Cendryva
Join the waitlist

Anomaly Detection and Forecasting

Use this article to read anomaly flags and forecasts correctly before acting on them.

Availability: Plan-dependent6 min readLast reviewed September 3, 2026

How to Interpret Output

  • Anomalies identify values or patterns that differ materially from an expected range.
  • Forecasts estimate possible future values based on available data and configuration.
  • Neither is a guarantee or an autonomous decision.
  • Users must review data freshness, model context, source lineage, confidence, and business events.
  • Forecasts become less reliable when history is sparse or operating conditions change.
  • Recommendations should be reviewed before creating or assigning work.

Illustrative Example

Illustrative only: a weekly booked-work statistic drops sharply and is flagged as anomalous. Before opening an operational investigation, the reviewer checks whether the source synchronized on schedule, whether the reporting period was complete, and whether a known holiday explains the change.

In this example the flag is a prompt to check, not a finding. The decision is still made by a person who can see the business context.

Seasonality Is Not an Incident

Recurring patterns can be flagged as anomalous when history is short. Record known seasonal behavior in the statistic's description or related playbook so reviewers can dismiss it quickly.

Troubleshooting

Insufficient history.
Expect wide ranges and frequent flags until enough completed periods exist. Treat early output as directional.
Missing periods.
Gaps distort expected ranges. Confirm every period was collected before interpreting a flag.
Stale source data.
Check the last successful synchronization for each contributing source; a stale source can look like a real decline.
Sudden definition changes.
A formula or source change shifts the series. Version the change and expect flags around the effective date.
Low-confidence forecasts.
Use them to frame planning ranges, not commitments. Do not set a target solely from a low-confidence projection.
Expected seasonal behavior flagged as anomalous.
Document the seasonal pattern and review whether the statistic has enough history to represent the cycle.

Related Articles