Purpose and context
Humanitarian aid increasingly operates in continued crises, where response efforts are longer-term and less reactive than for sudden disasters. In these contexts, the demand for goods and services tends to be more stable and predictable, creating opportunities for supply chain planning through better demand forecasting. Demand planning ensures an efficient resource allocation, supports responsiveness to the needs, and drives strategic decisions on sustainability (e.g. local sourcing, pooled procurement). Yet, our previous research (t’Serstevens et al., 2024) underscores that demand planning remains limited within the humanitarian sector, with little investment in data analytics and forecasting methods.
This paper challenges the prevailing assumption that humanitarian demand is inherently unpredictable. Through a case study from a hulo member humanitarian organization, we explore the potential of time-series forecasting—widely used in the commercial sector—as a data-driven approach to support demand planning. Our goal is to encourage greater adoption of data analytics as a key component of demand planning and broader supply chain planning. We also hope to inspire a shift towards more data-driven decision-making in humanitarian supply chains, paving the way for operational excellence.
Key findings
Analyzing four years of procurement data for medical items, we find that while humanitarian demand can be erratic, accurate forecasting is possible in the humanitarian sector. The key learnings include:
1. The forecast accuracy depends on the product granularity and the time horizon Short-term forecasts (e.g., for Q1) are generally more reliable than annual projections, particularly at the item level, where a portion of the predicted volume showed strong accuracy. However, a significant share of the demand remains difficult to predict, highlighting the need for a segmented approach. Surprisingly, the aggregated forecasts at the primary category level were less accurate than the item-level forecasts, suggesting that the product categorization should be carefully reviewed.
2. Data quality has an impact on the forecasting quality, and outlier detection can help achieve stronger accuracy Inconsistent procurement data distorts the forecasts, reinforcing the need for structured data validation, high data quality, and actual demand data collection. While outlier detection reduces the extreme forecast errors, it can also over-smooth the demand fluctuations, affecting the accuracy. A targeted approach to identifying and correcting the data entry errors could further improve the forecasting reliability.