Most lists of AWS cost optimization best practices are long and unordered, yet the order decides whether the savings stick. Buy Savings Plans before you delete idle resources and you lock that waste in for a year or three. Flexera's 2026 State of the Cloud report puts self-estimated waste at 29% of IaaS and PaaS spend, reversing a five-year decline, so the problem does not fix itself.
Here are the ten practices in the order I'd apply them, each mapped to a maturity stage:
- See where the money goes with Cost Explorer views by service, account and usage type (Stage 1).
- Set budgets and anomaly alerts as your first cost controls (Stage 1).
- Tag resources and activate cost allocation tags so every dollar has an owner (Stage 1).
- Delete the waste that keeps coming back: unattached volumes, old snapshots, idle IPs, logs kept forever (Stage 1).
- Right-size compute and storage with Compute Optimizer recommendations (Stage 2).
- Stop non-production environments outside working hours (Stage 2).
- Buy Savings Plans or Reserved Instances only after cleanup, and use Spot where interruptions are fine (Stage 2).
- Re-architect for price: Graviton, Lambda tuning, data transfer and NAT gateway paths (Stage 3).
- Enforce cost control across accounts with tag policies, SCPs and Budgets actions (Stage 4).
- Make cost management a habit with showback, unit costs and a fixed review cadence (Stage 5).
If you would rather have someone find the savings for you, our AWS Cost Optimization review traces your bill to the resources behind it and ranks each saving by dollars a month and production risk. The fixes are paid from a share of the verified savings, so no savings means no fee.
Cost Optimization vs Cost Management vs Cost Control
The three terms get used interchangeably, but they describe different work, and a healthy AWS account needs all three:
- Cost management is the ongoing practice of seeing, allocating, budgeting and forecasting spend. It answers "where does the money go, and who owns it?" Billing hygiene such as consolidated billing and activated tags lives here.
- Cost control is the set of guardrails that stop spend before it happens or flag it when it drifts: budgets and their alerts, Budgets actions, service control policies, tag policies.
- Cost optimization is changing resources and pricing models so you pay less for the same outcome: rightsizing, commitments, Graviton, architecture changes.
Most AWS cost optimization strategies fail on sequence. Without cost management you can't tell which instance to resize. Without cost controls, last quarter's cleanup grows back. And commitments bought before cleanup look great on a coverage report while you pay for capacity you were about to delete. The five stages put the three in an order that avoids those traps.
The Five-Stage AWS Cost Optimization Maturity Model
Each stage builds a capability the next one depends on:
- Stage 1 (Visibility): you can see where the money goes.
- Stage 2 (Tactical optimization): you right-size, schedule and commit to steady usage.
- Stage 3 (Strategic optimization): you make architecture decisions that lower the price of running a workload.
- Stage 4 (Governance): you enforce cost controls across accounts and teams.
- Stage 5 (FinOps culture): cost is part of how engineering, finance and leadership work every week.
Most teams sit between Stage 1 and Stage 2. That's fine: each stage pays for itself, so steady progress beats rushing ahead. For a longer list grouped by maturity level, use the AWS cost optimization checklist.
How the Stages Map to the Well-Architected Cost Pillar
The Well-Architected Cost Optimization Pillar gives you AWS's vocabulary for this work. Its five design principles are: implement cloud financial management, adopt a consumption model, measure overall efficiency, stop spending money on undifferentiated heavy lifting, and analyze and attribute expenditure. Its five best-practice areas map onto the stages like this:
| Pillar best-practice area | Where it lands |
|---|---|
| Practice Cloud Financial Management | Stages 1 and 5 |
| Expenditure and usage awareness | Stages 1 and 4 |
| Cost-effective resources | Stages 2 and 3 |
| Manage demand and supply resources | Stages 2 and 3 |
| Optimize over time | Stage 5 |
If you plan a formal review against the pillar, see what a Well-Architected review costs before you book one.
Self-Assessment: Identify Your Current Stage
You're at the first stage where any answer is "no":
Stage 1:
- Is Cost Explorer turned on and checked at least weekly?
- Can you name your top five services by spend without looking?
- Do budget alerts fire before a bad month ends?
Stage 2:
- Have you acted on rightsizing recommendations in the last 90 days?
- Do non-production environments stop outside working hours?
- Is your steady On-Demand usage covered by Savings Plans or Reserved Instances?
Stage 3:
- Have you tested Graviton for your main compute workloads?
- Do you know what NAT gateways and cross-AZ traffic cost you each month?
- Are high-traffic Lambda functions memory-tuned?
Stage 4:
- Do Service Control Policies enforce cost guardrails?
- Can each team see the spend it is accountable for?
- Is cost allocation tagging enforced in every account?
Stage 5:
- Do developers see the cost of a change during design and code review?
- Are cost KPIs reviewed next to performance and reliability metrics?
- Do engineering and finance meet on cost at a fixed cadence?
Stage 1: Cost Visibility and Awareness
Stage 1 gives you a clear view of where the money goes and an alert when spending drifts. It costs almost nothing to set up and usually surfaces the first round of waste on its own.
Setting Up AWS Cost Explorer
Cost Explorer is not on by default. You enable it by opening it once in the Billing and Cost Management console of your management account (or your only account), and member accounts can only use it after the management account has done so. AWS then prepares data for the current month and the previous 13 months.
Set up three saved views:
- Daily unblended cost by service, to see which services drive the bill and spot a jump the day it happens.
- Monthly cost by linked account. In a multi-account architecture managed with AWS Organizations, this shows which accounts carry the bill.
- Cost by usage type, which splits a service into its parts. For EC2 you'll see compute, EBS, data transfer and public IPv4 addresses as separate lines.
Watch automation: the Cost Explorer API charges $0.01 per request, and each page of a paginated result counts as a request. Scripts that poll it in a loop add up.
Budgets and Anomaly Alerts: Your First Cost Controls
AWS Budgets turns visibility into your first cost control. Configure alerts on both actual and forecasted spend, because the forecast alert warns you while there is still time to act. I'd start with three budgets:
- An account-level budget at 110% of your expected monthly spend, with alerts at 80%, 100% and 110%.
- Service budgets for your top three services by spend.
- A daily budget. If an average day costs $1,000, alert on any day above $1,500.
Budgets pricing costs little: monitoring and notifications are free, the first two action-enabled budgets are free, and each one after that costs $0.10 a day.
Add AWS Cost Anomaly Detection next to it. It is available at no additional charge, runs about three times a day, and can take up to 24 hours to flag an anomaly, so it complements budgets rather than replacing them. Turning on Cost Explorer also sets up a default services monitor with a daily summary alert. These belong with the other cost-management basics every account needs.
Basic Tagging Strategy for Cost Allocation
Without tags you see total spend but can't attribute it to teams, projects or environments. Start with four keys:
| Tag key | Purpose | Example values |
|---|---|---|
| Environment | Separate production from the rest | production, staging, development |
| Project | Attribute costs to initiatives | website-redesign, mobile-app |
| Owner | Identify the responsible team | platform-team, data-engineering |
| CostCenter | Map to financial tracking | CC-1234, engineering-ops |
Tags are case-sensitive, so Environment and environment are two different keys. Write the convention down before you roll it out.
Tags don't show up in billing data until you activate them as cost allocation tags. A new key can take up to 24 hours to appear for activation and up to another 24 hours to activate. If you activate late, a backfill request (console, or aws ce start-cost-allocation-tag-backfill) applies the current activation status to up to the previous 12 months. Backfill changes which keys count, not which resources carried them, so tag at creation.
Quick Wins: The Waste That Keeps Coming Back
With visibility in place, the same handful of items shows up in almost every account. If your AWS costs keep increasing and you need to stop the bleeding this week, clear these first:
- Idle EC2 instances. Compute Optimizer marks an instance idle when peak CPU stays below 5% and network I/O below 5 MB a day over 14 days, and on Business Support or higher, Trusted Advisor cost checks surface similar candidates. Check the workload before terminating, since some low-utilization systems are legitimate.
- Unattached EBS volumes. They cost the same as attached volumes, based on provisioned size and type. Non-root volumes default to
DeleteOnTermination=false, so every terminated instance can leave one behind. - Old EBS snapshots. Snapshots are billed separately, and deleting a volume does not stop its snapshot charges. Monthly or yearly snapshots you keep for 90 days or longer and rarely restore can move to EBS Snapshots Archive; archiving daily incremental snapshots can cost more.
- Public IPv4 addresses. Since February 1, 2024, AWS charges $0.005 per IP per hour for every public IPv4 address, attached or not. Use this script to clean up unused Elastic IPs across all regions.
- gp2 volumes. gp3 is 20% cheaper per GiB and includes 3,000 IOPS and 125 MiB/s at any size. The saving shrinks if you buy extra gp3 performance to match a large gp2 volume; CloudBurn's EBS pricing breakdown has the rates.
- Dev and test databases running 24/7. Schedule them with the rest of non-production.
- NAT gateways in dev and test VPCs. They bill per hour and per GB processed; see data transfer and NAT gateway costs.
- CloudWatch log groups with no retention. Logs never expire by default; set a retention period between 1 day and 10 years. CloudBurn's CloudWatch pricing breakdown shows what storage costs over time.
Stage 1 Checklist
- Cost Explorer turned on in the management account and reviewed weekly
- Account, service and daily budgets plus Cost Anomaly Detection alerting
- Environment, Project, Owner and CostCenter tags applied and activated
- Idle instances, unattached volumes, old snapshots and unused IPs cleaned up
These four are the minimum. The full AWS cost optimization checklist covers each area in more depth.
Stage 2: Tactical Optimization
Stage 2 acts on the obvious opportunities: resize what's too big, stop what isn't needed, move data to cheaper storage, then commit to the usage that's left. None of it requires architecture changes.
Right-Sizing with Compute Optimizer
AWS Compute Optimizer analyzes utilization metrics and recommends better-sized resources across EC2 instances, Auto Scaling groups, EBS volumes, Lambda functions, ECS services on Fargate, RDS and Aurora, NAT gateways and more. It is free, except for the paid enhanced infrastructure metrics option.
By default it looks back 14 days, with a 32-day option, and enhanced metrics extend that to 93 days for EC2, Auto Scaling groups and RDS. EC2 instances need at least 30 hours of metrics in the lookback window before you get a recommendation.
EC2 instances are classed as over-provisioned, under-provisioned or optimized, and every EC2 recommendation carries a performance risk rating from very low to very high. To act on them:
- Start with very low risk recommendations in Cost Optimization Hub or the Compute Optimizer console.
- Check memory and network metrics in CloudWatch, not only CPU.
- Resize stateless workloads in a maintenance window, and migrate stateful ones to a new instance.
Once the management account opts in to Cost Optimization Hub, savings estimation mode includes your Savings Plans and Reserved Instance discounts for member accounts too. Without it, member account estimates use On-Demand prices and overstate the saving.
Storage Classes and Lifecycle Policies
S3 costs creep when data sits in S3 Standard forever. For rates per class, see CloudBurn's S3 pricing breakdown; the decisions that matter are these.
For data with unknown or changing access, use S3 Intelligent-Tiering. It moves objects to Infrequent Access after 30 days without access and to Archive Instant Access after 90, with no retrieval charges. Objects under 128 KB are never auto-tiered and pay no monitoring fee; larger objects pay a small monitoring charge.
For predictable patterns, lifecycle rules are cheaper. Two rules trip people up: objects must be stored at least 30 days before they can move to Standard-IA, and since September 2024 S3 doesn't transition objects under 128 KB by default, because each transition is a billed request. Run S3 Storage Class Analysis first to see real access patterns.
Scheduling Non-Production Workloads
Development, test and staging environments rarely need to run around the clock. The Well-Architected pillar's own example: a dev environment used 8 hours a day runs 40 of 168 weekly hours, a potential saving of 75%.
Instance Scheduler on AWS starts and stops EC2 instances, RDS instances and Aurora clusters by tag, and AWS cites up to 70% savings when a weekly schedule drops from 168 to 50 hours. If you only need EC2, the Resource Scheduler in Systems Manager Quick Setup is simpler, but it doesn't cover databases.
Plan around two database details. A stopped RDS instance restarts automatically after 7 days, so the schedule has to stop it again. Aurora Serverless v2 can auto-pause at zero capacity on recent engine versions, which suits dev databases that can tolerate a short resume delay; you still pay for storage. CloudBurn's RDS pricing breakdown shows what an idle dev database costs.
Commitment Strategy: Savings Plans vs Reserved Instances vs Spot
Commitments give the biggest discounts, which is exactly why they come after cleanup. Here's how to choose:
Per the Savings Plans documentation, EC2 Instance Savings Plans save up to 72% but tie you to an instance family in one Region, while Compute Savings Plans save up to 66% and also cover Fargate and Lambda. CloudBurn's Compute vs EC2 Instance Savings Plans comparison covers the trade-off, and its Savings Plans guide explains terms and payment options. For databases, Database Savings Plans (December 2025) save up to 35% on Aurora, RDS, DynamoDB, ElastiCache and other database services, on a 1-year term with no upfront payment.
Spot Instances save up to 90% and suit batch jobs, CI/CD runners and containers built for interruption. EC2 gives a two-minute interruption notice, and a rebalance recommendation can warn you earlier.
Size commitments from your steady On-Demand baseline. Savings Plans discount eligible On-Demand usage, Spot is priced separately, Reserved Instances apply before Savings Plans, and unused hourly commitment doesn't roll over. AWS gives no fixed coverage percentage. Its recommendations come from 7, 30 or 60 days of On-Demand history, and the Purchase Analyzer's target coverage is meant to leave room to rightsize and clean up first. I'd start below your current baseline and add commitments in small steps.
Stage 2 Checklist
- Very low risk rightsizing recommendations applied
- Lifecycle rules or Intelligent-Tiering on your largest buckets
- Non-production EC2 and databases on an off-hours schedule
- Commitments sized from post-cleanup On-Demand usage, Spot used for interruptible work
Stage 3: Strategic Optimization
Stage 2 works inside your existing architecture. Stage 3 changes the architecture so the same workload costs less to run. The effort is higher, but the savings recur every month.
Graviton Migration for Better Price Performance
AWS says Graviton-based instances cost up to 20% less than comparable x86 instances, and its prescriptive guidance cites up to 40% better price performance. For per-hour comparisons, see CloudBurn's EC2 pricing breakdown.
The migration path depends on the workload:
- Containers: build multi-architecture images and run them on Graviton nodes in EKS, ECS or Fargate.
- EC2: test in staging first. Most Linux workloads on interpreted or JVM runtimes need no code changes, but native dependencies must have arm64 builds.
- Databases: RDS and Aurora offer Graviton instance classes, and read replicas or non-production databases are a low-risk place to start.
- Lambda: switch the architecture setting to arm64 unless the function ships x86-only binaries.
Start with stateless, horizontally scaled services you can roll back quickly, then expand.
Serverless Tuning and When EC2 Beats Lambda
Lambda bills for what runs, but configuration still decides the price. CPU scales with memory, and at 1,769 MB a function gets one vCPU, so more memory sometimes costs less because the function finishes faster. AWS Lambda Power Tuning runs your function at different memory sizes and shows the trade-off.
Switching to arm64 lowers the duration price by 20% per GB-second, according to the AWS launch announcement. Match provisioned concurrency to measured peaks, because idle provisioned concurrency is pure cost. For DynamoDB, on-demand mode fits unpredictable traffic and provisioned capacity with auto scaling fits steady traffic.
Lambda is the cheaper choice for spiky, event-driven work that scales to zero. Sustained, high-throughput, long-running work can be cheaper on EC2 or containers, but there's no universal break-even point. Compare with your own request volume and duration, and include Compute Savings Plans, which discount Lambda by up to 17% on duration and provisioned concurrency. CloudBurn's Lambda pricing breakdown walks through the math.
Data Transfer and NAT Gateway Costs
Data transfer is easy to overlook because it hides inside other services' line items. The NAT gateway is the usual culprit:
A NAT gateway charges per hour and per GB processed, whatever the source or destination. That includes traffic to S3 and DynamoDB, which a gateway endpoint carries with no hourly or processing charge. Interface endpoints for other services charge per AZ-hour and per GB, but usually less than sending the same traffic through NAT. For current rates, see CloudBurn's NAT gateway pricing breakdown.
Placement matters too. Traffic within one Availability Zone is free, while cross-AZ traffic costs $0.01 per GB in each direction. Keep chatty services and their NAT gateway in the same AZ where your availability design allows it. CloudFront in front of static content cuts origin data transfer.
Auto Scaling That Follows Demand
Target tracking policies can use high-resolution metrics with periods down to 10 seconds, which shortens the time you pay for capacity after a spike. Predictive scaling forecasts daily and weekly patterns and scales out ahead of them, but it never scales in, so pair it with dynamic scaling. Mixed instance types widen your Spot pool, and Compute Optimizer also flags oversized or idle Auto Scaling groups.
Stage 3 Checklist
- Graviton tested on at least one stateless workload
- High-traffic Lambda functions memory-tuned and on arm64 where compatible
- Gateway endpoints for S3 and DynamoDB in every VPC with a NAT gateway
- Production Auto Scaling groups on target tracking or predictive scaling
Stage 4: Cost Control and Governance Across Accounts
Stage 4 moves cost control from people to policy, so new accounts and teams start inside the guardrails instead of being cleaned up later.
Multi-Account Cost Strategy with AWS Organizations
AWS Organizations is the foundation of multi-account cost governance: consolidated billing plus policies that apply per organizational unit. Consolidated billing aggregates usage across accounts for volume discounts, and by default Savings Plans and Reserved Instances bought in one account apply across the organization. Separate production, non-production and sandbox OUs so each can carry a different level of control.
Cost Allocation with Tags and Cost Categories
Stage 1 introduced tags. Stage 4 enforces them. Tag policies standardize keys and allowed values across the organization, and an SCP can deny resource creation when required tags are missing.
Cost Categories group spend by business logic on top of tags and accounts:
- Combine tag values into one category, such as "team:platform" or "team:infrastructure" as "Platform Engineering".
- Roll projects up into business units and cost centers.
- Give untagged spend a default category instead of leaving it unallocated.
The result is a report finance can use without perfect tagging discipline from every developer.
Implementing Showback and Chargeback Models
Showback reports costs to teams for awareness. Chargeback moves those costs into team budgets. Start with showback and switch once teams trust the numbers.
| Model | Effort | Accuracy | Best for |
|---|---|---|---|
| Account-based | Low | Medium | Separate accounts per team/project |
| OU-based | Low | Medium | Costs grouped by organizational unit |
| Tag-based | High | High | Granular resource-level attribution |
Account-based allocation gives the best effort-to-accuracy ratio for most teams. Each team gets its own accounts and costs attribute themselves through consolidated billing, which fits a multi-account strategy built for workload isolation.
Preventive Cost Controls: SCPs and Budgets Actions
Reactive cost management catches problems after the money is spent. Preventive controls stop them first. Service Control Policies set guardrails that even administrators in member accounts can't bypass:
- Deny Regions you don't use.
- Limit non-production accounts to smaller, cheaper instance families.
- Block expensive services in sandbox accounts.
The SCP examples guide has production-ready versions, including cost control policies. Budgets actions add a second layer: when spend crosses a threshold, they can apply an IAM policy or SCP that blocks new launches until someone reviews it.
Stage 4 Checklist
- All accounts in AWS Organizations with consolidated billing
- Tag policies and Cost Categories in place
- Showback reports shared with every team
- At least one cost SCP and one Budgets action deployed
Stage 5: FinOps Culture and Ongoing Cost Management
Stage 5 makes cost management part of the normal working week instead of a yearly cleanup project. Teams that jump here without Stages 1 to 4 tend to hold meetings with nothing reliable to discuss.
How Often to Review AWS Costs
A fixed cadence catches drift before it compounds:
- Daily: anomaly alerts routed to the owning team. Cost Anomaly Detection runs about three times a day, so treat it as a next-day signal, not real time.
- Weekly: each team compares actual spend with its budget and looks at the trend.
- Monthly: engineering leads, finance and leadership review spend by team, unit costs and the open savings backlog.
- Quarterly: review Savings Plans coverage and utilization before buying more, plus the Graviton and architecture roadmap.
I'd run showback for 60 to 90 days before any chargeback, so teams can question the data while it carries no budget consequence.
Shift-Left Cost Estimation
Put an estimate in every architecture review, and check infrastructure changes before merge, for example with the open-source CloudBurn CLI, which runs cost checks against Terraform and CloudFormation. For one-off designs, the AWS Pricing Calculator now runs in the console and can include your discounts and commitments. Give developers sandbox accounts with budgets so they can experiment within limits.
For more depth, see the complete guide to AWS cost estimation and this comparison of AWS cost estimation tools.
KPIs and Metrics for Cost Optimization Success
Track metrics that change behavior, not only the total:
- Unit cost, such as cost per request, per customer or per GB processed. If it falls while the business grows, you're winning.
- Commitment coverage and utilization together. Raise coverage only after rightsizing, or you're committing to waste.
- Waste ratio: idle and over-provisioned spend as a share of the total, trending down.
- Time to remediation: how long recommendations sit before someone acts.
Stage 5 Checklist
- Anomaly alerts routed to owning teams
- Monthly cross-functional cost review on the calendar
- Cost estimates in architecture and pull request reviews
- Unit cost tracked next to performance SLOs
AWS Cost Management Tools by Maturity Stage
You don't need every billing tool AWS offers. These are the ones that carry each stage:
| Tool | Stage | What it does |
|---|---|---|
| Cost Explorer | 1 | Visualize and analyze spend |
| AWS Budgets | 1 and 4 | Thresholds, alerts and automated actions |
| Cost Anomaly Detection | 1 and 5 | Flags unusual spend, no additional charge |
| Trusted Advisor | 1 and 2 | Idle and underused resource checks |
| Compute Optimizer | 2 and 3 | Rightsizing and idle recommendations |
| Cost Optimization Hub | 2 and 3 | 18 recommendation types, deduplicated across accounts and Regions |
| Savings Plans Purchase Analyzer | 2 | Models commitment purchases and target coverage |
| Cost Categories | 4 | Business-level cost grouping |
| Cost and Usage Reports | 4 and 5 | Line-item data for custom analysis |
Cost Optimization Hub is free and accounts for your existing discounts, so it is the best single place to start reviewing recommendations.
Recent Tool Updates (2025-2026)
A few recent additions change how you work through the stages:
- Database Savings Plans (December 2025) extend commitment discounts to managed databases.
- The AWS Pricing Calculator became generally available in the console (May 2025), with support for discounts and purchase commitments.
- Cost Explorer added Cost Comparison (May 2025), which explains what changed between two months.
- Analyze with Amazon Q in Cost Explorer (June 2026) explains cost changes in plain language at no additional charge.
- Trusted Advisor gained 16 new checks from Cost Optimization Hub on Business Support or higher, once you opt in to Cost Optimization Hub and Compute Optimizer.
When to Consider Third-Party Tools
Native tools cover most needs. Third-party tools earn their cost in a few situations:
- Multi-cloud estates that need one view across AWS, Azure and Google Cloud.
- Kubernetes clusters where you need cost per namespace or pod.
- Automation beyond what native recommendations and Budgets actions can do.
Exhaust the native tools first, because a third-party platform adds its own bill and its own data to maintain.
Do It Yourself or Bring In Help?
Most of Stage 1 and Stage 2 is within reach of an in-house team. The native tools are free or cheap, and the work is mostly reading recommendations and acting on them. Do it yourself if one engineer can own cost work every week, you have few enough accounts to review by hand, and nobody is pushing for a large commitment purchase this quarter.
It tends to stall in three places:
- Cost work loses to the product roadmap. Cleanup is never urgent until the invoice is, so recommendations sit unread for months.
- Commitments are hard to undo. Compute and EC2 Instance Savings Plans run for one or three years, Database Savings Plans for one, and unused hourly commitment doesn't roll over. A wrongly sized purchase costs you every hour of the term.
- Cross-account questions need access individual teams don't have. Who owns the NAT gateway traffic, or which account a shared Savings Plan is covering, is a management-account question.
If that sounds familiar, consider an AWS cost review that ranks savings by production risk. We start from a read-only IAM role, trace the bill to the resources and architecture decisions behind it, and rank every change by monthly savings, effort and production risk. Your team can ship the fixes from that roadmap, or we implement them with you. Even on well-managed accounts we typically find 15 to 30 percent of waste.
To see the process from the inside, read what happens during a cost optimization assessment.
Conclusion and Next Steps
These AWS cost optimization best practices reward steady, repeated work. Three ideas carry the whole model:
- Order matters: visibility, then cleanup, then commitments, then architecture, then controls and culture.
- Commit only after cleanup, and size commitments from the usage that remains.
- Put cost controls in place before you scale to more accounts and teams.
Your next action by stage:
- Stage 1: open Cost Explorer today and create your first budget.
- Stage 2: apply the very low risk rightsizing recommendations this week.
- Stage 3: test one stateless workload on Graviton.
- Stage 4: enforce cost allocation tags and deploy one cost SCP.
- Stage 5: book a monthly cost review with engineering and finance.
In a multi-account setup, cost governance and security governance share the same tools. See AWS Organizations best practices for the full governance picture.
Next step