A while back FORLOOP was asked to analyse data for an airline client around supplier warranty claims. Our client was curious about the number of actual warranty claims they had made over a given period vs what they could have made, and why.
Our first step when looking at a problem is to examine existing business processes and the business impacT
What we discovered was that the impact is high. The processing of su
pplier warranty claims often equates to millions of dollars due to the high cost of maintaining a commercial aircraft fleet and the number of parts ordered in any given period.
The airline warranty claims process is very human centric
The warranty claims process involves the identification, lodgement and tracking of warranty claims using data from various sources and spreadsheets. While some MIS might be able to identify eligibility via part number, it cannot indicate which warranty clause is applicable, nor provide supporting evidence easily. The administrator still needs to do a bit of hunting around and collating, and manual tracking. Herein lies the problem...
The human brain can only process so much data
The human brain can often only process data from up to 5-7 data points, after which our capacity is capped and we become inefficient. Any process requiring accessing multiple data sources makes it a perfect candidate for Machine Learning.
Machines can process from hundreds or thousands of data points. In fact the more data points or more complex the data, the better candidate for Machine Learning. The advantage of Machine Learning is that it upgrades with the speed of the data. There is no need to wait for human reviews or cyclical reports.
The more complex the better
When tasks are simple, the benefits of building a machine based system is minimal. But when the task is complex, the machine can often be far more efficient than humans.
In the case of supplier warranty claims, the complexity was in the Warranty Contracts. When an airline is in receipt of a contract, they often have their legal team look over the T&Cs before sign off (ie a specialist). However when it comes to adhering to the T&Cs, airlines often use a team of non-legal specialists for the ongoing interpretation of complex T&Cs and waterfall clauses. This process relies on the knowledge and memory of the administrators involved.
The added complexity for airlines is that there are 6 different contract types - Standard, Reliability, Rogue Units, Direct Maintenance Costs, Shop Processing, and Freight. Each has their own complexities, and differ by vendor. This made our investigation learnings an ideal candidate for Machine Learning!
When examining the warranty claims process, we realised there were a few human limitations which could be overcome by machines. These were as follows:
- Humans are limited by their capacity and hours in the day;
- Humans often rely on memory, or rote learn a process;
- The human brain becomes less efficient when using data multiple data sources; and
- Humans make mistakes
My challenge to the team was to build a Machine Learning solution which could overcome these limitations, to provide a more efficient way of doing things for airlines and MROs.
Machine Learning advantages
When a machine is programmed to do something, it has no bias. Algorithms are written and performed, then tweaked if the desired result is not forthcoming. What we wanted was a machine that would improve with each warranty claim.
A effective machine learning system is one that improves each time it makes performs a process. If managed, it never makes the same mistake twice.
When a human makes a mistake or discovers an obscure way of doing things better, the knowledge is often lost across the team. The advantage of a machine is that it "is" the team, so once detected and tweaked, it then becomes the routine process.
The second advantage is that machines can process and analyse data faster and more efficiently than any human. And because it is data, if programmed correctly it can then churn out management reports with a drag and drop. This gives management and users visibility and speed unattainable through manual processing.
If you're interested to read more about this case study, please click here
If you want to know the difference between AI and ML. Click here