Aviation industry Analytics Startup: We will get nowhere without data partnerships

There is a paranoia about sharing data, explains the entrepreneurs behind an intelligent forecast system for the aviation industry.

Magnus Boye, @magnboye
Read original article in danish on Ingeniøren DataTech.

Danish data analytics startup Beep Analytics ApS has developed an ML-based platform that forecasts airline use of spare parts – providing the basis of a more focused sales strategy in the complex market for aircraft spare parts.

The software, launched as FlightDeck this month, is completely dependent on external data providers - and their willingness to share their potentially precious data. This does not come without problems, says Jens Peder Pedersen, founder of Beep Analytics.

"The industry is afraid of platforms that potentially push companies backwards in the value chain", he points out.

"We have started to encounter situations where we discuss with the data-owners if we are partners or competitors. They could potentially get the perception that we put ourselves in a position where we gain control over their data. Others think that maybe they should do the same as us."

After two and a half years of work on getting the analytical platform up and flying, it is clear to Jens Peder Pedersen, that partnerships are a key factor for success. To achieve these partnerships, you actively have to seek them out, he says.

"If you're a little arrogant and don’t talk about what you're doing then it's much harder to get past the concerns about sharing data. If you want to succeed with these things, you shouldn’t sit in a basement for three years until you have something you can actually show. It's not effective", says Jens Peder Pedersen.

"It is much better to develop partnerships and find the right consensus that benefits everyone. If you can do that, I think the chance of getting people on board with you is much bigger. "

Paranoia about data sharing

The idea behind ​​Beep Analytics’ solution is that manufacturers and aftermarket distributors of aircraft spare parts can develop an intelligent prediction of which and how many spare parts you are able to sell to an airline like SAS or United Airlines.

The forecast is based on historical sales data as well as airline fleet data, aircraft types and variants, and which routes they operate. Much of it is data that the company has to retrieve from external data providers.

The company pays for the most critical source of data - who operates which aircraft and who maintains them. Other data sets are retrieved through partnerships – among them a partnership with a company which has provided a search engine for aircraft spare parts since the 1980s.

With regards to the airlines themselves, the door is often closed, says Jens Peder Pedersen.

"The aviation industry has always had a kind of paranoia about data exchange," he says, and continues:

"Airlines will not uncritically share which parts they buy and where they buy them. Lufthansa does not want to share proprietary information about their aircraft maintenance processes, which they have optimized through more than 30 years. Everyone protects their data. It is a problem in relation to us. "

Cooperation with University has been essential

Therefore, it has been essential for Beep Analytics with an additional partnership – with a customer. In this case, a customer who shared the vision, and understood that the system was still under development.

"We have committed to not sharing their data with others. We guarantee our customers that their data runs in a totally isolated silo and we will not exchange any information between different silos”, says Jens Peder Pedersen.

"However, we would like to use our customers’ data to optimize the underlying algorithms in the system, and so far we have been met with an understanding for this requirement."

Finally, Beep Analytics has been dependent on a collaboration with the Technical University of Denmark - DTU, which saw the potential of allowing research assistant Valentin Liévin to develop the complex machine learning algorithms that form the core of the FlightDeck analytical platform. One of the key functions is an ensemble deep learning model that can determine which aircraft a spare part is fitted to.

"The partnership we have had with DTU has been crucial for us. We had never reached this level of maturity without that cooperation", says Jens Peder Pedersen.

Customer education

As the Beep Analytics' system has been put into use, the eight-person team behind the company has encountered a new challenge – how to create a culture of trust in the ML-algorithm predictions.

"In some cases, we received feedback from experienced sales managers who look at the prediction data, and decide it does not fit - without speaking to the customer. That's something we're going to work with", says Jens Peder Pedersen.

Another question is how to avoid creating a tool that management uses to hit the sales department over the head.

»It's not a question that I can answer clearly. We have to educate our customers. They have to understand that this is a new tool that does not create a revolution from one day to the next. But it will put our customers in a new position, as a more competent partner for their customers, the airlines and MROs.«

Extend with master data management

Over time, the platform has the potential to be extended to handle master data management. With a background of more than 25 years in the aviation industry, Jens Peder Pedersen has at first sight found that the quality of master data is often far below the target of suppliers and manufacturers of aircraft parts.

"Masterdata is typically very inaccurate. It's appalling how bad the data quality is", he says.

In some cases OEMs have had a de-facto monopoly on supplying specific aircraft parts, and the pressure to develop a proactive and customer centric sales strategy has been practically non-existing. Now, increasing competition from Boeing and Airbus among others, push suppliers to be much more proactive in order to make money in an aftermarket, they’re suddenly deeply dependent upon.

"Then it hits you with full force if you do not have your data under control. When you have a product portfolio with 100,000 different part numbers, and you don’t know what you can sell to whom, then it's hard for you to go out and push sales", says Jens Peder Pedersen.