Implementing demand forecasting for a global healthcare distributor

InfoVision partnered with a global healthcare distribution leader, operating across 32 countries, to implement an AI-driven demand forecasting solution. This enabled precise weekly order predictions, optimized inventory management, reduced costs, and enhanced customer satisfaction.

About the customer

The customer is a global leader in healthcare product distribution.  Operating across 32 countries with an extensive customer network, the business model required precise inventory management and order prediction capabilities to run at optimum efficiency.

Business need: Precision inventory management

The customer sought to transform their inventory management system through advanced predictive modelling. The project aimed to:

  • Forecast demand for 1,000 top products, identified through recency-based analysis

  • Predict weekly orders for 300,000 active customers in the distribution network

  • Optimize inventory levels with weekly prediction cycles

The primary objectives were to enhance inventory optimization, minimize stockouts, reduce operational costs, and elevate customer satisfaction. This data-driven approach promised to transform the company's supply chain efficiency and overall customer experience.

Solution delivered

InfoVision implemented a comprehensive AI-driven demand forecasting solution leveraging advanced analytics:

Data preparation and processing

  • Utilized 13 months of historical data (September 2021 to September 2022) for model training.
  • Validated accuracy with October 2022 data checks.
  • Optimized BigQuery ETL pipelines for processing customer-product records.

Customer segmentation

Applied RFM (Recency, Frequency, Monetary) analysis for nuanced customer segmentation: - Recency: Recent purchase dates | Frequency: Order patterns | Monetary: Revenue contributions

Feature engineering

Developed 40 production features, incorporating:

  • Historical order trends
  • Product-specific demand cycles
  • Customer-product interaction patterns

Model development

  • Primary Model: XGBoost with hyperparameter tuning for large-scale accuracy.
  • Secondary Model: PyCaret Random Forest for validation and benchmarking.

Model maintenance

  • Weekly retraining with updated data.
  • Continuous performance verification and iterative refinements.

Tech stack

  • Cloud Platform: Google Cloud Platform
  • Data Warehouse: BigQuery
  • Machine Learning: XGBoost, PyCaret
  • Programming: Python

Business impact

Delivered transformative AI-driven outcomes including:

  • Prediction accuracy: 90% predictive accuracy enabling precise inventory management
  • Operational efficiency: 4X accelerated prediction processing through BigQuery optimization.
  • Cost optimization: 40% reduction in total cost of ownership (TCO) by minimizing stockouts and enhancing operational efficiency.

Transform your business with data-driven decisions that enhance customer satisfaction and operational efficiency at scale. Reach out to us at digital@infovision.com if you have a similar case to discuss.