Google

Total compensation was excellent (I was able to FI/RE after less than a decade there).

Also, your work has real-world impact on a large scale.

The company advertises itself as data-driven, but politics determines whether a project moves forward and whether or not you will have financial success (promotions, bonuses, stock, etc).

Example #1: Two projects I worked on had neutral metrics but a significant cost increase in terms of infrastructure and computational cost. They both were launched anyway because the higher-ups believed that in the future the metrics would become better (even though the existing data did not support that conclusion). In both cases, the payoff never happened. (One of the features was later removed; the other remained, still adding costs without providing benefits, when I left years later.)

Example #2: Another project I worked on reduced AI model size by ~30% while delivering the same accuracy. Higher-ups did not believe this was possible and asked for more data. After months of back and forth, gathering all the data they required and proving the results were solid, they switched tactics and stated a new technology was coming that would replace this project shortly, so they shelved it. Two years later (when I finally left), my results were still on the shelf with this new technology never being deployed.

Example #3: We developed two AI models, B and C, both better than the current model A. C was much better than B, so we should launch C, right? No. That's not the Google way. Instead, launch B, then launch C just a bit later. That way, it looks like you're doing a lot of work (constantly launching new and improved models) and you're more likely to get promoted (because two good launches is better than one), even though you're actually costing the company money while B is in production (because C was substantially better). Basically: looking busy is better than accomplishing something.

I have worked in multiple research-heavy companies, both industry and academic (and somewhere in-between), and this was the most politically skewed of all of them. The cherry-picking of data was ridiculous.

My wife also worked there, and she had a similar experience (though she was much more adept at navigating the politics). For instance, she proved that a $XXX million project had zero positive impact on the bottom line because it had just cannibalized another Google project (i.e., moving the money from your left pocket to your right pocket). The cover-up was ridiculous (because the right-pocketers still wanted their promotions despite delivering no benefits to the company), though in this case, I have to give credit to the top higher-ups that they actually saw through it.

Ultimately, my biggest financial benefits and promotional awards came from work on features that did not improve the company's situation, while my most worthwhile work was shelved.

So in the end, I had to leave because I didn't like that I was losing my integrity. Basically, I was learning to nod along (or even do nothing!) to cash a bigger paycheck, instead of producing work of actual value.

Enforce go/no-jerks.