Solar home system payment patterns show intermittent access and late-term defaults in pay-as-you-go contracts

Solar home systems, propelled by the pay-as-you-go model, have spearheaded rural electrification across Sub-Saharan Africa for over a decade. However, while private distributors collect vast amounts of proprietary data, this information is rarely shared with external stakeholders or analysed for real-world behavioural impact, restricting sector-wide oversight. Here we use an unsupervised deep learning model to synthesize historical payment data from over 220,000 solar home system users across Rwanda and Kenya. We find that monthly-paying users face a two to five times higher risk of early default due to behavioural preferences, while gender and age differences highlight the need for differentiated services. Crucially, our analysis exposes a persistent share of intermittent energy access and late-term contract defaults. These patterns are currently unaccounted for but, if left unaddressed, could significantly undermine confidence in a pay-as-you-go driven rural electrification agenda. Behavioral differences across user groups reveal affordability constraints in pay-as-you-go solar, according to an analysis of over 220000 payment time-series with unsupervised deep learning.