Aluminum reduction cells have a limited life expectancy, and a significant proportion therefore needs to be replaced each year. This is a large expense for smelters, requiring careful long-term planning. Accurate forecasts can be hard to achieve, however, as they depend on the age distribution, design and operation conditions of the cells. In this work, we present a statistical model for forecasting the cell replacement rate. Pots are split into distinct populations, and a statistical distribution is then fitted to each one and used to produce detailed predictions. Special conditions, such as amperage creep, constraints on the start-up rate, or pot euthanasia, can also be taken into account, and the complete model is easily accessible through a web interface. Potential applications include producing detailed predictions for the coming years, exploring differences across designs or periods, rapidly detecting variations in the life expectancy, and planning replacement campaigns for future cell designs.