AI analytics and better forecasting lift FMCG supply-chain performance: practitioner survey links stronger AI capability to higher efficiency, while optimization models show routine constraint-based planning can reduce costs by about 10% and raise service levels even during demand surges.

Main Finding

AI-enabled analytics capability—measured as an organizational capability combining data, people, governance, and tools—is a strong, measurable driver of FMCG supply chain efficiency in both normal and high-demand regimes. Survey evidence (n=168) and prescriptive LP/MIP optimization runs converge: improving analytics, forecasting, inventory practice, and supplier coordination materially lowers total cost, raises service levels, and reduces stockouts, and constraint-based optimization remains feasible and robust under demand surges and modest forecast noise.

Key Points

• Sample & measurement: Survey of 168 supply-chain professionals; constructs showed acceptable reliability (Cronbach α 0.79–0.88). Mean scores: AI analytics capability 3.84 (SD 0.62), forecasting effectiveness 3.71 (SD 0.66), overall efficiency 3.52 (SD 0.67).
• Statistical associations:
  • Supply chain efficiency correlated strongly with AI analytics capability (r = 0.62, p < 0.001).
  • Multiple regression explained 54% of variance in efficiency (R² = 0.54, p < 0.001).
  • Standardized predictors (betas): AI analytics capability 0.34 (p < 0.001), forecasting effectiveness 0.21 (p = 0.001), inventory optimization practice 0.17 (p = 0.005), supplier coordination 0.14 (p = 0.017). Logistics planning positive but not significant (p = 0.075).
• Optimization (LP/MIP) validation:
  • Under normal demand: total cost fell 11.8% (from 1.72M to 1.52M), service level rose 92.1% → 96.0%, stockouts down 18.4%.
  • Under high demand: total cost fell 9.3% (2.05M → 1.86M), service 88.4% → 93.2%, stockouts down 15.1%.
  • Solutions feasible across scenarios; reported optimality gap 0.8%.
  • Robustness: ±10% forecast noise led to only a 2.1% cost increase and 0.9 percentage-point service decline.
• Conceptual point: AI (ML/forecasting) primarily improves input parameters and regime identification; LP/MIP remains the prescriptive decision core for constrained operational choices. Value arises from the complementarity of prediction + constrained optimization, embedded in disciplined governance and execution.

Data & Methods

• Research design: Quantitative, cross-sectional, case-based study combining perceptual survey data with prescriptive optimization models calibrated to the same FMCG planning context.
• Survey:
  • N = 168 supply-chain professionals.
  • Constructs: AI analytics capability, forecasting effectiveness, inventory optimization practice, logistics & distribution planning quality, supplier coordination, lead-time reliability, and perceived supply chain efficiency.
  • Analyses: reliability testing (Cronbach’s α), descriptive stats, Pearson correlations, multiple regression.
• Optimization:
  • Cases: cloud and enterprise planning cases modeled under normal and high-demand regimes.
  • Formulations: Linear Programming for continuous allocation/flow decisions; Mixed-Integer Programming to capture discrete operational choices (e.g., shipments, activation of policies).
  • Constraints represented realistic capacity and lead-time limits; objective included total cost and service/stockout penalties.
  • Diagnostics: feasibility checks across regimes, optimality gap reporting (0.8%), and sensitivity/stress tests (±10% forecast noise).
• Validation approach: triangulation—compare perceptual/statistical relationships with objective optimization outcomes to strengthen causal plausibility.

Implications for AI Economics

• Measurable returns to analytics capability: The study quantifies how organizational AI analytics capability translates into cost and service improvements, supporting ROI arguments for investments in data, models, and governance—especially in volatile, high-demand FMCG markets.
• Complementarity and bundling: Value accrues not from ML alone but from bundling forecasting with constraint-aware prescriptive optimization (LP/MIP) and disciplined inventory execution. Economics of adoption should account for complementary investments (OR modeling, process change, training, governance).
• Robustness lowers downside risk: Small sensitivity to ±10% forecast error suggests moderate robustness; firms and policymakers can treat prescriptive systems as resilient against typical forecasting noise, reducing risk premia on working-capital buffers.
• Efficiency vs. resilience trade-offs: Optimized plans reduce cost and stockouts simultaneously, indicating that properly designed AI+optimization can relax the classical trade-off—however, feasibility checks revealed binding constraints, so benefits depend on accurate representation of capacity and lead-time limits.
• Policy and labor implications: Organizational capability matters (skills, governance), implying labor-market demand for analytics, optimization, and operations expertise. Public and private actors might prioritize workforce development and standards for model transparency/auditability.
• Competitive and market structure effects: Firms that operationalize AI+LP/MIP effectively can sustain higher service levels and lower costs during demand surges—potentially increasing competitive asymmetries in FMCG markets and affecting pricing/availability across regions.
• Research and evaluation guidance: For economists evaluating AI interventions, the paper illustrates the value of combining perceptual (survey) measures with counterfactual or prescriptive model outputs to estimate treatment effects under realistic constraints and uncertainty.

Limitations to note for economic extrapolation: cross-sectional, case-based design in an FMCG/high-demand setting (context-specific); external validity requires replication across product categories, geographies, and differing network structures.