Managing Anomalies

Definition

Anomaly Management is the ability to detect, identify, clarify, alert and manage unexpected or unforecasted cloud cost events in a timely manner, in order to minimize detrimental impact to the business, cost or otherwise.

Managing anomalies typically involves the use of tools or reports to identify unexpected spending, the distribution of anomaly alerts, and the investigation and resolution of anomalous usage and cost.

In the context of Cloud FinOps, anomalies are levels of spending that are different from the normal or expected spend.

Anomaly detection identifies data points, events, and/or observations that deviate from a dataset’s normal behavior. Detection tools & procedures allows the FinOps team to react quickly in order to maintain spend levels that an organization expects. To quickly find those needles in your cloud haystack, using automated, machine learning–based anomaly detection is key. These tools are generally offered by cloud providers and third party platforms.

Maturity Assessment

---

Crawl

• Understanding that anomalous spending might occur
• Manually checking for anomalous spending
• Reacting to anomaly active post billing
• Using budget alerts vs an anomaly detection service
• Limited granular detection minimizing context insights
• Unexpected spending is manually investigated and routed when necessary to a suspected owner for resolution

Walk

• Some form of automated detection or reporting or tooling (usually provided by the Cloud Service Provider in use
• Use of anomaly detection tooling in most or all departments and teams
• Context-relevant thresholds (percentage of spend change, single item spend amount ceilings, forecast breach alerts, etc.)
• Cost allocation metadata provides context to segment anomalies
• Unexpected spending automatically routed to responsible teams
• Create initial KPIs – # of alerts & spend associated with alert

Run

• Mature Anomaly detection tooling embedded across the organizations
• Automated handling of anomalous spending alerts, with appropriate severity
• Connected to event management or ticketing systems.
• Granular context-related anomaly alert thresholds linked directly to service components
• Alert thresholds iteratively updated in line with service lifecycles
• Analysis results in full RCCA post-mortem / future prevention

---

Functional Activity

As someone in a FinOps Practitioner role, I will…

• establish requirements for anomaly detection tool selection that is appropriate for cost monitoring and capable of defining, refining, detecting and alerting unexpected cloud spending events (cost anomalies).
• work with stakeholder teams to establish anomalous detection thresholds and reporting/notifications frequency
• document and communicate anomaly detection mechanism and thresholds to all stakeholders
• ensure that anomaly detection is tied appropriately to cost allocation metadata
• generate reports that surface all and/or alerted anomalous spending

As someone in an Engineering/Operations role, I will…

• ensure anomaly detection tooling has access to raw spending data at appropriate real-time granularity and frequency
• ensure that my team is aware of the correct processes and actions to respond to and address a cloud cost anomaly
• provide feedback to FinOps on the correctness of established thresholds and detection coverage
• determine the cause and resolve issues with anomalies

As someone in a Finance role, I will…

• help establish anomalous percentage thresholds that trigger financial review of budget to actual
• provide contingency capacity within forecasts for cloud costs to accommodate anomalous spending trends that align with established thresholds

Measure(s) of Success & KPI

Measures of success are represented in the context of cloud costs and may include one or more key performance indicators ( KPI ), describe objectives with key results ( OKR ), and declare thresholds defining outliers or acceptable variance from forecasted trends.

• The count of anomalies within a period of time; including consistent identification of a proper anomaly
• Spend associated with alert within a period of time; represents total anomaly detection
• Time to detect the anomaly; this is the time it takes from the anomaly reporting to someone acknowledging it.
• Duration of the anomaly; the length in time for resolution of the anomaly
• Time to investigate and address an identified anomaly; time of investigation of a true anomaly is real time wasted cost in many cases
• Educating teams on understanding how variable types of cloud spending can be by product, as to better define what is anomalous and who is accountable
• The count of actioned anomalies and spending amount avoided (to nearest billing period)
• The count of unactionable anomalies and categorized but justification to ignore (i.e. new service, performance testing, customer peak, false alert).
• Tracking number of alerts suspended (ignored) to identify teams who might not be adhering to protocol or policy

Inputs

• Cloud cost and usage data (API or File based)
• Anomaly detection tooling
• Cost allocation metadata established and aligned to the organization’s reporting needs
• Anomalous spend notification to stakeholder teams
• Stakeholder real-time visibility into cost and usage reporting data
• Documentation of triage process and expectation of personas

Get Involved

Get involved and contribute to the community by sharing your real world experiences related to this Capability in the form of a story or providing a playbook for how you have implemented best practices in your organization. Your real world experiences can be provided in the context of:

• one or more cloud providers
• the types of cloud services used (compute, storage, database, etc…)
• describe a combination of tooling, platform or vendor, and processes including KPIs
• the industry the organization belongs to
• the complexity of the organization (global enterprise, start-up, etc…)
• the [FinOps personas](https://www.finops.org/framework/personas/) involved / organizational roles

Join the conversation about this Capability in Slack . You can submit stories, how-tos and suggest improvements using one of the options for contributing here.