My experience is that you can "sell" R when the statistical or modelling technique is not (or not well) implemented in python. Which still includes a lot of potentially useful statistical techniques! R should/could lean into being the tool that python programmers reach for when they need something a little less mainstream, if they made it easier for non-R programmers to do so.