How does a firm decide whether a repeated Claude task is worth turning into a formal skill?

One firm (three people) described a two-part filter: (1) tasks they personally hate doing get prioritized for conversion into a skill, and (2) tasks that are done more than once and follow a similar process each time are candidates. They weigh the time investment of building the skill against the expected payoff. As an example, they built a tax return review skill that takes a full copy of the tax return, all source documents, the organizer, and client messages, cross-references everything, checks the math, and flags discrepancies -- essentially automating the 'tick and tie' review process between the return and the workpapers/Excel tracking.