week or eight week initiative—it’s not that much money down the drain if you close it down.
It also an efficient way to start. Everyone knows, from the start, that it’s a limited initiative.
What’s your advice to airline peers exploring the benefits of AI?
My view is that we should have a center of competency around AI—everybody should be more knowledgeable about it. We started as a competency centre to serve as a catalyst for the organisation. But there is a transformation and change that needs to happen in the organization. We can only facilitate the change.
I don’t think we’re even seeing the beginning of AI and machine learning. We’re right at the start of it. The more we can do now to be prepared—that’s the key.