Share

Client

The client is a U.S.-based retail chain operating more than 50 physical stores alongside an e-commerce channel. The company manages product inventory and sales across both offline and online channels, requiring demand planning at the individual product and store levels.

Its forecasting process relied heavily on manual work. Managers exported sales and inventory data from SAP ERP into Excel, used basic formulas or previous-year figures, and adjusted forecasts based on experience. Promotional activity was not consistently incorporated into demand planning.

Partnership goal:
  • Improve demand forecasting accuracy. Replace manual Excel-based forecasting with ML models that incorporate sales, inventory, promotions, and e-commerce data at the SKU and store levels.
  • Optimize inventory planning and replenishment. Reduce excess stock and stockouts while providing planners with accessible forecasts and ERP-integrated replenishment capabilities.

Challenge

The retailer’s manual forecasting process made it difficult to accurately predict demand across stores and individual SKUs.

Reliance on historical sales figures, Excel calculations, and subjective adjustments resulted in inconsistent forecasts, particularly when promotional activity affected purchasing patterns.

These inaccuracies contributed to excess inventory for some products and stockouts for others, increasing storage costs and causing missed sales opportunities.

The client needed a more accurate and automated forecasting approach that could combine data from multiple sources and provide actionable insights for supply chain and category planning teams.

Solution

What We Built

Techbar developed an ML-driven demand forecasting system that generates predictions at the SKU and store levels.

The solution integrates sales, inventory, promotional, and e-commerce data from SAP Business One, Shopify, and the client’s internal promotions system. Azure Data Factory and Apache Airflow manage ETL pipelines, while the existing PostgreSQL data warehouse stores the consolidated information.

Demand forecasting models based on LightGBM and Prophet were implemented using Azure Machine Learning, with weekly retraining to incorporate updated data and demand patterns.

Forecasts are delivered through Power BI dashboards, allowing planners to review demand predictions without technical expertise. A REST API integration with the ERP system supports automated replenishment based on forecast data.

How We Worked

Techbar collaborated with the client’s supply chain and category management teams to validate forecasting results against operational requirements.

The system was rolled out across all physical stores and the e-commerce channel within six months.

Following implementation, monthly forecast accuracy reviews supported ongoing model evaluation and improvements, while weekly retraining kept forecasts aligned with updated sales and inventory data.

Results

The ML-driven forecasting system improved forecast accuracy by 20% compared with the retailer’s previous Excel-based methods.

More accurate demand predictions helped reduce excess inventory and associated storage costs while decreasing stockouts and supporting higher sales of top-performing products.

Power BI dashboards provided planners with clear demand insights, reducing reliance on manual forecasting and subjective adjustments.

The solution was deployed across more than 50 stores and the e-commerce channel within six months, with ERP integration supporting automated replenishment workflows.

More Client StoriesView all case studies

Do Not See Your Scenario Here?

Most of our work stays under NDA. Describe your engineering challenge and we will send relevant examples along with pre-vetted candidate profiles within 48 hours.

Dedicated software engineers