Order Demand Forecasting for a Leading Automotive Supplier

Client

The client is an American automotive and e-mobility company. It is an original equipment supplier to most major automotive OEMs worldwide and develops and manufactures products for conventional gas, hybrid, and electric vehicles.

Challenge

The client was looking for a solution to manage large amounts of unstructured order and shipment data to optimize their inventory management operations.

LTM Solution

• Built two forecasting models with order and shipment data to optimize inventory level. 
• Built with the stacking of regression and time series approaches. 
• Performed data cleansing in Snowflake and prepossessed data in Python.
• Resampled data to weekly and daily frequency to validate the model performance. 
• Feature-engineered the dates into multiple independent variables to build a robust model.
• Implemented the salient features of the model, including 12 major Machine Learning (ML) methods, which cover 20 algorithms/runtime selection of best algorithm/self-learn and self-train model.
• Calculated threshold values with statistical methods by rolling 24 weeks of order data using Quartile values to identify the upper and lower threshold levels of customer orders and intern alert the abnormality of the customer behavior.

Business Benefits

• Adopted to individual parts/customers automatically, irrespective of the part count.
• Built converging models for shipment vs. orders and past vs. future.
• Identified statistical thresholds, patterns, and anomalies.
• Achieved 86% accuracy by implementing the forecasting models.

Tools

Snowflake |Python| ML | Qliksense