Abstract Stochastic integer programming (SIP) has recently been studied to manage the risk caused by geological uncertainty when solving mine planning and production scheduling problems of open pit mines. However, similar to other mathematical programming techniques that deploy integer variables, the main obstacle of applying SIP on real-life datasets stems from the enormous number of integer variables required by its mathematical formulation, which is a function of number of mining blocks being processed and lifespan of the mining project. In this paper, a new framework is proposed for stochastic mine planning process which makes the application of SIP on large-scale datasets tractable. Firstly, mining blocks of simulated orebody models are clustered using TopCone algorithm to significantly reduce the scale of the data. A new SIP model is then developed to work on aggregated blocks so not only the net present value (NPV) is maximised and the risk of not meeting production targets is minimised, but also solution can be obtained in a practical timeframe. The scheduling result of the new SIP model is also compared to an integer programming (IP) model to highlight the ability to manage risk and generating higher NPV on a case study of a large-scale multi-element iron ore deposit in Pilbara region, Western Australia.