Artificial Intelligence for Accurate Demand
Forecasting and Planning



ERP auto-generated forecast still required a need for manual adjustments by the operations team based on gut feel.

This complex and error-prone forecasting and inventory management processes led to overstock and out-of-stock situations that impacted sales.

Its ERP demand forecasting could not account for multiple warehouses, outliers such as data related to the global pandemic, customer groups or seasonality.


Cleansed data with ETL processes and data engineering to ensure data is in the right format for modelling.

Created an AI model with products and customer groups as separate series for more granular forecasting.

Developed ARIMA and exponential smoothing time series models and used RMSE to compare their performance.

Identified the most accurate model by series comparing forecast to actual.

Optimized model performance – reduced processing from 6 days to 11 hours.


Projected inventory cost savings of 10%.

Average monthly variance of forecast VS. actual reduced from –28.1% to -5.3%.

After a year in production, the AI forecast has less variance to actual compared to the ERP unadjusted forecast and the ERP manually adjusted forecast.

The operations team has more time to review forecasts and make more proactive decisions.

Faster model performance enabled faster decision making.

Improved inventory management, purchasing, and overall business planning resulted in cost savings and increased profitability.