What is a common mistake with Kubernetes autoscaling?

Scaling only on CPU when workload demand depends on queues, latency, or external metrics may produce slow or ineffective scaling.