Skip to main content

Anomaly Detection

Anomaly detection identifies when a signal deviates significantly from its established baseline. It is the primary mechanism for catching unexpected cost spikes, usage changes, and resource count shifts.


How anomalies are detected

For each tracked signal, frugally.app compares the current value against the baseline bounds:

If signal value > baseline upper bound → anomaly (spike)
If signal value < baseline lower bound → anomaly (drop)

The baseline bounds are determined by the sensitivity configuration and the signal's historical variance.


Threshold configuration

You can fine-tune anomaly detection with additional threshold rules:

Threshold typeDescriptionExample
PercentageTriggers when the signal exceeds the baseline by a percentageAlert when daily spend is 20% above baseline
AbsoluteTriggers when the signal exceeds the baseline by a fixed amountAlert when daily spend is $100 above baseline
Standard deviationsTriggers based on statistical deviationAlert at 2 standard deviations (default)
Minimum amountOnly trigger if the absolute value exceeds a floorOnly alert if the anomaly represents at least $50 in additional spend

Configure thresholds at Settings > Monitoring > Anomaly Detection.


Suppression rules

Suppression rules prevent known, expected patterns from triggering anomalies:

Rule typeUse case
Service suppressionSuppress anomalies for a service with known variable costs (e.g. Lambda during batch processing)
Account suppressionSuppress anomalies for a sandbox or testing account
Time window suppressionSuppress during a planned event (e.g. migration week, Black Friday, load testing)
Minimum thresholdOnly alert on anomalies above a minimum dollar impact
Day-of-week suppressionSuppress weekend anomalies if your workload is weekday-heavy

Managing suppression rules

Navigate to Settings > Monitoring > Suppression Rules.

Each rule has:

FieldDescription
NameDescriptive label (e.g. "Batch processing — Lambda spikes expected")
ConditionWhat to suppress (service, account, time window, etc.)
ExpiryWhen the rule expires (optional — for time-limited suppressions)
ReasonWhy the suppression exists (for audit purposes)
tip

Always set an expiry date on time-limited suppression rules. Forgotten suppressions can mask real anomalies.


Cross-reference with Intelligence

Anomaly detection in the Monitoring section focuses on detection and alerting configuration. The Intelligence — Cost Anomaly Detection page focuses on investigating and resolving detected anomalies using the Cost Explorer.

AspectMonitoringIntelligence
PurposeConfigure detection rules and thresholdsInvestigate and resolve anomalies
ActionsSet sensitivity, create suppression rulesAcknowledge, dismiss, drill into cost data
ViewSignal-centric (baselines and thresholds)Cost-centric (spend breakdown and trends)

Both views show the same anomalies — they are two perspectives on the same data.


Anomaly lifecycle

Detected → Open → Acknowledged → Resolved / Dismissed
StatusMeaning
OpenAnomaly detected, not yet reviewed
AcknowledgedSomeone is investigating
ResolvedThe signal returned to within baseline bounds
DismissedClosed without action (with a reason)

Anomalies that resolve on their own (signal returns to normal) are automatically moved to Resolved status.