Improved purchase order forecasting & expense category volatility to create reliable forecasting regimen

About the client

A leading US healthcare provider

Problems:

• The customer was witnessing a significant variance in the Purchase order (PO) issued and total spend at the hospital level resulting in incorrect capital allocation.
• Lack of granular forecasting of PO submissions by its hospital partners broken down by on-contract, off-contract, and expense categories (EOCs)
• Poor understanding of PO volatility across expense categories (i.e., EOCs) for each hospital

Solution:

• Innover developed ensemble ML model (XGB + Seasonal ARIMAX) to accurately capture the Total PO trend, including weekly/monthly growth and seasonality.
• Incorporated lagged values of the previous three weeks to account for recency effect
• Addressed week-to-week volatility through noise component adjustment
• Developed causal models to estimate EOC level share of expense and the week-over-week volatility index