We rebuild the page nearest the purchase and run it head to head against the one you use now, on your own visitors. Judged on revenue per visitor, not opinions.
Usually the product page, because that is where the money is decided. One page, rebuilt properly, rather than a list of opinions about your site.
The new page runs against your current one on your own traffic, both live at the same time. Up to five iterations until the result is clear.
Two or three tests a month on the pages closest to the purchase, each measured the same way, so the gains compound instead of resetting.
Every figure we report was measured in a live split test on the brand's own traffic.