"Image compression" is used as a general term, but there are actually two very different types of compression — and understanding the difference saves you a lot of trial and error.
This type reorganizes image data more efficiently without deleting any visual detail — the resulting image is pixel-for-pixel identical to the original, just with a smaller file size. PNG uses this kind of compression by nature, so there's a much smaller margin for shrinking it further.
This type discards fine visual details the human eye barely notices, in exchange for a much bigger reduction in file size. This is the basis of JPG, and it's why any compression tool has a "quality" slider — the lower the number, the more detail gets dropped and the smaller the file.
Rule of thumb: quality between 70% and 85% gives an excellent balance of size and visual quality for most images — you rarely need 100% quality for ordinary web photos.
If the image has text or sharp edges and fine lines (like a screenshot or a chart), lossless compression or the PNG format is usually the better choice. Regular photographic images — products, people, landscapes — handle lossy compression at medium-to-high quality with no noticeable difference to the naked eye.
Always start around 80% quality, and compare the result to the original on the same screen. If the difference isn't noticeable, try dropping to 70% or lower for a bigger saving. If you start noticing artifacts around edges or a clear loss of detail, go back up.
There's no single "correct" quality number that fits every image — but starting at 75-80% and eyeballing the result is the fastest way to find the right balance for your own images.