One of the most challenging tasks in potable water production is the cost-efficient and consistent operation of water treatment plants (WTPs) that treat raw water of variable quality and quantity. To increase process stability and optimize the usage of resources, two data-driven models simulated coagulation in two WTPs. The datadriven models were successfully trained on monitoring data collected from the two WTPs (mean errors of effluent turbidity were below 0.5 NTU in both case studies) and were subsequently employed in the optimization of two historical periods of the WTPs. During this model-based backtesting of the WTPs, multiple operating scenarios were investigated on a daily time step in search of chemical doses that deliver a quality threshold for treated water at the minimum usage of chemicals. Results from the application of this model-based approach for WTP optimization indicated that a reduction of chemical costs equal to 6% and 8% would be probable for the two case studies respectively, without hampering the efficiency of raw water treatment. This work underscores that the large quantity of passive data that are amassed daily during the operation of WTPs can be turned into actionable intelligence that supports decision-making and enhances adaptive planning for water utility operators.