Journal of Marketing Analytics Маркетинг
Validating customer segmentation for longitudinal marketing analytics: evidence from four years of supermarket receipt data
Аннотация
Customer segmentations estimated from accumulated transactions are often reused for longitudinal monitoring, although cross-sectional quality does not establish their suitability for repeated comparison. This study evaluates the scope and limits of a fixed segmentation reference using 2021–2024 Danish supermarket e-receipts: 20,994,240 retained item lines, 2,484,736 receipts and 2,999 households. A six-feature K -means reference was projected onto active household-month observations. Temporal agreement was assessed through a 2021-2023-only reconstruction frozen before projection to 2024, alongside analyses of assignment geometry, temporal aggregation, clustering resolution, discount-feature representation and recording conditions. For K = 3, training/full-period adjusted Rand indices were 0.682 for households and 0.786 for 2024 monthly assignments. K = 4 achieved higher agreement but lower recurrence and separated large-basket from high-discount prevalence changes. At a matched three-month lag, recurrence increased from 65.08% for monthly observations to 74.40% for quarterly means. Removing discount-line share materially changed assignments and prevalence magnitudes. Prevalence changes remained similar among households active in all 48 months, while recording indicators also varied. Fixed definitions support descriptive comparison, but interpretation depends on temporal aggregation, segmentation specification and the measurement environment. The findings concern recorded purchasing configurations rather than identified customer migration.