Discrete Choice Models (DCMs) have a distinct advantage over Machine Learning (ML) classification algorithms, in that they employ a highly interpretable linear structure. However, a key drawback of DCMs compared to ML is the need to specify the utility functions manually in advance of fitting the model. This process does not scale well with increasing data complexity. As such, ML is increasingly being applied to problems typically tackled using DCMs, in particular where large, complex datasets are available. Whilst ML automates the model specification process, ML algorithms suffer from a crucial limitation in that the resulting models are difficult to interpret and to check for behavioural consistency. In this talk, I present an assisted specification framework which addresses the respective limitations of the current DCM and ML solutions. The framework uses the aggregate structure of a fitted Ensemble Learning (EL) model to inform the utility functions in a DCM. The resulting assisted-specification model is found to have greatly improved performance for a mode-choice prediction problem, outperforming a suite of benchmark DCM and ML models, whilst maintaining an interpretable linear structure.