What metrics would you look at to evaluate the performance and latency of a critical API integration?