This paper shows how Grounded Theory (GT), a method for domain understanding predominantly used in the social sciences, can be useful for the design of a Domain Specific Modelling Language (DSML). Using a pilot study from the smart grid domain, we discuss how GT can be used to systematically derive the abstract syntax of a DSML from domain data. From this, we derive lessons learned from the application of GT, the most relevant being that (1) with GT, one systematically derives an abstract syntax of a DSML, reflecting domain commonalities and variation points, (2) in line with its explorative character, with GT one gains a grounded domain understanding, and the domain goals that a DSML should satisfy, (3) GT does imply a notable time investment, which one needs to weigh against its prospective benefits. Finally we present a concluding outlook in terms of implications for DSML design mechanisms.