Water pricing is an economic instrument traditionally used to reduce water demand. However, its effective implementation requires knowledge of the extent to which users reduce water consumption with increasing water prices. The price elasticity of water demand has been estimated using econometric regression, which relies on cross-sectional and time-series water data. As an alternative, we propose the use of agent-based modelling, which does not require reliable historical data on water prices and consumption and enables the simulation of multiple scenarios with different consumer profiles, behaviour profiles and water price changes, thereby allowing comprehensive understanding of price elasticity estimates. To illustrate the potential use of agent-based modelling for the estimation of water demand price elasticity, we performed an empirical application to a residential area in Chile. Price elasticity estimates ranged from −0.0159 to −0.1036 (mean −0.0250), indicating that residential water consumption is inelastic to price changes. This result is consistent with previous findings. Agent-based modelling is an alternative for the ex-ante assessment of the potential effectiveness of water pricing policies intended to reduce residential water demand.

This study addresses the challenge faced by urban water utilities in translating technical water quality parameters into actionable insights on customer perception, using a data science approach in the Barcelona Metropolitan Area. Ten years of customer complaint data (2013–2023) from the water utility, Aigües de Barcelona, were integrated with corresponding water quality measurements. A comprehensive data pipeline prioritised high-coverage parameters for predictive analysis, including residual chlorine, conductivity, pH, and temperature. Exploratory Data Analysis revealed strong correlations among mineral-related parameters and identified residual chlorine and turbidity as the most distinctive variables in complaint-associated cases. Predictive modelling with Random Forest and XGBoost achieved consistent discrimination (AUC–ROC = 0.75 on a temporally held-out test set). Threshold optimisation enabled a recall exceeding 0.90, aligning with the operational objective of early detection of potential customer dissatisfaction. Interpretability analyses using SHapley Additive exPlanations confirmed that residual chlorine and conductivity, in interaction with temperature and pH, were the most influential drivers of complaint probability, exhibiting non-linear effects. These patterns were translated into actionable thresholds for operational improvement through heatmaps and logistic regression on discretised quartiles. Overall, the data-driven machine learning approach successfully links technical water quality variables to consumer dissatisfaction, providing actionable insights for utility management.