Watchdog & Alerts
Catch a performance drop before it quietly spends a month of budget, and get told when a test is ready to call.
Overview
Tailor runs two kinds of alerting, and they answer different questions.
Watchdog
Something went wrong. It watches ad spend and traffic patterns and flags anomalies: a sudden spend drop, a CPA spike, a source that stopped sending anyone.
Next best action
Something needs a decision. It watches your live tests and tells you when one is ready to call, stalled, or starved of traffic.
Both land in the same inbox, so the open items across alerts and recommendations are one list rather than two places to check.
Watchdog Rules
Watchdog compares each window against the equivalent window before it rather than against a flat threshold, so an ordinary Monday does not read as a collapse. Rules include:
Traffic drop
A page or source is sending materially fewer visitors than its own recent history would predict.
Source gone dark
A source that reliably sent traffic has stopped entirely. Usually a paused campaign, a broken link, or a billing failure.
UTM attribution loss
Traffic is still arriving but the campaign parameters are not, so spend can no longer be tied to outcomes. Often a redirect that drops query strings.
Low dwell
Visitors are landing and leaving faster than usual, which is how a broken or mismatched page shows up before conversions move.
Any rule can be turned off for your account if it is not useful to you. Alerts you have acted on or dismissed move to archived rather than disappearing, so the history stays readable.
Next Best Action
The recommendation engine watches the tests you already have running. Each rule can be disabled on its own, and the engine can be turned off entirely.
| Rule | Fires when |
|---|---|
| Winner ready | A variant has enough evidence to call. This is the one that turns a test into a shipped change. |
| Clear loser | A variant is losing clearly enough that leaving it running costs conversions. |
| Stale test | A test has been running long past the point of being interesting and nobody has closed it. |
| Test starved | The segment is too small to ever reach an answer, so the test needs a broader audience or a bolder change. |
| Test flat | The arms are indistinguishable, which is itself a result worth acting on. |
Starved and flat are the two most people ignore, and they are the two that waste the most time. A test nobody can read is not a test.
Tuning The Noise
An alert channel that fires on every ordinary swing gets muted, usually on the day it would have mattered. Two settings control that.
- Sensitivity sets how far a metric has to move against its own history before it is worth telling you about.
- Minimum visitors stops low-traffic pages from firing on swings that are just small numbers being small.
Start conservative. It is easier to raise sensitivity after a quiet fortnight than to win back a channel everyone has muted.
Where Alerts Arrive
Alerts appear in the app, and can be routed to email and to a Slack channel. Slack is worth wiring up: an alert that needs a human decision is far more likely to get one in the channel the team already reads.
Related: Account Lists for alerts when a target company visits, and saved agents for scheduled reports rather than event-driven alerts.