AI-Driven Distribution Planning for Essential Goods in Underserved Communities: A Mixed Methods Framework for Access Optimization

Abstract

This study examined how underserved communities experience moderate to low and uneven access to essential goods due to last-mile distribution planning that is constrained by unreliable data, partial workflow embedding, and volatile operating conditions. The purpose was to quantify how AI-driven distribution planning capability improves access optimization within cloud and enterprise distribution contexts, while accounting for technology, organizational, and environmental conditions. Using a quantitative, cross-sectional, case-based design, a structured five-point Likert survey captured system-level perceptions from 312 valid respondents (78.0% usable) out of 400 invited and 320 completed, spanning planners/coordinators (22.4%), field and delivery staff (27.6%), retail or community distribution points (18.9%), NGO or partner actors (11.5%), and beneficiaries (19.6%). Access Optimization (AO) was modeled as the outcome, with AI Planning Capability (AIPC), Data Quality and Integration (DQI), Organizational Readiness and Embeddedness (ORG_READY), Environmental Support (ENV_SUPPORT), and Constraint Severity (CONSTRAINTS) as predictors. The analysis included reliability testing, descriptive statistics, Pearson correlations, and multiple regression with collinearity diagnostics. Measurement reliability was strong across constructs (AO α = .91; AIPC α = .88; DQI α = .86; ORG_READY α = .89; ENV_SUPPORT α = .82; CONSTRAINTS α = .79). Baseline access was constrained (AO M = 2.84, SD = 0.76) alongside high operational constraints (M = 3.74, SD = 0.62), moderate AI capability (AIPC M = 3.27, SD = 0.71), and moderate organizational readiness (M = 3.12, SD = 0.69). Equity profiling revealed meaningful disparity, with the lowest access cluster reporting AO M = 2.41 versus 3.18 in the highest access cluster (gap = 0.77; inequity index = 0.21). Bivariate results aligned with the conceptual model, showing positive associations between AO and AIPC (r = .52), DQI (r = .47), ORG_READY (r = .55), and ENV_SUPPORT (r = .32), and a negative association with CONSTRAINTS (r = −.41). The regression model was significant and explanatory (R² = .45; Adj. R² = .44; F (5,306) = 49.8, p < .001; VIF 1.22–2.05), with ORG_READY as the strongest predictor (β = .29), followed by AIPC (β = .24) and DQI (β = .18). Results indicate that AI-enabled planning can improve access, but realized gains depend on data integration, organizational embedding, and targeted interventions to narrow subarea gaps under persistent constraints.