Measure first: a 20-minute performance audit
A repeatable routine for finding the one thing that is actually slow.
Performance work goes wrong when it starts with a solution. Somebody has read about a technique, applies it, and the page is exactly as slow as before because the bottleneck was elsewhere.
My routine fits in twenty minutes. First, load the page on a throttled connection with an empty cache and write down three numbers: time to first byte, largest contentful paint and total transferred bytes. Not scores, numbers.
Second, look at the waterfall and find the longest chain. A chain is anything that cannot start until something else has finished: a stylesheet that imports a font, a script that injects another script. Chains matter more than sizes.
Third, repeat the load with a warm cache. If the second visit is not dramatically faster, the caching headers are the problem and no amount of minification will help.
Only then pick a fix, and pick exactly one. Apply it, repeat the three measurements, and keep or revert. It sounds slow. In practice it is much faster than applying five changes at once and trying to work out which of them broke the layout.
The audit is deliberately low-tech. A browser, a notebook and the discipline to measure the same way twice will find most problems before any specialised tooling is needed.