Considering that the majority of Australian longwall mines are currently roadway development constrained, understanding of the key geotechnical parameters that determine the roadway roof behaviour is often critical to the success of modern longwall operations. Crucial to improving this understanding is geotechnical characterisation. This process typically evolves with time and experiences at any underground coal mine and is necessary to understand the variation of the rock mass across a mining area. Based on local experience it is possible to improve forecasting of how similar geotechnical areas will behave and subsequently, the types and densities of support required to maintain roadway serviceability. There are numerous methodologies to characterise a rock mass and determine geotechnical domains using site-based data, which can vary from a simplistic single variable back analysis to more complex multivariate approaches. In a relatively isotropic and benign roof environment, simplistic models have been proven to be effective to provide an acceptable understanding of the change and variability in the rock mass and associated roof behaviour. However, with more challenging, weaker rock masses, where there are multiple independent features driving roof behaviour, the more complex statistical based back analysis approaches are more appropriate in order to accurately define different geotechnical domains. In this case study at Grosvenor Mine, a novel application of complex multivariate statistics in the form of a neural network analysis is shown to provide a useful and significant improvement in forecasting of the as-mined roadway conditions. This example indicates that in complex and challenging geotechnical environments, the application of complex analyses to characterise and understand the ground conditions is a promising potential area of further research, particularly with the advances being made in artificial intelligence more broadly.