Alerts

When to use it

• Know when sign-up conversion drops, without checking a chart every morning.
• Hear when a source stops sending, before a week of data is missing.
• Watch a launch closely for a few days.

Create an alert

1. Go to Alerts and click New alert. You can also press ⌘K and choose New alert, or pick Watch this… from a board tile’s menu.
2. Under Watch, pick a metric:
   • Overview KPI: Pageviews, daily or Sessions, daily.
   • Saved funnel: any funnel’s conversion, such as Checkout conversion.
   • Retention: Week 1 retention.
   • Source volume: Events from any of your sources.
3. Under Tell me about, choose a sensitivity:
   • Big changes only (the default): roughly one alert a month on a settled metric.
   • Moderate: roughly one a week. A few will turn out to be nothing.
   • Everything unusual: several a week. Useful while you are watching a launch.
4. Check What we have to work with. It shows how many days of history exist and whether the baseline is solid or still learning.
5. Under Delivery, in-app is always on for every member. Tick Email and choose who gets it. You are selected by default.
6. Click Start watching. The alert is named after its metric.

How the range is built

Traffic on a Monday doesn’t look like traffic on a Saturday, so each day is compared with the same weekday over the previous six weeks:

• The expected value is the middle (median) of those six same-weekday values.
• The range around it is based on how much those values usually vary. Big changes only gives the widest range, and Everything unusual the narrowest.
• Very small numbers can’t fire an alert. For example, a count needs enough volume and a real change behind it before a dip counts.

A metric becomes available once there is at least one earlier matching weekday, so a new project waits a few days before its first alert. Monitors are checked every few minutes.

Baseline learning

Until a metric has 42 days (six weeks) of real history, its baseline is learning. While learning, the range is held wider, and a larger change is needed before an alert fires. Monitors and alerts show a baseline still learning chip until then. After six weeks the baseline is solid and the range tightens by itself.

When an alert fires

Fired alerts appear on the Fired tab, under Needs a look or Resolved itself. Open one to see:

• Actual vs Expected range: the metric drawn against its band.
• What changed: tables of where the gap is concentrated, by segment, with expected, actual, gap and share of the drop, plus links to the events, funnel or segment. This is a filtered count against the same-weekday baseline, with no model. It shows where the drop is, not why.
• The timeline, and this monitor’s sensitivity and baseline.

Acknowledge tells your team someone is on it. Monitoring stays open. An alert closes itself, as Resolved itself, when the metric comes back inside its range. Emails go out when it fires and when it resolves.

Watching, muting and email

• The Watching tab lists every monitor with its status: Nothing unusual, Outside its range or Not checked yet.
• Mute monitor pauses checks. Resume monitor starts them again. Muted monitors live on the Muted tab.
• To change sensitivity, open an alert and use Adjust sensitivity. The next check uses the new range.
• The mail button on a monitor opens Alert emails: pick recipients and Save recipients, or Stop emailing me. Anyone who turned off operational email in their notification settings is skipped.
• Search and sort the list by Most recent or Name A–Z.

Tips and gotchas

• Start with Big changes only. A quiet alert you trust beats a noisy one you ignore.
• “All quiet” doesn’t mean nothing happened. It means nothing left its usual range for that weekday.
• Agencies see every project’s alerts in one inbox. See Agencies.

Recipes