Pretty much all the AWS services can post messages to an SNS topic, and CloudWatch is no exception. With CloudWatch, we can collect metrics from our operating system, as well as from our applications. On top of the metrics, we also have the ability to create alarms. When creating an alarm, we would need to define a threshold for a certain metric and the duration of the threshold breach. 

For example, when monitoring our EC2 instances in our environment, we can set a threshold for the CPU usage of 80% and a duration of 10 minutes. When an EC2 instance's CPU usage goes above 80% for more than 10 minutes, a CloudWatch alarm will trigger. That trigger can send a message to an SNS topic, which can in turn notify a certain email address or perform other actions, such as triggering a Lambda function and so on. It is also fairly easy to design the alarm notifications, as we can customize which topic each alarm will post messages to, so we can divide the tasks among teams responsible for their own region or account, or aggregate all our accounts and regions into one email: