The question here is how many of the potential customers you've already reached? How do you define the target group? How do you limit it, especially in very early stages?
At first, startup has 0 customers. Apple has a bit more. Both define customers differently.
For new products, we try to validate the idea with at least 10 people that are wiling to pay. We start with a narrow niche (as narrow as we can make it).
When it answers a need that a sizeable percentage of your users have.
When it makes things easier for a sizeable percentage of your users.
When it can be done at a reasonable cost.
When someone with significant clout specifically hires you to. - In enterprise software it sometimes makes sense to build things to order for a specific client, even if just as one-off project.
When it saves you money behind the scenes.
And, to a certain extent, when you can rapidly prototype it and don't know what will happen - you can sometimes spin it off into a separate product if you're large enough.
Note that need and easier are not things that the users necessarily know themselves.
Sometimes your roadmap is for sale, it happens (http://www.bothsidesofthetable.com/2013/04/15/how-to-sell-yo...). But you know when to say yes by having:
1) A vision of what your product is going to become
2) Understanding of your customers and your market
