Business model/plan(ing) for research heavy early stage ML startup

Most information business planning aspects of tech startups applies to commercializing ideas that are closer to an MVP, are based on just engineering on top of some conventional science/technology or are reasonably close to execution.

For a startup that is embarking on years of basic scientific research before we even have a prototype, how can they maintain focus in strategy and execution , what should that look like in a business document for grants etc? Thanks!