Alcohol-specific mortality has risen in England, yet behavioural drivers remain hard to measure due to stigma and survey bias. We investigate spatial inequalities in alcohol mortality across English Lower-tier Local Authorities by integrating novel supermarket transaction data with established socioeconomic predictors. Using machine learning with spatially stratified cross-validation, we test whether shopping behaviours improve prediction beyond traditional demographics. Starting from a combined 28-feature set (13 demographic, 15 shopping), a refined 15-feature model achieved \(R^2=0.374\) , outperforming both a demographic-only model and the full combined set. Key shopping features–spirits purchasing intensity (quadratic), store-level alcohol concentration, and spatial consumption spillovers–ranked among the strongest predictors. A Bayesian bootstrap with a \(\pm 0.01\) \(R^2\) ROPE indicated practical superiority over the full model. Retail behavioural signals capture consumption patterns that demographics miss, offering policymakers timelier tools to identify at-risk communities and target interventions.