A well-defined problem is important for a successful analytics project.
Understanding the decision or action you would like to influence is an important milestone.
Discussions with subject matter experts that focus on decisions and the value of improving them can get you there. A request to improve a prediction that would improve a decision is something a data scientist could really sink their teeth into.
Important inputs to a successful implementation include knowledge of the decision to be influenced, where it fits in the process, downstream impacts, upstream support, education or system modifications needed, and especially who the key stakeholders are.