How deep should an issue tree go, and how do you know when to stop?

How to answer

Drill until each leaf is a question you can actually test with available data or a single analysis - usually 2-3 levels. Stop when the next split no longer changes what you'd recommend or do. Going deeper wastes the partner's time; staying too shallow leaves the recommendation unactionable. Depth should track decision-relevance, not symmetry or completeness.

Key idea: Building a symmetric, fully-expanded tree for its own sake instead of pruning to the decision-relevant branches.