Knowing a domain and being able to predict success in it are two very different things.
EDIT: A better example. Before Ford released the F-150, it seemed to be a solution in search of a problem. It was somewhere between the F-100 and F-250 and none of the pre-market reviewers really seemed to like it. Ford released it anyway. The F-150 went on to become the most popular variant of the best selling truck series for 34 years.
That's an interesting example, but it's not from a startup. Ignoring that, it could still hardly be considered an idea Ford wouldn't know if it would work at all, because they already had success with similar ideas.
It's not from a startup but there's an interesting parallel. It's as if Ford had a $10/month plan and a $25/month plan. They didn't see a need for a $15/month plan but they created one anyways and it took off.
The thinking behind its success is likely similar: the highest tier of service often acts as a decoy for the one you really want them to pick... the second highest plan. It gives customers a reason to reject the "luxury" option and settle instead for the middle of the pack option which is only slightly more expensive than the cheap option, allowing them to feel like they made the sensible choice.
EDIT: A better example. Before Ford released the F-150, it seemed to be a solution in search of a problem. It was somewhere between the F-100 and F-250 and none of the pre-market reviewers really seemed to like it. Ford released it anyway. The F-150 went on to become the most popular variant of the best selling truck series for 34 years.