The challenge of a WAF is it has no context of the application behind it. It doesn't really know what "good" input is for a specific application, so instead it tried to block "bad" things with a blacklist.

If the WAF ships with too strict of a blacklist, legit input gets block (All the people named "O'Brian" that can't use a web app...) If an attacker can make an attack look close enough to pass the blacklist, they win. In this example, the attacker had to remove whitespace, remove a math operation sign (=), and send english letters with a few special characters that could be in normal text (quote, forward slash, asterisk).

WAFs will never catch everything, and if they are advertised as such, that is snake oil. But WAFs can help by providing a ready made blacklist that can supplement input validation inside the app, which does have context about what valid input should look like.