An experienced architect will always do this. Often the most risky subproblem is the one that is least understood (unknown unknowns), which needs to be investigated, even if it turns out not to be a great risk. In agile-speak this is called 'spike solution'. Also see lean architecture methods, such as:

- 'Just Enough Software Architecture', by George Fairbanks.

- 'Risk and Cost Driven Architecture', by Eltjo Poort.