As someone who's currently trying to carry their company into data science (I know, we're late to the party, but homebuilding usually is), this is something tha

As someone who's currently trying to carry their company into data science (I know, we're late to the party, but homebuilding usually is), this is something that keeps me up at night. I've been hammering the idea into my superiors that if we're going to effectively administer a scientific process, we have to not only change how we manage expectations, but really, throw them out entirely. We have to see the general pursuit of knowledge as it's own ROI, because odds are that tangible value won't come from a data science team for a long time. But having seen how poorly we adopted agile, I'm skeptical this will happen (we devolved to "waterfall w/sprints" due to not properly managing expectations with our users). If we can't get people to move away from hard deadlines for regular releases, how are we going to make them wait for a properly verified hypothesis? One solution I've seen is organizationally "hiding" science teams from users, and only allowing a select few drive the direction of the team. But it still comes down to that select few properly managing everyone else's expectations.