Polaris Uses Machine Learning to Power Up Financial Planning
Polaris, a powersports vehicle maker, turned to machine learning to improve its financial planning and forecasting processes. Here’s how the company approached the new technology.
![financial charts, business team working on computer financial charts, business team working on computer](https://eu-images.contentstack.com/v3/assets/blt07f68461ccd75245/blt0e9a93c118bbd861/6617f96cfe97402b0f82b4d5/financial_20charts_20business_20team_20working_20on_20computer.jpg?width=1280&auto=webp&quality=95&format=jpg&disable=upscale)
Powersports enthusiasts know the importance of well-crafted and innovative on- and off-road vehicles for a peak experience. What they might not appreciate is the effort that goes into the underlying technology.
Polaris, a powersports vehicle manufacturer based in Medina, Minn., has spent the last several years expanding its business through acquisitions. At the same time, Polaris has also sought to upgrade its technological capabilities across the organization. For the finance department, specifically, those upgrades aimed to support several capabilities, said Melanie Hermann, Polaris finance director of process and systems. The key objectives were to enhance forecasting for parts and materials, fine-tune projections for revenue and costs, foster better collaboration with partners, and facilitate the company’s growth trajectory.
Polaris’ finance team paved the way for achieving its goals by focusing first on devising a more efficient systems management strategy. For example, the company operated on about 15 ERP systems, most of which were obtained through acquisitions over the years. Having so many ERP systems posed challenges in terms of holistic planning and forecasting at an enterprise level.
Revamping ERP Strategy
Polaris decided to tackle its ERP complexity by moving off from Oracle Hyperion systems for capital expense planning and financial management. The finance team opted to use equivalent systems from OneStream. The shift involved the migration of data between applications and the upkeep of metadata across those platforms.
Next up, the finance team’s focus turned to replacing its manual processes for demand forecasting with a more intelligent, automated system. The finance team began to explore DataRobot's AI and machine learning tools, which were already being used by Polaris’ data science department.
Polaris desert and dune off-road vehicles
According to Rob Kugel, senior vice president and research director at Ventana Research, AI and machine learning present compelling use cases for an organization’s finance department.
“Today, we spend so much time preparing a budget, and some of that is just organizational,” Kugel said. “But what if we could [reduce the time] significantly? In theory, instead of taking two or three months to put together a budget or annual forecast, you could do it in a few weeks.”
Soon after the finance department started experimenting with AI and ML technology, OneStream introduced a machine learning product called Sensible ML. The product is not only fully integrated with the existing OneStream platform but also with DataRobot. Hermann saw the launch of Sensible ML as a sign – it was time to seriously commit to incorporating AI and ML technologies into Polaris’ financial operations.
“We wanted to understand better what indicators drive our business, correlate better with all of our businesses, and do a better job with financial planning,” Hermann explained. For example, the team aims to model what its financials might look like under different scenarios over the next two years.
Testing Out Sensible ML
To determine whether Sensible ML would meet its requirements, Polaris tested various scenarios, focusing on the off-road vehicle segment of its business. Testing began with the ingestion of raw data from a mixture of sources, including ERP systems, its Snowflake data warehouse, and selected outputs from DataRobot, into the OneStream relational database.
The team then used that data to train the model by combining historical patterns and data for similar products with information already present in the OneStream database (e.g., presold orders and build-to-ship durations), explained Matt Kohorst, a finance system architect at Polaris.
Additional experiments involved the inclusion of economic indicators like steel and aluminum pricing. The results were then incorporated into a driver-based forecast.
“You can load up a non-machine learning forecast and then run comparisons over a long time horizon to compare to what the machine learning is generating, and then you can formally score it to see which is more accurate,” Kohorst said. “We could see that the ML system over this known historical period is forecasting more accurately than our traditional non-ML forecast. On that basis, you know you’re going down a good path.”
Embracing ML Approach
Once the finance team became convinced that the ML approach was the way to go, it moved forward with what Hermann referred to as a “2.0 version”. That meant migrating all data and related Extract, Transform, and Load processes into OneStream using OneStream’s relational engine and data integration tools. The migration included additional internal data around warranties and promotions, traditionally considered drivers of the business. “The more types of data you add, the more accurate the model will be,” Hermann said.
The next step is ensuring that the entire process is automated and works smoothly so it can run monthly. Once that happens, the team plans to transition from an 18-month financial forecast to a 24-month forecast. The team also plans to extend financial forecasting from the off-road segment of the business to the on-road segment.
About the Author
You May Also Like