Sales Prediction System

Who is the Client?

Moto-Profil Ltd. specializes in the import and distribution of car spare parts and accessories, as well as equipping car workshops. Through a network of over 1,400 business partners, the company reaches more than 15,000 workshops across Poland and Central Europe, emerging as a solid industry leader in recent years.

Goal of the Project

The project's goal was to develop and implement a system capable of predicting demand for Moto-Profil's distributed products. This would allow for optimal purchasing—ensuring sufficient stock to meet anticipated orders while avoiding overstocking, thus reducing storage costs.

Result

The developed sales prediction system is extensive, non-monolithic structure. It is based on multiple microservices responsible for:

• collecting data from various sources, including internal systems such as multiple client databases and dedicated interfaces, as well as external sources,
• transforming and processing the data,
• integrating the data,
• storing the data,
• predicting sales trends using artificial intelligence, including neural networks, recurrent neural networks, and convolutional networks.

Furthermore, the system provides data visualization through an informational portal and integrates seamlessly with the ordering system.

Technologies

• ASP.NET Core
• .NET Core
• Python
• AI
• ML
• DeepLearning
• Keras
• Darts
• MS SQL
• Redis
• Docker
• Cloud
• Azure
• Mongo
• Terraform

Solutions

The sales prediction system is like an iceberg – only a small part is visible to the end user. Most of the system operates in the background on a vast cloud infrastructure, making its presence invisible to users. The user, however, can access the system through a dedicated online portal or order integration system, where they can place orders based on the system's suggestions.

The client's goal is to minimize warehouse stock. Items in stock should suffice for sales but not exceed necessary levels to avoid surplus. The key to achieving this ambitious goal is predicting future sales, product availability from manufacturers, and delivery times – tasks managed by the system.

Thanks to AI-driven sales trend predictions based on historical analysis and external factors, combined with predictions of product availability and delivery times from manufacturers, the system can accurately forecast sales trends for the coming weeks and optimally stock the warehouse. This solution ensures smooth sales operations, prevents potential losses, and minimizes storage and purchase costs.

Challenges

• Developing an AI model for sales trend prediction
• Integrating data from a large number of diverse systems
• Processing vast amounts of data
• Achieving satisfactory prediction accuracy to avoid financial losses for the client

Future

The system has been successfully implemented into production. Currently, we provide comprehensive technical support services. The system is also continuously developed and improved, including enabling sales predictions for products with very short sales histories.