I think they can be an over-simplistic solution when more effective solutions exist and can work with a little more effort, particularly for a company as large as Google where you have the volume to have meaningful data about hiring patterns.

- Outreach to under-represented groups is a no-brainer (I assume Google is already doing this)

- Look at ways to make your interview / recruiting process as blind as possible - coding exercises, resume review and even behavioural questions can be done in a way that doesn't reveal someone's gender or race.

- Where blindness isn't possible, use data to figure out where there may be bias. Do certain individuals / teams / departments show bias in who they advance through the hiring process? If at least part of your process is blind this is even easier - look for evidence of candidates who did well until the process could no longer be gender / race blind and see if there's any bias introduced at that point.

OKRs mandating a fixed gender ratio are the worst way to go about things. It's supposedly common knowledge that any metric is going to be gamed, and doubly so if your career performance is based heavily on it, so it shouldn't be surprising that basing it heavily on a 50/50 ratio can often result in "hiring to quota". Long term, this ends up casting doubt on eminently qualified women and racialized folks.