I was working on the Google Ads backend, and we had a strict 20ms timeout. The managers didn't care about the algorithms themselves, but my change had a 0.1ms i

I was working on the Google Ads backend, and we had a strict 20ms timeout. The managers didn't care about the algorithms themselves, but my change had a 0.1ms impact on latency in the experiment arm, for launching that you need director level exception approval. My tech lead actually helped me by saying that the trick is that instead of trying to speed up my own code base (which was really hard), it's easier to find something to improve outside my change to balance for the 0.1ms impact inside the same experiment.