Sort every thin page onto one of two branches before you touch it: fixable-and-worth-ranking, or useful-to-visitors-but-never-meant-for-search. The branch decides the action, and almost no page belongs in a “maybe” pile between them.
Branch one is improve. Ask whether the page has a real job in search, a query it should answer, and whether you can make it genuinely satisfy that query. If both are yes, the thinness is a content gap, not a reason to hide the page. You expand it, answer the question fully, and let it earn its place. Noindexing a page like this throws away a ranking opportunity you could have won with an afternoon of work.
Branch two is noindex. Some pages are thin by design and always will be: thank-you and confirmation pages, internal search results, thin tag, filter, and category pages, utility URLs that exist for users mid-flow. They serve a purpose on the site, but they have no business competing in search and will never be worth indexing. These you noindex, which keeps them working for visitors while removing them from Google’s view of your content. The test that splits the branches is simply whether the page is meant to rank and could, with work, deserve to.
One exception overrides the sort. If a thin page has earned backlinks, do not noindex it on reflex. Those links are real equity pointing at a live, indexed URL, and cutting the page out of the index wastes them. Keep it indexed and improve it instead, so the page lives up to the links it already attracted.
The mistake on each side is a reflex. Noindexing every thin page sweeps away ones you should have fixed, and improving pages that were never meant for search pours effort into URLs that should quietly drop out. Decide the branch, then act on it.