Exam-style
A Bedrock summarisation feature has a seasonal, steadily growing spend curve: quiet in January, triple that by June, higher again next year. A monthly cost budget high enough to avoid false alarms never fires; one low enough to be useful fires every month. The team wants an alert within about a day when daily spend departs from its own moving baseline, with no threshold to re-tune and nothing to build. Which arrangement fits?
Reveal the answer
B. An AWS Cost Anomaly Detection monitor covering the feature: the AWS managed AWS services monitor, or a cost allocation tag monitor on the feature's tag, with an absolute threshold on the alert subscription
Cost Anomaly Detection fits this shape of spend: it evaluates weekly and monthly seasonality and natural growth, then alerts when the daily curve departs from the learned baseline. The AWS services monitor is AWS managed and evaluates every service independently, Bedrock included; there is no customer managed version of it. To isolate one feature, use a cost allocation tag monitor on the tag carried by its application inference profile, which is how on-demand invocation cost is attributed. Set an absolute threshold on the alert subscription, so a small departure stays under it. Detection reads Cost Explorer data, which lags up to twenty-four hours. A monthly budget compares spend against a number a human chose, which is what this team cannot keep current. A static CloudWatch alarm on Invocations fires in minutes and is a useful companion, but on a growing workload it eventually alarms on the growth. A saved Cost Explorer report is visibility with nobody watching. The Athena query over the Cost and Usage Report works, and has to be maintained.
Q. The feature’s spend is seasonal and growing, so a fixed budget threshold never fires or fires every month. What catches a departure from its own normal?
A. An AWS Cost Anomaly Detection monitor covering the feature, with an absolute threshold on the alert subscription. The monitor evaluates the seasonality and the growth, then alerts on the daily departure from that baseline, so there is no number to re-tune each quarter. The AWS services monitor is AWS managed and covers Bedrock automatically. A cost allocation tag monitor narrows it to one feature, which is what the tagging work makes possible.
Why? Three horizons, three tools. A CloudWatch alarm on Invocations or OutputTokenCount fires in minutes, catching the retry loop while it is still running. Cost Anomaly Detection reads billing data through Cost Explorer, so it lands up to twenty-four hours later. It answers a different question: has daily spend on this feature left its learned curve. A monthly budget compares spend against a number a human chose. That number is what keeps breaking on a curve that triples between January and June, even with the guardrail layering already in front of the workload. Run the alarm and the monitor together.