Insights

Insights says what the figures point to, and why. Find it under Insights in the sidebar.

Forecast

The forecast projects this month and the next, in the project's main currency.

• Each provider on its own, from its last complete day. Providers report spend a day late or later, and not all at the same pace.
• Each day left is estimated at its weekday's average over the four weeks before, since spend often runs lower at weekends. With under two weeks of a provider's history, the last seven days' average is used instead. A provider connected only days ago adds what it has spent so far, and no more.
• A range where the month lands four times in five, from how much spend has moved day to day. It never starts below what is already spent.
• A confidence: high with four weeks of history and steady spend, medium with two weeks, low otherwise.

There is no forecast before three complete days of the month. Budgets use the same forecast.

Anomalies

An anomaly is a day in the last month when a service spent outside its normal range:

• the expected figure is the median of the same weekday over the four weeks before;
• the normal range is three robust deviations either side of it, which a past spike does not widen;
• the move must also be at least 1 percent of the project's average daily spend, so a service moving by cents never counts.

Each provider is read through its last complete day, and needs two weeks of history. Both rises and drops are listed, each with what was expected, and Explore it opens Explore at that service and month.

What drove it divides the day's move by every dimension, as Explore divides a change between two months: the team, account, region, model and every other dimension, each with the values that moved most and the rest together. It measures the day against the average of the same weekday over the four weeks before, which it names, because an average divides exactly among what the spend was for and the median that finds the day does not.

Savings

Savings lists what you could save, largest first, each with where its figures come from:

• AWS's own recommendations, from Cost Explorer, priced at your account's rates: a one year Compute Savings Plan with no upfront payment, Reserved Instances for EC2, RDS, ElastiCache, Redshift and OpenSearch on the same terms, and EC2 instances to stop or resize. Each is based on the last 30 days, and read once a day.
• Azure Advisor's cost recommendations, such as virtual machines to resize or shut down and reservations to buy, with what Advisor estimates each saves a month at your rates, in every subscription CostReveal's application has Cost Management Reader on. See Connect Azure.
• Google Cloud Recommender's cost recommendations, such as machine types to change, idle resources and commitments to buy, from the recommendations export to BigQuery, for the projects your billing account pays for and the account itself. A saving Recommender states over a period other than 30 days is shown for 30 days. See Connect Google Cloud.
• Idle Kubernetes capacity: what each cluster's nodes billed last month for capacity no workload requested or used. Fewer or smaller nodes, or autoscaling, would save up to that.

Each cloud's recommendations are read once a day, and replace what it recommended before. Savings Plans, Reserved Instances and other commitments can cover the same usage, so their savings never add up; each is shown on its own.

Cost spikes and every alert

Cost spikes are the alerts raised when a day rises well above the seven before it, by the percentage set in Settings, Notifications. Every alert lists budget alerts and cost spikes as they were sent. See Budgets and alerts.