Does Bigger Data Make Better Plans? Decomposing the Value of Big Data Analytics in Strategic and Systematic Production Planning under Demand Uncertainty

Abstract

Big data analytics (BDA) is widely argued to improve strategic and systematic planning, yet the empirical evidence relates analytics capability to firm performance without isolating which component of an analytics system actually creates the value. This study decomposes that value for capacity and production planning under demand uncertainty. Seven planning regimes, ranging from judgmental annual planning to a full BDA regime combining cross-series machine learning, an external leading indicator, probabilistic forecasts and monthly revision, are embedded in an identical rolling-horizon two-stage stochastic planning model and evaluated on 200 real monthly industrial demand series (180 months each) drawn from the M4 competition. Because execution is identical across regimes, differences in realised performance are attributable solely to planning information. The BDA regime reduced planning-relevant cost per unit by 30.4% relative to judgmental planning and by 11.9% relative to quarterly statistical planning, and captured 70.7% of the gap between judgmental planning and a perfect-information bound. Decomposition shows that formalising the forecast contributes 48.8% of that gap, big-data forecast quality a further 21.0%, and representing uncertainty explicitly 13.8%, whereas increasing the revision frequency from quarterly to monthly does not help and slightly harms performance while increasing plan instability. Superior forecasts used deterministically were statistically indistinguishable from a simple quarterly statistical plan. The advantage of BDA increased with demand volatility (4.8% to 17.2% across volatility terciles) and with the cost of shortages, and decreased when capacity lead times exceeded the horizon over which the external signal was informative. The results indicate that analytics investment should prioritise decision-quality elements, namely uncertainty representation and cost-aware optimisation, over data velocity.