Compute & rightsizing

Rightsize or stop idle computeCompare CPU/memory utilization against instance size; downsize or stop instances running well under their provisioned capacity.CostMon CostMon surfaces per-service AWS spend trends so an oversized instance's cost stands out before you have to go hunting for it in the console.
Clear orphaned volumes, snapshots, and IPsAudit for unattached EBS volumes, snapshots past their retention need, and unassociated Elastic IPs — all billed whether anything uses them or not.
Schedule down non-production environmentsStop dev, staging, and test instances outside business hours instead of leaving them running around the clock.
Revisit autoscaling policies against real trafficCheck scale-out thresholds and max instance limits against actual peak traffic, not the peak you originally provisioned for.

Storage & data transfer

Move cold data to a cheaper storage tierIdentify objects and backups that haven't been accessed in months and move them to infrequent-access or archive tiers.CostMon CostMon's normalized, service-level view makes a growing storage line easy to spot over time, across providers.
Audit cross-AZ and egress data transferCross-region, cross-AZ, and internet egress traffic bills per GB — check whether services could be colocated to avoid it.
Set lifecycle policies instead of manual cleanupAutomate expiration and tiering rules on buckets and volumes so cleanup doesn't depend on someone remembering to do it.
Review backup frequency and retentionConfirm backup cadence and retention windows match your actual recovery requirements, not a default nobody has revisited.

Track cost per model, not one blended totalBreak spend down by model and by workload — a blended total hides whether an increase is a deliberate upgrade or a regression.CostMon CostMon's Anthropic connector pulls per-model, per-day usage so this breakdown doesn't require a spreadsheet.
Set an expected cost per request before shippingEstimate a rough cost per request or per task before a feature ships, so a regression is visible immediately, not a month later.
Watch for agent-loop and retry fan-outA single logic bug in an agent loop or retry policy can multiply call volume, not just per-call cost — watch call counts, not only spend.
Use prompt caching for repeated contextReuse cached context for calls that repeat the same system prompt or reference documents instead of resending it every time.
Compare models on cost-per-outcomeThe cheapest model per call isn't always the cheapest per completed task — compare on results, not the sticker price.

Revisit reserved instances & savings plans quarterlyCompare current commitment coverage against your actual footprint — commitments sized for last year's usage rarely fit this year's.CostMon A normalized daily view makes it easy to compare current usage against what a commitment was originally sized for.
Match commitment term to confidence in the workloadUse shorter or convertible commitments for workloads you're not confident will run steady for the full commitment term.
Check for underutilized or expiring commitmentsFlag reserved capacity sitting unused, or commitments expiring soon, so you can renew, resize, or let them lapse on purpose.
Revisit enterprise discount agreements as spend growsProvider discount tiers and enterprise agreements are usually renegotiable once your spend has grown past the terms they were set at.

Normalize spend into one daily viewPull every provider's billing data into the same schema, currency, and cadence so totals are actually comparable.CostMon This is what CostMon does first: aggregates AWS, Anthropic, and the rest of your stack into one deduplicated daily table.
Alert on deviation from a service's own baselineA single service moving sharply off its own historical trend is a stronger signal than total spend moving at all.CostMon CostMon watches each connected service's daily trend and can flag a deviation from its own baseline.
Route cost alerts to an owner, not a shared inboxAn alert that lands on a specific engineer's radar gets acted on; one that lands in a shared inbox decays into being ignored.
Put a recurring spend review on the calendarA short, regular review against a normalized view catches drift while it's small, instead of at invoice time.

Tag spend by service, team, or projectConsistent tagging at the point resources are created makes it possible to attribute cost, not just see a total.
Use read-only, cost-scoped credentials for toolingAny tool that only needs to read billing data should never be granted write access to workloads or infrastructure.CostMon CostMon only ever requests read-only, cost-scoped access to each connector — never write access to your infrastructure.
Audit unused SaaS seats and shadow toolsCross-check active users against paid seats for every SaaS subscription — an unused seat is pure waste with no offsetting benefit.
Decommission cleanly when a project endsKeep a checklist for tearing down a project's resources when it ends, instead of relying on someone remembering to look later.