The present disclosure provides a computer-implemented method of predictively preparing a water provision system installed in a building, the water provision system comprising a heat pump configured to transfer thermal energy from outside the building to a thermal energy storage medium inside the building and a control module configured to control operation of the heat pump, the control module having executing thereon a first machine learning algorithm, MLA, having previously been trained to determine a correlation between cold water usage and a subsequent heated water demand, the water provision system being configured to provide water heated by the thermal energy storage medium to an occupant of the building at one or more water outlets, the method being performed by the control module and comprising: receiving first sensor data indicating cold water usage at a first water outlet; determining whether the cold water usage at the first water outlet is correlated to a subsequent heated water demand at a second water outlet by inputting the first sensor data to the first MLA; and upon determining that the cold water usage at the first water outlet is correlated to a subsequent heated water demand at a second water outlet, preparing the water provision system for delivering heated water.