Relying solely on Average CPU Utilization might not be sufficient to guarantee the high availability of an application during peak times. Auto-scaling decisions should be based on a broader set of metrics that reflect different aspects of system performance and user experience to avoid potential bottlenecks or resource constraints. These can include metrics like network I/O, disk I/O, memory usage, request latency, error rates, and throughput. A combination of these metrics can more accurately trigger scaling actions that maintain the application's high availability during demand surges.