Continuing our coverage from the World Aviation Festival, we're moving on to one of the most talked about parts of the industry - artificial intelligence and machine learning.

The event provided a complete AI/ML track and great insights from industry experts.

The collective theme from these discussions is that it cannot be all about the science, but to be successful in AI/ML functions, an airline must equally blend in business strategy to create applicable real-world use cases.

Avoiding Failure

Before we dive into how to be successful, let's learn some lessons from what doesn't work. In his presentation, Dirk Jungnickel, Senior Vice President Enterprise Analytics, Emirates, provided two staggering statistics:

  • 87% of data science projects never make it beyond the initial vision into any stage of production.
  • Through 2022, only 15% of use cases leveraging AI techniques will be successful.
"There is a profound difference between doing something as a POC or as a pilot compared to doing it at scale."

So, the next logical question to ask is why is this the case? Jungnickel explains that is largely due to the big difference between having a proof of concept (POC) and launching something into a live environment.

"There is a profound difference between doing something as a POC or as a pilot compared to doing it at scale at a company like Emirates with thousands of employees where you need to run things in a stable technical environment and 'productionize' what you are doing. And that is where many companies struggle."

Having the Right Model

More specifically Jungnickel pointed out there is often a disconnect that can occur between the team building the models and the rest of the business.

"Data scientists and PhDs are very good at building algorithms, but they're not very good at understanding algorithms next to a business problem. You need to model for the right thing," said Jungnickel. "Machine learning and AI solutions almost in all cases means changing the business process."

"You have to understand what you want to get out of your models."

You can have all the data and technical sophistication in the world, but if you're not applying it in the right way to the right areas of your business, you are doomed to fail.

Since the possibilities are endless It's all about producing the right model for your business.

"You have to understand what you want to get out of your models," said Melissa Skluzacek, Director -Trading & RM, easyJet. "You have to have the subject matter expertise."

Turning Information & Data Into Insight

The best way to counteract these problems is by not only making information available, but making the right information available which provides insights across your entire organization in order to problem solve from a holistic approach. The term is popularly known as data democratization.

"One of our biggest accomplishments this year is we did a lot with data democratization by bringing in dashboards that were extremely visual and extremely user friendly and shared them with a number of people in the company and making it easy for them to digest," said Skluzacek.

"You can produce a model to do anything, it doesn't mean it's going to fit your business needs."

By democratizing data, it becomes easier for airlines to work cross-functionally and better meld together scientific intellect with business expertise.

"You can produce a model to do anything, it doesn't mean it's going to fit your business needs," said Skluzacek.

"One of the main things we do is you have to know what is the business problem you are trying to solve with the model. And working very closely with the business team and the data science team we make sure everyone understands what it is we're trying to solve."

As for where airline AI is heading, Manuel van Esch, Lead Consultant, ZeroG, offered up some insights.

"Any dream of any data scientist is automation."

However, van Esch did point out the importance of still having humans involved in the process, especially from an ethical perspective.

"Human oversight is very important. When you make that decision of having the human in the loop with the machine learning applications, those are the very moments to make sure things are being built ethically."