Automate tests that are repeatable, stable, high-risk and expensive to run by hand: regression, smoke, API, data-driven and cross-browser checks. Keep exploratory, usability, one-off and fast-changing scenarios manual. A practical scoring rule is execution frequency × business criticality × stability, and a useful stability check is to wait if a feature has changed shape twice in the last three sprints. Martin Fowler's guide also keeps exploratory testing in the mix because automation misses design and usability problems.

Sources: QAwerk, PractiTest, Fowler