Everyone wants their analytics investment to yield successful business results, but not everyone can show the kind of serious return on investment that captures the attention of C-suite executives. So when Southwest Airlines did a pilot program to look at the potential impact of investing in an analytics package, Doug Gray, director of analytical data services at the company, knew he needed to choose something with a big impact.
Of course, analytics work was nothing new to Southwest Airlines. The company had been working with analytics projects for about 20 years already. Such projects are absolutely necessary to manage the complexity of airline operations efficiently and cost effectively.
But with that complexity, there’s quite a lot of room to find more cost savings and efficiencies.
Airlines need their flights to depart and arrive on time. They need to manage a fleet of aircraft and have those aircraft be in the right places at the right times for their flights. They need to forecast customer demand for flights so that they operate that fleet of aircraft efficiently and profitably. They need to manage their crew member scheduling and know their crew member whereabouts for scheduling.
In the case of Southwest Airlines, that meant managing and forecasting 700 aircraft flying about 4,000 flights per day to over 100 national and international destinations.
Gray said Southwest Airlines uses over 100 terabytes of data and dozens of applications that involve analytics already, and when it looks to create a new application, it goes for the high value targets — the project has to yield millions of dollars in business value.
So for its pilot to test a new analytics platform and toolset in 2016, Gray’s group chose something that was the second largest expense for the company — fuel costs. Gray recounted the story of the project at the Gartner Data and Analytics Summit in March.
Depending on market pricing at any given time, every year Southwest Airlines spends between $4 billion and $6 billion dollars on fuel. That means any small percentage improvement in those costs will amount to a huge number.
Before the pilot began, Southwest Airlines’ fuel cost forecasting efforts relied on pulling information from multiple systems including Ariba, the Allegro fuel management environment, and then the company’s own enterprise data warehouse for historical data. Plus, Gray admitted, the company also had a lot of spreadsheets that the team used to store and manipulate quite a lot of data.
“We were feeding all that data into one big massive spreadsheet,” Gray said. With 100 airports served and running a rolling schedule for forecasting fuel every month of the year, the team was producing 1,200 fuel demand forecasts every month. It took one of the finance analysts three days each month to go through the process of generating forecasts that generally weren’t as accurate as the company would have preferred. So there was a lot of room for improvement.
The fuel consumption pilot project used Alteryx Designer, the platform’s gallery, and R, to build eight different predictive models that included time series regression modeling and neural networks, Gray said. For each month and each airport, the system was able to generate 9,600 forecasts. Gray couldn’t share the actual cost savings in dollar amounts publicly, but he said they were substantial. Benefits included reducing the amount of time required for data wrangling by 60%. Forecast accuracy improved. And Southwest learned it could negotiate a better deal on fuel if it gave all the business to a single vendor in Southern California rather than buying from multiple vendors there. The change also sped up the process of fuel purchasing.
Read the source article at InformationWeek.com.