Anomalies in Quantify: setup and interpretation
Use anomaly flags to investigate unusually high underlying values in a supported Quantify time series. Start with a Count analysis for the clearest interpretation. A flag is a prompt to inspect the data; it does not establish the cause of a change or guarantee statistical significance.
For general chart guidance, see Read and interpret Quantify results. For the meaning of Count, Percentage, and numeric measures, see Calculations on Quantify.
Show anomalies
Open or build a Quantify analysis with a grouping and time-based results. Review Plot, Filter, Duration, and Measured as.
Choose Trend + Anomalies from the chart types offered for the query.
Turn on Show Anomalies when available.
Inspect the flagged points and the supporting result table or records.
Chart and anomaly controls depend on the query. Some query shapes, including user/account measures and a single aggregate trend, do not expose the same controls. Check the available chart types and control messages if you cannot enable anomalies.
Set the threshold
Choose the gear beside Show Anomalies, labeled Configure anomaly threshold.
In Set Threshold, choose Default or Custom.
For Custom, enter a value from 1 to 3. The stepper changes the value by 0.1. The default is 1.5.
Choose Apply to commit the change and refresh the result. Closing the panel without applying does not commit an edited threshold.
A lower threshold can flag smaller upward deviations; a higher threshold requires a larger upward deviation. The threshold is a Z-score cutoff, not a confidence percentage or a guarantee that every meaningful change will be detected.
What a flag means
Flags focus on the latest completed day, week, or month, depending on the time buckets used by the query. They do not independently test every historical point shown in the chart. The detector uses recent completed time buckets; its comparison window can differ from the Duration displayed in the chart.
For a Count of feedback records, the calculation compares the bucket's count with a baseline mean and variation. A sufficiently high positive Z-score can be flagged when there is enough nonzero data. A decrease is not flagged simply because it is equally far below the mean. A sparse or flat series may not produce a flag.
Empty time buckets can contribute zeros to the anomaly baseline. This is separate from Mean's treatment of missing values in a numeric field.
For Percentage, detection uses the underlying numerator aggregate rather than the displayed percentage rate. Review the numerator and configured denominator before interpreting a flag; it does not by itself show that the percentage rate is unusual.
Recurring seasonal patterns, changing source coverage, and changes in overall feedback volume can affect the result. Do not treat the flag as proof that seasonality has been removed, that a release caused a change, or that a topic's underlying customer impact has increased.
Inspect the evidence
Hover over the chart to inspect its values. Click a flagged point for available expected-mean details and supporting-record actions. Open a linked result or View Feedback action when available to review the records. Check the category, time bucket, filters, and measured object used by that view.
Use the chart and table values to investigate the change. The current tooltips do not provide an exact Z-score for every flagged point.
Save, share, or report on the analysis
After applying changes to the anomaly settings, use Save or Update when permitted to store the analysis configuration. Reopen the saved analysis to check its chart, threshold, and selected values.
Review which values are selected: manually selected values remain tracked even when they are no longer anomalous. Use the offered Top selection when you want the result to follow the current top anomalies rather than a fixed list.
Use Copy Link to share a saved analysis with teammates who have access, or use an offered export action. For scheduled email or Slack delivery, add the analysis to a Dashboard and configure that Dashboard's subscription; see Reports in Enterpret.
Review the saved chart, selections, and subscription settings before relying on the report. A Dashboard subscription should not be treated as a dedicated alert containing only anomalies first detected since the previous delivery.