Prompt
Do I need incident detection for slow requests?
Latest observation
Nov 15, 2026
OpenAI APIWeb search: off
Usually, yes—if “slow requests” affect user experience, SLAs, or revenue, they should be treated as an operational signal and often deserve incident detection.
A good rule of thumb:
-
Need incident detection if slow requests:
- are user-visible
- exceed an agreed latency SLO/threshold
- happen across multiple requests or endpoints
- indicate a dependency problem, saturation, or regression
- could turn into outages if they worsen
-
May not need incident detection if they are:
- rare one-off outliers
- expected during known batch jobs, deployments, or peak periods
- already covered by performance monitoring and not actionable
- below an impact threshold
A practical approach is to define:
- a latency SLO (e.g. p95 or p99 response time),
- an alert threshold for sustained degradation,
- and a separate metric/alert for extreme outliers or timeouts.
If you want, I can help you decide whether to alert on:
- individual slow requests,
- sustained latency degradation, or
- both.