>>Microsoft did brain teasers like "How many ping pong balls fit on a bus?" questions. Then Google copied that to start. Google later realized that performance on these questions didn't predict performance on the job and developed a more data-driven approach that you see today (leetcode style) that other companies copy today.

I'm curious how did they realize that ping pong bus questions didn't predict performance. Was that just based on negative candidate feedback or they pulled the spreadsheets to realize they are below the hiring quotas.