The identification strategy relies on a cluster-randomized experiment. All informal settlement customers in Nairobi Water’s utility database will first be called to confirm contact details and receive the base intervention. Eligibility criteria into the study will include: (i) customers can be contacted and their contact details can be updated; and (ii) customers do not hold multiple accounts (multiple property owners) – which is expected to yield an eligible sample size of approximately 5,500 customers. After the base intervention has been implemented and contact details have been updated, all compounds with tenants (approximately 2600 customers) will be randomly assigned to receive face-to-face tenant sensitization (Treatment 1). Randomization will be stratified by location (settlement), whether landlord resides in the property, and whether the landlord has ever made a payment. After completion of the outreach campaigns, all customers in the study will be grouped into clusters which will reflect the street they live on (approximately 166 clusters). Half of the clusters will be randomized into treatment and half into control, stratified by settlement and cluster size. Within treatment clusters, half of all compounds eligible for disconnection based on their balance and repayment history, will be randomly assigned to receive disconnection notices, resulting in approximately 420 compounds receiving notices and subsequently disconnected if they fail to make payment. To test for direct effects of receiving disconnection notices, we will compare disconnection treatment customers in treatment clusters to disconnection-eligible customers in control clusters. To test for spillovers, we will compare repayment rates of control customers in treatment clusters to disconnection-eligible customers in control clusters. To test the effectiveness of bottom-up accountability through increased tenant pressure in low and high enforcement settings, we will compare compounds with tenants that receive treatment 1 (tenant sensitization) to compounds with tenants that don’t receive the treatment, in disconnection vs. non-disconnection clusters, effectively estimating the interaction effect size of treatments 1 and 2.