On this page
Monitors need Pro or above: 20 monitors and 2 notification channels on Pro, unlimited on VIP 1 and above.
Where: Product analytics › Intelligence › Monitors.
What a monitor does
A monitor checks one metric every hour, day, week or month and notifies you when its condition is met. It doesn't change collection or fix anything — it makes sure you hear about it first.
Metrics you can watch
Built in: pageviews, visitors, sessions, bounce rate, average duration, error rate and error count. You can also pick a custom metric you've saved, such as sign-ups or paid conversion rate.
Six conditions
| Condition | Fires when | Good for |
|---|---|---|
| Fixed threshold | The value is above or below a number | Errors above 50, sign-ups below 10 |
| Period change | The change from the previous period passes a percentage | Visitors down 30% on the same day last week |
| Forecast baseline | The value leaves the average of the last 7 days and the same weekday | Traffic with a clear weekly rhythm. Needs at least 14 days of history |
| Anomaly detection | The value leaves the normal range of the same weekday over 28 days | Noisy metrics where a threshold is hard to pick. Needs 30 days by default |
| Frustrated clicks | Rage clicks per thousand pageviews reach a number | Interaction problems after a redesign |
| Collection stopped | No data at all for a number of minutes | The snippet deleted by accident, a broken domain setting |
Create a "collection stopped" monitor in every project. A snippet that disappears during a site redesign, with nobody noticing for two weeks, is a more common accident than you'd think.
Dry-run first
Before saving, click Dry run. It calculates without sending and tells you how often the rule would have fired in the last 30 days. A rule that fires twenty times a month is soon ignored; one that never fires may be too loose.
Avoiding alert fatigue
- The same monitor doesn't notify again for about 60 minutes by default.
- After 3 consecutive hits it pauses itself so it doesn't keep paging you. Resume it once the problem is handled.
- Forecast baselines have three sensitivities: strict (about ±10%), medium (about ±30%) and loose (about ±50%). Start with medium.
Where notifications go
5 channels are supported: email, WeCom, DingTalk, Feishu and webhooks. A monitor without its own channel list notifies every enabled channel. Webhooks can feed your own on-call system.
Scheduled reports
Besides alerts when something goes wrong, Pro and above can subscribe to weekly and monthly email reports — the previous period's key numbers and changes delivered to your inbox, no login needed.
Permissions
Everyone can view monitors. Creating, pausing, resuming and dry runs need Analyst or above; deleting needs an admin or the owner.
Not enough history yet
Forecast baselines and anomaly detection skip their check until there is enough history — they aren't broken. They start working on their own once the days have accumulated.