Shrinkage methods like Lasso or Ridge regression are a valuable extension of classical general linear models (GLM). By their capability of continuous variable selection they can help automating parts of the actuarial workflow. In contrast to most other machine learning techniques, the output is transparent and can be reviewed in detail before further use.
However, there are some caveats one has to consider. There are of course the old challenges of GLMs that remain the same with shrinkage methods.
In addition especially standardising practically plays a major role for the quality of the output. This is not just an issue for practitioners, but it is rooted in the core of the construction of this methods and mandatory when using them.