Using Machine Learning to Reduce Long-Held Stock by 34% for a Car and Spare Parts Importer
About
A car and spare parts importer used machine learning to support product configuration and ordering decisions at individual points of sale. The classification model estimated the probability of selling a configured product within a defined period and contributed to a 34% reduction in products remaining in stock for too long.
Client
The client was an importer and wholesaler of cars and spare parts. Individual points of sale configured products and placed orders for their local markets.
The challenge
Ordering decisions were made independently by each point of sale. Employees had limited access to data and often relied on personal judgement when selecting product configurations.
The company could not estimate how long a specific configuration would remain in stock at a selected location. This increased the risk of ordering products with lower local demand and keeping them in inventory for extended periods.
The goal was to provide sales representatives with data-based recommendations during product configuration and order placement.
Solutions
SoftwareHut analysed historical stock data covering hundreds of thousands of products from the previous two years.
The analysis identified inconsistencies caused by differences in data entry. Workshops with the client’s business specialists helped prepare and validate the data.
The team tested regression, classification and clustering models using datasets with different levels of complexity. Classification models produced the most useful results because they could estimate the probability that a product would be sold within a specific period.
The selected model was implemented using SoftwareHut’s AI Farm software and added to the product configuration and ordering process.
Based on configuration parameters and the location of the point of sale, the model could:
- estimate the probability of sale,
- suggest changes to the configuration,
- warn that a product might remain in stock too long,
- recommend abandoning a configuration.
The model also supported employee decisions.
Results
The number of products remaining in stock for too long decreased by 34%.
Sales representatives gained access to recommendations based on two years of historical stock data instead of relying only on personal judgement.
Do you have any questions or ideas?
Drops us a line and we'll contact you about your project.
