Operationalising advanced demand forecasting in resource-constrained SMEs: evidence from a large-scale deployment in a packaging supply chain

Purpose – This paper examines how advanced forecasting analytics can be operationalised within resource-constrained SME supply-chain environments characterised by heterogeneous and intermittent demand, sparse histories and long-tail SKU portfolios.

Design/methodology/approach – The study reports a large-scale empirical deployment conducted through a multi-year Knowledge Transfer Partnership with a UK-based sustainable packaging supplier managing 1,474 SKUs. The proposed framework integrates demand-type-aware local statistical and intermittent-demand models with pooled global machine-learning and deep-learning approaches within a unified and auditable forecasting pipeline. Forecasting candidates are evaluated using rolling-origin cross-validation and selected through an operationally aligned champion selection mechanism balancing accuracy, bias and computational feasibility.

Findings – The adaptive framework substantially improves forecasting performance relative to the firm’s incumbent rule-based approach, with the strongest gains observed for intermittent and short-history SKUs. The findings further demonstrate that no single forecasting paradigm dominates across all operational settings. Instead, forecasting effectiveness depends strongly on demand structure, history length and operational context. Global models provide the greatest value for sparse and data-limited series, while local statistical approaches remain highly competitive for stable and mature SKUs.

Originality/value – This paper contributes to operations and supply-chain management literature by reframing forecasting as an operational capability rather than solely a prediction task. It provides rare empirical evidence on how advanced forecasting analytics can be embedded within SME planning environments and demonstrates the importance of adaptive forecasting governance in long-tail demand settings.