How do you handle a PhD student whose Romance language data is too thin?

I have a PhD student working on a lesser-documented Romance variety, and the corpus we can actually access is much smaller than we assumed when the project was designed.

We are now debating whether to broaden the research question, add a different methodology, or change the scope of the thesis.

I would like to hear how others have handled a similar situation: how did you decide whether to adjust the project or push for more data collection?