Prior to the supply chain disruptions of recent years, Polaris Inc., a leading provider of powersports equipment, forecasted production and shipments based on innovation and market demand. However, since these disruptions occurred, the business environment became constrained by supply. Recognizing a need for more speed and agility across planning processes, the Polaris Finance team turned to the power of OneStream’s Sensible Machine Learning (Sensible ML) solution to assist with demand forecasting.
In previous years, Polaris’ business units had relied on a highly manual financial planning model with inputs such as SIOP-generated shipped unit forecasts by product, product costs and MSRPs, freight cost, and dealer discounts to arrive at a gross margin view. This model was referred to as the “Driver-Based Revenue Model,” and it provided the perfect opportunity to incorporate machine learning-driven forecasting and transition to a unified planning process within OneStream.
Polaris decided to focus their Sensible ML project on their North American Off-Road Products GBU, looking at a 12-month forecasting time horizon with a focus on variables impacting their Shipped Units forecast. These variables included Commodity Prices, Presold Orders, “Clean Build” Percentage and Build-to-Ship Durations. Historic data representing these variables would be combined with historic shipped units to generate the ML models and their forward-looking forecasts.
The historic data model covered 181 products, with weekly units sold from 2016 through 2022. Sensible ML crunched through this data, combined with commodity prices for steel and aluminum, factored in events such as holidays, and generated over 2,800 models for comparison. The OneStream ML models proved to be the most accurate, based on the historic data. The ML forecasts were run monthly and were incorporated into a driver-based forecast.
The results were impressive. Not only were the forecasts more accurate than with prior approaches, but with Sensible ML, Polaris added speed and efficiency to their forecasting processes, reducing forecasting cycles from days to hours. Polaris also now has more transparency into what’s behind the ML models, including insights into the key forecast drivers for more informed decision-making.
It provides a finance-run ML forecasting process that integrates seamlessly across planning and forecasting processes in the same user experience used for financial close and consolidations, account reconciliations and reporting.
“The ability to quickly generate driver-based forecasts is essential to adapting to our changing business conditions,” said Melanie Hermann, Director, Finance Process & Systems at Polaris Industries. “Incorporating AI into our planning and forecasting through the OneStream Sensible ML solution accelerates the forecasting process and further elevates it with powerful ML data-driven forecasts. Sensible ML forecasts have shown to be more accurate, and the Value-Add Dashboard provides the business users with insights into the key features driving the forecast to easily manage, improve and enhance the model.”
The Polaris Data Science team was impressed with the process and results. “Sensible ML commoditizes the part of my job that can be commoditized and allows me to focus on where I can add value… with the output that Sensible ML provides,” said Luke Bunge, Manager Data Science Product. “It’s an incredible timesaver and gets you to the best answer possible. The team did a great job immersing us in the tool…as opposed to turning it into a black box.”
For companies across fast-changing industries such as CPG manufacturing, retail and hospitality, Sensible ML reduces the traditional barriers to ML forecasting and improves both the speed and accuracy of demand planning. This enables organizations to fine-tune production plans, optimize inventories as well as reduce volatility and fluctuations in labor planning.
To lean more, download the Polaris Inc. case study and contact OneStream if you are ready to learn how your organization can take advantage of the power of machine learning.
Download the Case Study