Demand forecasting using EZForecast

Manufacturer Transforms Forecasting Process With EZForecast

A leading heating systems manufacturer struggled with seasonal demand fluctuations, leading to production overloads in winter, downtime in summer, and inefficiencies from manual forecasting. These challenges caused workforce disruptions, high retraining costs, and underutilized capacity during the off season.

To address these challenges, the manufacturer partnered with Wavicle to leverage their advanced analytics expertise in improving predictive capabilities. As a part of the proof-of-value (PoV) phase, Wavicle utilized its proprietary EZForecast accelerator to develop a solution, demonstrating a seven-percentage improvement in forecasting accuracy. With this promising outcome, the solution has proven to be a valuable foundation for supporting the manufacturing company’s future efficiency and growth strategies.

Complex demand forecasting issues  

This manufacturer faced major operational challenges due to the seasonal nature of its business. Demand for heating systems surged during the winter, forcing the company to overutilize production lines, while the summer months saw demand plummet, leading to layoffs and retraining cycles for workers. This cyclical pattern resulted in inefficiencies, production disruptions, and increased costs.

Compounding these issues was the company’s reliance on a manual forecasting process, which failed to accurately predict demand for its portfolio of over 1,000 finished goods. This led to frequent stockouts during peak months and excess inventory during downtimes, impacting profitability. A significant portion of demand came from a key customer, which accounted for 25–30% of total demand. The manufacturer struggled to accurately forecast the key customer’s orders separately from those of other customers, further complicating its supply chain planning. 

Additionally, the manufacturer’s internal infrastructure was not equipped to support advanced analytics or machine learning solutions, creating a bottleneck in implementing modern forecasting solutions. This infrastructure gap meant that automation and predictive analytics, critical to solving their issues, could not be integrated into their operations.

Faced with inefficiencies from manual forecasting and seasonal demand fluctuations, the manufacturer turned to Wavicle to develop a comprehensive PoV to improve their demand planning process. Wavicle leveraged its EZForecast accelerator to develop the PoV, showcasing how automation and advanced analytics could enhance forecasting accuracy and streamline operations.

Building a demand forecasting PoV with EZForecast 

To address the challenges, Wavicle implemented its EZForecast accelerator during a structured PoV. This six-week initiative allowed the company to test the effectiveness of the demand forecasting PoV in a controlled environment.

The first step was to process the company’s historical data within Wavicle’s Azure-based environment, as the manufacturer’s infrastructure was not ready for advanced analytics. The solution was tailored to the company’s specific needs, providing monthly demand forecasts at a finished-goods level for both key and non-key customers. The EZForecast accelerator integrated internal historical data with external sources, such as Dodge Data, allowing the model to account for promotions, industry trends, weather patterns, and carried-over orders.

Wavicle employed a mix of univariate, multivariate, and transformer-based models to generate accurate forecasts. To evaluate model performance, Wavicle used its proprietary combined forecast error metric (CFEM), which assessed factors such as accuracy, error direction, and bias. The team also applied the rolling origin evaluation method, running the models across multiple timeframes to eliminate the possibility of results being influenced by luck. Results were shared via Power BI dashboards and CSV files, enabling direct comparisons with the company’s manual forecasts during the user acceptance testing (UAT) phase.

Wavicle’s demand forecasting PoV established a framework to address the manufacturer’s inefficiencies. By integrating multiple data signals and automating the forecasting process, EZForecast provided a scalable and adaptable solution for manufacturer’s business.

Wavicle’s demand forecasting PoV demonstrated the system’s ability to automate forecasting, reduce reliance on manual processes, and improve demand predictions, ensuring better alignment of production schedules with actual market needs.

Measurable improvements and path to long-term efficiency 

Wavicle’s proof-of-value phase proved to be a transformative step for the manufacturer. It demonstrated significant improvements, with an overall forecast accuracy uplift of seven percent compared to the manufacturer’s manual process that provided a solid foundation for operational improvements.

For high-demand products, which contribute 23–24% of overall demand, forecast accuracy improved by up to 15%. These results underscored EZForecast’s ability to handle the company’s diverse product portfolio, including low-demand and new products. The model’s ability to incorporate various data inputs, such as promotions, market trends, and weather signals, proved essential in generating more reliable forecasts.

With improved forecast accuracy, the manufacturer can expect to reduce production costs by optimizing line utilization, minimizing stockout risks during peak months, and avoid excess inventory during slower periods.

The success of the PoV phase has paved the way for next steps of developing a minimum viable product (MVP). It aims to focus refine the model further by incorporating additional data signals identified during collaborative sessions with the manufacturer’s team. The goal is to automate the process to reduce manual effort, improve process accuracy, and help the manufacturer fully adopt the EZForecast accelerator as a long-term solution.

Overall, by demonstrating the value of its approach, Wavicle has laid the groundwork for the manufacturer to transform its demand forecasting process, ensuring sustainable growth and operational efficiency.