Skip to content
Optima docs
ConsoleAgentsBuilt-in agents

Anomaly explainer

A built-in agent that emails you why a metric alert fired, with the breakdowns that explain most of the change.

The anomaly explainer is a built-in agent that investigates a metric alert as soon as it fires. It looks at what changed, splits the metric by the dimensions your data allows, and emails you a short explanation of the largest contributors and the likely cause.

Who uses it

Campaign managers and tenant administrators create the instance. The people who receive the email can be anyone on your team or any email address you enter. You need the Built-in agents feature in your licence and the Metrics feature, and you need at least one metric with an alert configured, because the agent starts from a fired alert.

Fields

ParameterWhat you enterDefault
recipientsThe people who get the email.Owner (you)
metricsThe metrics to explain. Leave empty to explain the alerts of every metric.Any
lookback_daysHow many days of the metric's history the agent considers. The agent reads at most the last 200 checks of the metric, so a metric checked every 15 minutes shows about two days of history whatever you enter.28

Actions

  • Reads: the metric alert that fired, the metric and its recent history, and breakdown queries over your data. It reads aggregates, not individual customers.
  • Sends: one email per alert to the recipients you choose.

The email is plain text of at most 25 lines:

  • One sentence on what changed, from what to what, and when.
  • Largest contributors: up to three breakdown values with their numbers in both windows.
  • Likely cause: one or two sentences with a confidence of high, medium or low.
  • Not checked: what the agent could not verify, such as a failed tool or a dimension that does not exist.

When an alert fires because a metric stopped receiving data, the email explains which inputs stopped arriving instead of breaking down a value.

Every number in the email is checked against what the agent read. If the agent cannot back a figure up, the run fails and nothing is sent.

Examples

Create the instance in dry run, then open a metric and temporarily set Change of at least (%) to a small number such as 1 and select Trigger. The run appears in the run log with the email it would send. Set the threshold back afterwards, and switch the instance to live when you are happy with the emails.

Common problems

  • No email. The metric's alert did not fire, the alert is within its quiet period, or the instance is stopped. Check the metric page and the instance.
  • The run failed with "numbers that no tool returned". Try a stronger model or look at the run in the run log.
  • The explanation says it could not check something. The agent's owner may lack access to the data it needed. Create the instance as someone who may read it.
  • The card is missing. Your licence has no Metrics feature, or no Built-in agents feature.

On this page