Turning on monitoring feels like progress. Week one often feels like regret. Alerts fire for deploys, for probes, for things you already knew. Or nothing fires and you wonder if it works. Both are normal. The job of the first two weeks is to earn trust.
Days 1–3: install the minimum
Use the day-one SaaS list — login, core app or API, money path, status page. One-minute checks on the critical ones. Put alerts somewhere you will actually see, with an owner.
Do not connect five chat tools yet. One channel. One human who cares.
Days 4–10: keep a calibration log
A spreadsheet or a note with four columns:
- Time
- Alert
- Real customer impact? (yes / no / regional)
- Change made
Every false page should produce a change: confirmation stricter, maintenance window for deploys, timeout adjusted, monitor removed. Every missed real issue should produce a new check or a better health URL.
If the log stays empty because you muted the channel, you are not calibrating — you are hiding.
Days 11–14: freeze scope, test the human path
Stop adding monitors. Run a planned failure if you can (staging is fine): trigger a page, practice ack, open the runbook, post a status note. Fix whatever was embarrassing.
Then ask: do we believe the next page? If no, keep tuning. If yes, carefully add the next customer-critical URL.
What "done" looks like for week two
- On-call knows who is primary.
- Pages are rare enough that people still read them.
- Status page (if you have one) matches what the monitors say.
- You can explain the last three alerts without guessing.
Monitoring is not a checkbox you tick in setup. It is a habit you install. The first two weeks are when the habit either forms or dies.
