With that many requirements, a manual review usually ends up as a quick scan, since nobody can carefully check every item for every vendor. The result is that some requirements get real attention and others get a tick without much thought.
Scoring works better when each requirement has clear criteria for what counts as acceptable evidence, and the evidence is checked against those criteria one requirement at a time. AI can do the first pass across all of them and show which ones are met, which are partial and which are missing, so the reviewer can focus their time on the requirements that actually need judgment.