Undisclosed Global Pharmaceutical Supplier

A global pharmaceutical supplier faced a demand forecasting crisis at scale: over 400 planners manually calculated supply requirements for more than 10,000 medical components and test kits shipped to hundreds of facilities worldwide. Each forecasting cycle consumed approximately seven days, and the manual approach produced inconsistent results that chronically missed desired accuracy targets. In pharmaceutical supply chains, where stockouts can delay diagnostics or patient care and overstock drives costly waste of time-sensitive materials, inaccurate forecasting carries consequences beyond operational inefficiency. The company had no automation, no scalable ML infrastructure, and its data sources remained siloed — with no mechanism to monitor forecast quality or model performance over time. The cumulative impact: annual losses reaching tens of millions of dollars.

RST Data Software designed and built an end-to-end automated ML forecasting pipeline replacing the manual process entirely. The system ingests historical demand and contextual data from the company's existing SAP BW environment into Snowflake, where it prepares ML-ready datasets at scale. A model selection layer evaluates both statistical models and neural networks to identify the best-performing approach for each product-location pair, then routes monthly supply predictions — covering over 6,000 products across thousands of global facility locations — directly back into SAP for operational use. Crucially, the pipeline retrains itself automatically each month using the latest demand trends, logistics constraints, and contextual signals such as regional viral outbreaks or seasonal surges. No manual intervention or rework is required when conditions shift, and the architecture is designed to extend to new products and facilities without additional tooling.

The automated pipeline compressed a seven-day manual planning cycle into a 1.5-hour fully automated workflow — a 10× improvement in forecasting speed — while forecast accuracy consistently reached 80%, representing up to a 5 percentage point gain over the prior baseline. The financial impact is material: higher accuracy unlocked up to $50M in annual waste reduction across global supply chain operations by eliminating systematic over- and under-supply of medical components. The system now covers 60% of the company's global portfolio without requiring additional planners or duplicate effort across local markets. Planners previously consumed by routine forecast cycles were redeployed to exception handling and strategic planning, compounding operational gains beyond the headline speed metric.

• Unified data infrastructure — consolidating SAP BW inputs into a governed Snowflake environment — is a prerequisite for ML forecasting at enterprise scale; model quality is bounded by data accessibility.
• Self-retraining monthly pipelines outperform static models in pharmaceutical supply chains, where demand can shift abruptly due to outbreaks, regulatory changes, or regional logistics disruptions.
• Automating high-volume routine forecasting does not eliminate the need for human planners — it redirects them toward higher-value exception management, improving both speed and judgment quality.
• Covering 60% of a global portfolio without headcount growth demonstrates that ML pipelines scale economically in ways manual planning cannot; architecture choices at design time determine how far that coverage can extend.