Get notified

Metric alerts

Rules that watch a call metric across many calls, plus the record of every fire.

A metric alert watches a number across many calls rather than reacting to one. They live on the Alerts page in the workspace sidebar, alongside a record of everything that has fired.

Rules are evaluated every fifteen minutes. One rule fires at most once an hour so a bad run does not flood your channels.

Build a rule

Choose New alert on the workspace Alerts page, then turn on a ready-made alert or build your own. New alert needs a plan that includes alerts. To build your own:

  1. Metric. What to watch, from the shared metric catalogue.
  2. Name. Used as the title in the alerts table and in Slack.
  3. Scope. All agents, or one.
  4. Alert type. Trend fires when the metric drifts away from its usual level. Threshold fires when enough recent calls cross a value you set.
  5. For a trend rule, pick a sensitivity: Conservative catches only large changes, Balanced catches clear changes, Aggressive catches small changes too. Then pick a direction: both, up only, or down only.
  6. For a threshold rule, pick a condition (above a value, below a value, or outside a range), the value or values, how many recent calls must breach it (Min breaches), and how many calls to look back over (Window (calls)).
  7. Delivery, and optional From name, Email subject and Email heading overrides.

Setting a threshold value

Alert type:       Threshold
Condition:        Value < threshold
Threshold value:  0.8
Min breaches:     5
Window (calls):   20

Watch the unit in the value's label: some metrics are stored as a fraction (0 to 1) even though the product shows them as a percentage.

When a trend rule stays quiet

A trend rule stays quiet until it has enough calls to establish a baseline, and while the metric has no spread to measure. Conservative needs 100 measured calls within 30 days, Balanced 50 and Aggressive 25.

Where a fire is delivered

  • Alerts page only records the fire without sending anything.
  • Email goes to the addresses on the rule, from notifications@verticalai.com.au, with the reason, metric, observed value, baseline, and a link to the triggering call.
  • Slack posts to an incoming webhook you connect under Delivery settings, with the same detail and a Go to call button.

Built-in notifications

Six ready-made rules sit alongside your own. Their thresholds are fixed. Spend approaching limit and Agent performance are on until you switch them off. The other four start off.

NotificationWatches for
No calls in 24 hoursAn agent that normally gets calls goes 24 hours without one.
Frequent callerOne number calls the same agent five or more times in an hour.
Transfer spikeAn agent hands off three or more calls to a human within an hour.
Spend approaching limitUsage reaches 80%, then 100%, of the spend limit.
Agent goes liveAn agent is published for the first time or assigned a phone number.
Agent performanceAn agent reaches a call milestone, has three tool errors in a day, or goes a week without a call.

Each keeps its own recipient list. Leave the list blank and it goes to the Default recipients under Delivery settings, or to the workspace owner when that is blank too. Removing a ready-made row switches it off, returns it to New alert and resets its recipients and wording. Switching it off keeps both.

Dashboard reports

Two more ready-made rows, marked Report, email the dashboards you have flagged for the daily report. Daily report goes out once a day and Weekly summary is a seven-day digest sent on Mondays.

Their schedule is fixed. Unlike call summaries, they are not per-agent and there is only one of each.

Reviewing what fired

The History tab on the workspace Alerts page covers the last 24 hours, 3 days, 7 days (the default) or 30 days, or a custom range. It shows total triggers, how many alerts triggered, the most triggered alert and an activity chart, plus one row per fire with its reason and a link to the triggering call.

Inside one fire

Opening a fire shows the full reason, when it fired, the observed value, the usual level, the measurement window, and delivery status (Sent, Failed to send, Not sent or Sending, or Alerts page only for a rule that sends nothing).

Below that is the rule's Trigger history for the last seven days. Once a rule has fired at least three times recently, Summarise groups those fires into a headline and a few themes.

To stop a noisy rule, retune it or switch it off.

On this page