EDIT I\'ve just started skimming Codd\'s famous 1970 paper that started it all, that Oracle was based on (A Relational Model of Data for Large Shared Data Banks [pdf]),
One key was the self contained products - you no longer had to manually define and maintain your key files (indexes) and the ability to change the data model with less effort. Combine that with the SET based structures made it a compelling product set to work with. Combine the SQL language on top of that to return data and it was a win-win situation over traditional ISAM data structures primarily associated with COBOL languages.