The problem is rarely that files are badly named. It is that they are distributed: some in cloud storage, some on a laptop, some attached to email threads, and some in whatever tool the work happened in. Any single one of those locations is workable. The combination is what makes finding something a search rather than a lookup, and the cost is paid every time.
Choose one home for everything and move the rest into it. Cloud storage registered to the business rather than to a personal account, with the desktop folder syncing to it so there is no distinction between local and stored. That single decision resolves the majority of the problem, and it also means a lost laptop is an inconvenience rather than an event.
Structure by client or project rather than by file type, because that is how you will look for things. A folder per client, subfolders per project, and consistent subfolder names inside those. Filing by type, with all contracts in one place and all images in another, matches how files are created and not how they are retrieved.
Put dates at the front of file names in year month day order, which makes chronological sorting automatic in every system without any additional effort. Then a short description, then a version if you need one. A name like 2026-08-18-acme-proposal tells you what it is and when at a glance, and it sits in the right place in any list.
Solve the final version problem explicitly rather than through naming. Files called final, final revised, and final v3 are universal and unresolvable, because the naming cannot express which is current. A folder called current containing exactly one version of each live document, with everything superseded moved to an archive folder, removes the ambiguity structurally.
Archive rather than delete, and archive on a schedule. Completed projects move out of the active area into a year based archive, which keeps the working space navigable while preserving everything. The active area should contain only what is live, and if scrolling it takes effort, the archiving is overdue.
Name things as somebody else would search for them, including the client name in the file rather than relying on the folder to supply it. Files get moved, attached, downloaded, and separated from their context routinely, and a document called proposal is useless the moment it leaves the folder that explained it.
Decide what happens to the things that are not documents, because those are what people forget. Photographs of completed work, screenshots, recordings, and exports all accumulate somewhere unstructured and are exactly what you will want in two years. Giving them a home in the same structure at the start costs nothing.
Then do a single pass rather than intending to organise gradually. Gradual reorganisation does not happen because there is never a moment when it is the most pressing task. An afternoon spent moving everything into one structure, with anything ambiguous dropped into a folder called to sort, produces a working system immediately, and the to sort folder can be worked through whenever.
Check that whatever you choose has a working search across file contents rather than only names, because at a few thousand files the structure stops being how you find things and search takes over. Structure remains worth maintaining, since it determines whether the results make sense, and search is what makes the system usable once the volume exceeds what anybody can navigate deliberately.