What does Lossy compression mean for your PDF?
JPEG is lossy: it stores an approximation, and each re-encode approximates the approximation. PNG and Flate are lossless: they pack the same pixels into fewer bytes, and decompress to exactly what went in.
In a PDF, photographs and scans are usually JPEG, while text, line art and transparency are lossless. That is why compressing a PDF works on the pictures and leaves the text alone — the text costs nothing and cannot be improved.
Quality settings around 75–80 are the practical floor for documents: below that, edges around text and fine lines start to show halos.
Why does compressing some PDFs do almost nothing?
Because the size is not where you think it is. Compression works on images, and a PDF exported from Word or a browser is mostly text and vector shapes — which are already stored efficiently and cannot be approximated. Squeezing a 400 KB text document gives you a 390 KB text document.
The files that shrink dramatically are the ones dominated by pictures: scans, photographed pages, brochures, anything with full-page background images. Those routinely drop by 70 to 90 per cent, because that is where the waste lives.
Before compressing, it is worth knowing which kind you have. If your PDF is a scan, the images are the whole file. If you can select the text, most of the file is already as small as it goes, and the honest answer may be that it cannot get much smaller.