Reinventing Forecasting and Smart Scaling

Situation

A global technology company faced challenges in forecasting total warranty repairs for their diverse product portfolio. The client wanted a forecasting framework that was scalable, automated, and explainable. The company’s primary objective was to ensure accurate accruals of repair costs for the finance team.

Challenges

Multiple obstacles in forecasting warranty repairs led to setbacks in financial planning and inefficient decision-making.

• Lack of data: The absence of structured historical data made forecasting difficult.
• Manual & time-consuming process: Generating a forecast took 15 days, delaying critical financial decisions.
• Lack of explanation for variations: Forecast fluctuations lacked clear business-driven explanations, reducing trust in the process.

Solution

To address these challenges, we designed a robust forecasting system that:

• Adopted Generalized Additive Models (GAM): These models efficiently captured non-linear trends using segmented curves.
• Created a Man-Machine Framework: This hybrid system allowed automated forecasts while enabling business rule interventions.
• Provided a Single Forecast Figure: A monolithic forecast number gave the finance team a reliable accrual estimate.

Impact

• Increased forecasting accuracy to 95% – 98%
• Reduction in forecasting time from 15 days to 5 hours
• $200M in savings

Scalability: Extended the forecasting model beyond finance

The adoption of the forecasting system was limited. Other teams, like inventory planning and product quality monitoring, also impacted by repairs, relied on separate forecasts, leading to fragmented decision-making.
So, we scaled our solution.

• Developed a single forecasting system to meet the needs of finance, inventory, and product quality teams.
• Combined GAM, Linear Regression, and Exponential Smoothing to account for sparse data and seasonality.

Expanded Outcome

• $130 M savings through improved inventory management.
• Optimized shipping during repair logistics transfers.
• 25% faster recall program due to early defect detection.