Speed starts with the cause.
One test result is not a diagnosis. Useful work needs repeatable measurements, device context and a network and code review.
Measure before change
Key pages, devices, conditions and repeated results are recorded to separate persistent problems from temporary variance.
Largest cost first
Heavy images, blocking fonts, unnecessary scripts and a slow initial response are handled by impact rather than fashion.
Regression after repair
Content, layout, features and the same measurement path are checked again. Faster must not mean broken.