SQL, data modeling, data quality, EDA, visualization · insight-1073
Page on actionable conditions only, and keep everything else on the dashboard
- Evidence type
- documented fact
- Difficulty
- easy
- Roles
- Data analyst, Data scientist, ML engineer
- Industries
- cross-industry
Summary: situation, action, reason, boundary
Situation
A data-platform team routes every failed check, slow query and late table to a chat alert, and the channel is muted within a month.
Action
Alert only when a human must act now and a runbook step exists, for example a freshness SLA breach on a key table; send the rest to a dashboard reviewed on a schedule. Review which alerts led to action each month and retire the others.
Reason
Attention is limited, so alerts must be rare, actionable and tied to impact for the signal to survive.
Boundary
Data pipelines, model monitoring and analytics platforms.
Symptom or risk
Noisy alerts train people to ignore them, so the one failure that corrupts a revenue table is missed, and responders cannot tell urgent from informational.
Exceptions
Early warm-up of a new pipeline, when broad alerting teaches you what normal looks like.
Trade-offs
Fewer alerts mean some problems are found later, at the next review.
Pitfalls
Alerting on causes (a job restart) instead of symptoms that users feel.
Lessons and concepts
Claims and their sources
Google's SRE monitoring chapter says pages should be urgent, actionable and require human judgment, that noisy alerts get ignored, and that non-urgent issues belong on dashboards.