Problem Statement
FMCG demand forecasting efficacy is fundamentally constrained by current AI/ML models that are insufficiently dynamic and fail to integrate the influence of crucial exogenous variables (such as macroeconomic fluctuations, population shifts, and market disruptions) and the wave coming from social media.
While the automation of forecasting has delivered incremental improvements, the lack of an externally aware predictive framework compromises model robustness and leads to suboptimal inventory and resource decisions. Quantifying the benefit, the strategic integration of these factors into AI/ML models is expected to yield substantial returns, potentially translating to a 5-10% reduction in forecasting error and the associated cost of lost sales or excess inventory, compared to current internal-data-only methods.
Research Question
How do key exogenous variables (macroeconomic, demographic, and disruptive) impact the predictive accuracy of AIML algorithms for FMCG demand forecasting, and how should the model be augmented to contextualize these factors for improved efficacy?
Research Methodology
The research approach involves three core steps:
- Data Gathering: Collect data on product demand alongside the identified relevant exogenous variables.
- Algorithm Development: Develop and train AI/ML algorithms.
- Accuracy Assessment: Assess the predictive accuracy of the algorithms when the exogenous variables are considered versus when they are excluded, as well as the learning feedback loops.
Key Results
The analysis of the AIML algorithm for FMCG demand forecasting revealed two key points:
- Efficacy of Augmentation: The Augmented Prophet Model (APM), incorporating macroeconomic and other external variables, proved superior, reducing MAPE by an average of 11.2% compared to the baseline.
- Human-AI Pairing is Critical: The success of the APM relied heavily on human domain expertise for selecting and contextualizing optimal variables and interpreting results, underscoring that superior forecasting requires the synergy of experts and AI.
References
Macias, J., et al. Integrating Exogenous Variables into Demand Forecasting: A Value Quantification Approach. (Work in progress)