Mixed logit models with unobserved inter- and intra-individual heterogeneity hierarchi-cally extend standard mixed logit models by allowing tastes to vary randomly both acrossindividuals as well as across choice tasks encountered by the same individual. Recentwork advocates the use of these methods in choice-based recommender systems underthe premise that mixed logit models with unobserved inter- and intra-individual hetero-geneity afford personalised preference estimation and prediction. In this research note, weevaluate the ability of mixed logit with unobserved inter- and intra-individual heterogene-ity to produce accurate individual-level predictions of choice behaviour. Using simulatedand real data, we show that mixed logit with unobserved inter- and intra-individual het-erogeneity does not provide significant improvements in choice prediction accuracy overstandard mixed logit models, which only account for inter-individual taste variation. Wemake these observations even in scenarios with high levels of intra-individual taste vari-ation and when the number of choice situations per decision-maker is large. Also, theestimation of mixed logit with unobserved inter- and intra-individual heterogeneity re-quires at least ten times as much computation time as the estimation of standard mixedlogit models. Informed by recent advances in machine learning and econometrics, wethen discuss alternative modelling approaches, which can capture richer dependenciesbetween decision-makers, alternatives and attributes.