We propose a comment about the article “What makes a VRP solution good? The generation of problem-specific knowledge for heuristics” (Arnold and Sörensen, 2019) by Florian Arnold and Kenneth Sörensen. In the original contribution, the authors implemented several Machine Learning (ML) algorithms in order to predict good vs. not good solutions. Then, some outcomes of the algorithms were discussed in terms of the predictive power of the solutions features. The purpose was then to use the extracted knowledge to improve existing heuristics. The first contribution of our comment is to validate and complement some of the conclusions of the authors. Then, we argue than most of the extracted knowledge can be retrieved by classical data reduction methods such as Principal Component Analysis (PCA). Hence, instead of ML-based predictions, a factorial analysis provides a powerful and synthetic view of the variables inter-dependencies in the light of solution quality. Thanks to the datasets provided by the authors in the original article, new experimental results are conducted. Finally, the integration of these results into future “boosted” heuristics is discussed.