So far I see more value in ontologies than graph databases. When using established ontologies naming does not become an issue and exchanging / understanding data becomes easier. You also push some of your code to the database "as data" and can reason on it better.

We are actually working with semantics / ontologies. The graph database is just our engine for storing the data. I am starting to think whether an ordinary SQL database could be just as useful.

Very interesting. What did they used to use back in the day? SPARQL? This wiki article indicates that it should be possible to store triples in a RDBMS

https://en.wikipedia.org/wiki/SPARQL#Advantages