Forecast latency, throughput and headroom before every deploy.