Hmm, there might be couple of reasons, If after checking that your metrics are set properly, It could be the cooldown period after a scale-out event before scale-in can occur. Check your scaling policy's cooldown settings. And also, it could be that the fluctuation between scaling states is rapid and this is will lead to Thrasing. You can try reducing the datapoints maybe 3 instead of 15 and then test with controlled load, gradually reduce the load to below the scale-in threshold (e.g., 50%) and observe CloudWatch metrics/alarms. Use the AWS CLI to check alarm states: aws cloudwatch describe-alarms --alarm-names <scale-in-alarm-name>