4.3.1 · Lesson 11 of 16
Define GenAI Operational Metrics
Define system, model, and business metrics that make GenAI cost, latency, and quality observable.
WHAT
Gain operational visibility into token consumption, latency, interventions, and cost anomalies.
HOW
- Track CloudWatch metrics:
InputTokenCount,OutputTokenCount,InvocationLatency,InvocationsIntervened,ThrottlingException. - Tie token metrics to cost ownership and budgets.
Key points
- InputTokenCount / OutputTokenCount
- InvocationLatency
- InvocationsIntervened
- ThrottlingException
- Business/quality signals
Exam focus
Monitor Bedrock metrics such as InputTokenCount, OutputTokenCount, InvocationLatency, InvocationsIntervened, and ThrottlingException.