Optimising Inventory with AI

Problem

Agentico was asked to optimise the parts inventory of a machinery business with 25,000 lines in stock over 4 sites.

Stock demand was driven by seasonal factors such as rainfall and was uneven across sites. Furthermore, demand for some components was not market driven but set by warranty alerts from manufacturers.

Finally, suppliers offered substantial discounts for orders made six months in advance of seasonal demand, the business made profit almost exclusively on parts ordered during that window.

Solution

A stock order proposal system was developed through extensive business process analysis and stakeholder consultation.

The system employs an ensemble machine learning model combining both neural network and traditional statistical approaches to forecast demand.

This was augmented with custom logic for warranty handling and stock distribution optimization.

The solution was implemented gradually, starting with stakeholder interviews, moving through prototype testing, and culminating in a full deployment with ongoing user support and maintenance.

Recipe

• Process:

  • Microsoft Team Data Science Process

• Model:

  • 1D convolutional neural network for time series forecasting, in ensemble with ARIMA models

• Data pipeline:

  • Extraction from source system (proprietary database) to MS SQL Server

• Deployment:

  • AzureML

• Comms:

  • Email system for user communications

• Impact:

  • The model saved 10% in required stocking levels and reduced parts obsolescence, paying for itself in the first year of usage.