Skip to content
FinOps workspace
Overview / AWS cost review / CloudWatch
COST LEDGER / DECISION GUIDE

AWS CloudWatch cost optimization: logs, metrics and queries

Published: 6 November 2026

Review log ingestion, retention, metrics and queries while preserving the evidence your operations team needs.

Review my CloudWatch costs
Server cabinet in the mint and deep green Cost Ledger illustration style
Connect monitoring usage with the workload decision.

If CloudTrail feeds your logging route, the CloudTrail cost optimization guide helps separate its collection charges from the CloudWatch part of the bill. Keep a shared inventory of the destinations being reviewed.

Build a CloudWatch cost inventory

The official CloudWatch pricing page separates monitoring features and their charging units. Record the applicable Region, billed quantity and rate for the features actually enabled. A single monitoring subtotal does not identify the change to make. Keep the review inside your AWS cost optimization scope so that a change in observability is assessed alongside the application it supports.

Review areaEvidence to collectDecision to validate
Log collectionSources, incoming volume and useful event typesWhich events must the application emit?
RetentionLog groups, effective retention policies and storageHow long is each category needed?
Metrics and API usageCustom metric definitions, consumers and request patternsWhich measurements and polling intervals serve a decision?
Queries and dashboardsGroups selected, query windows and refresh settingsWhich scans support active investigation?

Ask application owners to explain the largest lines before reducing collection. Use the EC2 workload inventory and the EKS cost review to connect monitoring demand with the resources that generate it. A workload migration or traffic increase can change both application and monitoring costs; record it in the baseline.

Separate ingestion, retention and log class decisions

AWS documents indefinite log retention by default for Standard and Infrequent Access log groups, with configurable retention at log group level. An applicable account-level retention policy overrides the group setting. Establish the effective policy before estimating a change. Agree the required investigation history with the operations and security owners, then document the approved retention for each category. Shortening retention can delete existing history as well as affect future events. Approve that data-loss boundary before applying the policy.

  • List each log source and the person who approves its collection policy.
  • Separate diagnostic detail from events needed for operational evidence.
  • Test a logging change during representative application behavior.
  • Record which investigation steps must remain possible after the change.

The CloudWatch log class documentation distinguishes Standard from Infrequent Access. Infrequent Access has a reduced feature set; its lower ingestion price does not mean lower storage or Logs Insights query charges. The same documentation describes a separate Delivery class for forwarding Lambda logs to Amazon S3 or Amazon Data Firehose: it keeps events in CloudWatch for a fixed two days and does not support Logs Insights queries. A log group's class cannot be changed after creation. Treat a new class as a logging design decision and validate the required features before routing new events to it.

Review query scope and monitoring consumers

The Logs Insights query guidance recommends selecting the necessary log groups and the narrowest useful time range. It also explains that each dashboard refresh starts a new query. Review unattended dashboards and their refresh intervals with their users. Preserve the queries needed for active incident investigation, and measure the actual effect of an adjusted schedule.

The CloudWatch cost analysis guide covers custom metrics, API requests and other monitoring features. Use its billing dimensions to identify the services and consumers behind spend. Review metric definitions and external polling patterns before changing them. Ask the monitoring owner to specify the resolution and visibility required by the response process.

  1. Capture the billed usage and the monitoring configuration for a representative period.
  2. Select one source, query pattern or monitoring consumer with its owner.
  3. Define the required diagnostic evidence and test the candidate configuration.
  4. Apply an approved change, reconcile its billed effect and record any workload variation.

Measure the result and prepare the handover

Keep monitoring savings in the realized savings register with the original usage basis, approved configuration and measurement period. Compare equivalent operating requirements. Lower spend following a traffic reduction is a different explanation from lower spend following a monitoring change. Keep migration effort and recurring administration in the cloud TCO comparison when evaluating a different logging arrangement.

A quote should state whether the work includes diagnosis, application logging changes, infrastructure configuration or ongoing review. Ask for queries and inventory exports your team can maintain. Identify who can authorize retention changes and who owns the updated runbook. The handover should explain the accepted configuration and the evidence needed to detect a later cost regression.

Questions before a CloudWatch cost review

Does shorter log retention reduce every CloudWatch charge?

No. Retention concerns stored history. Review ingestion, queries, metrics and other enabled features separately using their applicable billing units.

Can I switch an existing log group to Infrequent Access?

AWS states that a log group's class cannot be changed after creation. Assess the required features and future collection design before creating a group in a different class.

Can dashboard refresh settings affect query cost?

For Logs Insights widgets, each refresh starts a new query. Review the selected log groups, time range and refresh frequency against the dashboard's operational purpose.

Get my FinOps quote