Rules for Cost Anomalies

Cost Anomaly Detection identifies potential anomalies but does not take any actions to correct them. It is not accountable for the underlying causes of anomalies or any associated losses. Cost Anomaly Detection uses algorithms to forecast your costs based on your consumption model. The forecast is for reference only and may not be entirely accurate. For more information about forecasting, see Forecasting.
Cost anomalies are not identified in real time. There is some delay. For details, see Delay in Generating Cost Anomalies.

Cost Anomaly Detection has a specific monitoring scope. For details, see Monitoring Scope of Cost Monitors.
The cost type for cost anomaly detection is the net original cost (actual payment), also referred to as the actual cost.

Cost anomalies are not recorded in real time. Cost Center supports user-created cost monitors . The delay details are as follows:
- User-created cost monitors: There is a T+1 delay for the latest identified anomaly to appear, where T represents the day an anomaly occurs. For example, if a cost anomaly occurs on April 08 and meets the detection rules, Cost Center will generate an anomaly record on the afternoon of April 09. Anomalies are identified one day after they occur.
The delay in calculating the cost impact is consistent with that in generating cost anomalies. The details are as follows:
- User-created cost monitors
  - Suppose you found a cost anomaly in pay-per-use resources. If the anomaly was identified on April 10 and persisted until April 14, the anomaly lasted for five days. The cost impact equals the sum of the differences between the actual costs and the maximum forecasted costs from April 09 to April 13.
  - Suppose you found a cost anomaly in yearly/monthly resources. If the anomaly was identified on April 10 and persisted until April 14, the anomaly lasted for five days. The cost impact equals the cost difference between April 01 to April 13 and March 01 to March 13.

Cost Anomaly Detection applies to both pay-per-use and yearly/monthly costs.

Monitoring scope for yearly/monthly costs: costs whose bill type is Expenditure-new purchase, Expenditure-renewal, Expenditure-change, Expenditure-auto-renewal, or Expenditure-monthly payment.
Monitoring scope for pay-per-use costs: costs whose bill type is Expenditure-use.

Cost Anomaly Detection does not track cost anomalies in monthly-settled cloud services, such as CDN billed by 95th percentile bandwidth.

Anomaly Cost Detection identifies anomalies in your pay-per-use and yearly/monthly costs.

Detection rules for pay-per-use costs: AI algorithms are used to identify cost anomalies. If the actual cost is greater than the maximum forecasted cost and the difference is greater than $1 USD, a cost anomaly will be recorded.
Detection rules for yearly/monthly cost: If the MoM growth rate of MTD costs is greater than the configured threshold (20% by default, which can be changed) and the difference is greater than $1 USD, a cost anomaly will be recorded.

Cost impact on pay-per-use resources

Cost impact = Actual cost on the current day – Maximum forecasted cost

If the cost anomaly persists, the impacts will accumulate. During the anomaly period, the cost impact equals the sum of the daily differences between the actual costs and the maximum forecasted costs.
Cost impact on yearly/monthly resources

Cost impact = Cost for the current month – Cost for the same period in the previous month

There are three levels of severity for cost anomalies, depending on the cost impact percentage.

Minor: > 0% and < 20%
Major: ≥ 20% and < 50%
Critical: ≥ 50%
There are slight differences in how the impact percentages for pay-per-use and yearly/monthly resources are calculated.
- Impact percentage of pay-per-use cost anomalies = (Actual cost – Maximum forecasted cost)/Maximum forecasted cost
- MoM growth rate of yearly/monthly costs = (Cost for the current month – Cost for the same period in the previous month)/Cost for the same period in the previous month