A small Nairobi survey finds AI-driven demand forecasting strongly associated with better supermarket supply-chain performance, but the cross-sectional design and limited, non-random sample prevent causal claims.

Main Finding

AI-driven demand forecasting has a statistically significant and sizable positive effect on supply chain performance in large supermarkets in Nairobi City County. The study reports an estimated regression coefficient β1 = 0.714 (p = 0.000 < 0.01), indicating that higher adoption/effectiveness of AI demand-forecasting predicts better supply-chain performance outcomes.

Key Points

• Scope and focus: Effect of AI-demand forecasting on supply chain performance in 10 large supermarkets in Nairobi City County, Kenya.
• Theoretical framing: Technology Acceptance Theory (Davis, 1989) and the Hybrid Intelligence model (Hoos, 2016) — emphasis on user acceptance and human–AI augmentation rather than full automation.
• Descriptive results:
  • Accuracy of sales predictions: mean = 4.19 (SD = 0.62) on 5-point Likert items, indicating strong agreement that AI tools improve sales-prediction accuracy.
  • Reduction in stock-outs: mean = 3.63 (SD = 1.04), respondents generally agreed AI forecasting reduced stock-outs but with more variation.
• Inferential result: Regression coefficient on AI-demand forecasting β1 = 0.714; p-value = 0.000. Authors conclude AI-demand forecasting significantly improves supply-chain performance (operationalized via cost reduction, improved efficiency, and real-time decision-making).
• Data collection: primary data from structured questionnaires administered to supply-chain staff; questionnaire pretested on 7 respondents from two Naivas supermarkets (Kiambu County).
• Sample: target/sample = 70 employees across supply-chain departments of the 10 largest supermarkets in Nairobi City County.
• Analysis tools: descriptive statistics (percentages, means, SD), correlation and regression analyses using SPSS v30.

Data & Methods

• Research design: Positivist, cross-sectional, descriptive design using structured questionnaires (closed-ended Likert items).
• Population and sample: Employees in supply-chain roles (managers, supervisors, officers) across 10 large supermarkets in Nairobi; n = 70 (the study treats this as the study sample/target population).
• Pretest: 7 respondents from Naivas supermarkets in Kiambu County.
• Measurement:
  • Independent variable: AI-demand forecasting (measured via items on accuracy of sales predictions, use of historical and real-time POS data, market-trend incorporation, forecasting lead times).
  • Dependent variable: Supply chain performance (measures stated include cost reduction, improved efficiency, and real-time decision-making; paper reports perceived improvements).
• Statistical analysis: Summary statistics (means, SDs), correlation, and regression analysis (SPSS v30). Reported regression: β1 = 0.714, p = 0.000.
• Limitations in reported methods (implicit/observable):
  • Cross-sectional, observational design — limits causal claims.
  • Small, localized sample (70 employees in 10 supermarkets) — limits external validity.
  • Reliance on self-reported Likert-scale measures and perceptions rather than firm-level objective performance metrics (sales, inventory turnover, costs).
  • Paper does not report model specification details (controls, R², robustness checks, potential endogeneity handling), so estimate interpretation is subject to omitted-variable and reverse-causality concerns.

Implications for AI Economics

• Productivity and value creation: The study provides micro-level evidence that AI demand-forecasting is associated with measurable improvements in supply-chain performance (accuracy, fewer stock-outs, faster decisions). For economists, this supports models where AI adoption raises firm productivity through improved matching of supply to demand.
• Consumer welfare and market functioning: Better demand forecasts reduce stock-outs and overstocks, likely improving consumer surplus (higher availability, fewer lost-sales) and reducing waste. At scale, these effects could alter retail pricing dynamics, inventory costs, and market competitiveness.
• Investment and diffusion in developing economies: Positive evidence from Nairobi supermarkets suggests potential high returns to investment in AI analytics even in lower‑income settings, conditional on data availability and human capacity to integrate tools. However, realized gains will depend on integration costs, data quality, and organizational adoption.
• Labor and hybrid outcomes: The paper emphasizes hybrid intelligence and augmentation; implication is firms may realize productivity gains without wholesale layoffs if human-in-the-loop workflows are adopted. Economists should consider heterogeneous labor effects (upskilling vs. displacement) and distributional outcomes across occupations.
• Costs, heterogeneity, and scalability: While the estimated effect is large, broader policy and economic appraisal should include implementation and training costs, integration complexity, and varying returns across firm sizes and contexts. Mixed findings in prior literature (some negative or insignificant effects) point to important heterogeneity in net benefits.
• Policy and infrastructure priorities: To maximize economic gains from AI in retail supply chains, targeted policies could support (a) data infrastructure and POS integration, (b) training programs for supply-chain staff (to enable hybrid intelligence), (c) standards for data sharing and interoperability, and (d) incentives or support for small/medium firms to adopt forecasting tools.
• Research gaps for AI economics:
  • Need for causal evidence using panel data, natural experiments, or randomized rollouts to estimate causal impact on firm-level outcomes (e.g., inventory turnover, stock-out rates, margins).
  • Cost–benefit analyses that quantify implementation costs and ROI over time.
  • Macro/market-level modeling of how widespread AI forecasting changes price dynamics, employment composition, and welfare in retail sectors of developing economies.

Notes for readers: the study offers useful localized evidence of perceived and statistically associated benefits from AI demand forecasting in Nairobi supermarkets, but limitations in design and measurement mean the estimate should be treated as suggestive rather than definitive causal proof. Further, objective firm-level outcome studies and broader samples are needed to generalize and to inform policy and investment decisions.