Cloud Cost Optimization: A Practical Guide to Cutting Your AWS, Azure or Google Cloud Bill

By Muhammad Sami · Nov 3, 2026 · 6 min read
Cloud bills usually grow through many small decisions. Someone leaves a test environment running, a database stays sized for a launch that passed long ago, and logs pile up because no one set a retention rule. Together, these items can make up a large share of the invoice.
Cloud cost optimization means finding that spend and keeping it from coming back, without hurting reliability or slowing your team down. This guide covers the order we recommend working in, for teams on AWS, Azure or Google Cloud.
Step 1: Make the spend visible
Before changing anything, work out which product, team or customer each pound or dollar pays for.
- Use the native tools first. AWS Cost Explorer, Azure Cost Management and Google Cloud's billing reports are all included and show where most of the money goes.
- Apply a tagging standard. Agree on a small set of tags, such as environment, product and owner, and apply them to every resource. Enforce them through infrastructure as code so nobody has to remember them.
- Separate environments. Separate accounts or projects for production, staging and development make costs easier to read and limit the damage of mistakes.
- Set budgets and alerts. A budget alert or anomaly notification can catch a runaway resource within a day or two, well before the end of the month.
Sort the results by cost and start at the top. A few services usually account for most of the bill.
Step 2: Remove obvious waste
Start here, because you are removing things nobody uses and the risk is low.
- Unattached storage volumes left behind after someone deleted the servers.
- Old snapshots and backups beyond your retention requirement.
- Idle load balancers and unused public IP addresses, which providers can still bill when nothing is behind them.
- Forgotten proofs of concept, especially GPU instances from AI experiments, which are expensive to leave running.
- Non-production environments running 24 hours a day. If developers work a normal week, a schedule can stop dev and test systems overnight and at weekends, which removes a large share of their running hours.
Before deleting anything, confirm the owner and take a final backup if the data might matter. The tags from step 1 make owners quicker to find.
Step 3: Rightsize what remains
Teams often size servers and databases in a rush, for a worst case, and never look at them again. Rightsizing means matching capacity to real usage over a representative period.
Compute
Look at CPU and memory use over several weeks, including your busiest periods. AWS Compute Optimizer, Azure Advisor and Google Cloud's equivalent all offer recommendations. Treat those as a starting point, since they cannot know about a seasonal peak or a planned launch. Newer instance generations often give better price and performance, and ARM-based options are worth testing if your software supports them.
Autoscaling
If your traffic varies during the day, scaling automatically usually costs less than running at peak size all the time. Your application must be able to run across multiple stateless instances, a pattern we cover in building web apps that scale.
Databases
Databases are often the largest and least flexible item on the bill. Before moving to a bigger instance, check for slow queries and missing indexes, since fixing those can remove the need for an upgrade. Non-production databases rarely need read replicas or high-availability settings, so review those too.
Step 4: Fix storage and data transfer
These costs grow slowly and are easy to miss in a monthly review.
- Storage tiers. Use lifecycle rules to move infrequently accessed objects to cheaper tiers and delete what you no longer need. Check retrieval charges first, since cold tiers can be costly to read from.
- Log retention. Logging and monitoring can become a major cost. Set a retention period for each log type, and turn off verbose logging in production unless you are debugging.
- Data transfer. Providers can charge for traffic between regions, between availability zones and out to the internet. Managed NAT gateways are a common surprise on AWS. Placing services that exchange a lot of data close together and putting a CDN in front of static content can help.
Step 5: Commit once usage is stable
All three major providers offer lower rates in exchange for a commitment. AWS has Savings Plans and Reserved Instances, Azure has reservations and savings plans, and Google Cloud has committed use discounts. These suit steady workloads.
The order matters, because committing before you rightsize locks in the waste. Clean up first, find the baseline usage you are confident will continue, and commit to that level rather than your peak. Choose shorter terms if your architecture is still changing.
For interruptible work such as batch jobs, data processing and some CI pipelines, spot or preemptible capacity is much cheaper. The provider can take that capacity back at short notice, so the workload must be able to stop and resume safely. Most customer-facing databases and single-instance services are a poor fit.
Step 6: Keep it from creeping back
After a one-off clean-up, the bill usually starts to climb again. FinOps addresses this by making engineering, finance and product share responsibility for cloud spend.
- Review costs on a regular schedule. A short monthly meeting with named owners is enough for most small and mid-sized businesses.
- Track a unit cost. Cost per customer, per order or per thousand requests tells you more than the total. A growing bill can be healthy if the unit cost is flat or falling.
- Make cost part of engineering decisions. Include a cost estimate in architecture reviews, and show teams the cost of what they run.
- Use infrastructure as code. When code defines every resource, reviewers can catch oversized resources and teams can tear down environments easily. Our DevOps consulting guide explains how this fits into a wider delivery process.
Common mistakes to avoid
- Cutting redundancy to save money. Removing a backup or a standby can make an outage far more expensive. Decide what downtime is acceptable before you change anything.
- Chasing small items first. Work through your largest services before spending time on tiny line items.
- Buying commitments too early. See step 5.
- Ignoring engineering time. Moving to a cheaper service saves nothing if it takes more staff time to run than it saves on the bill.
- Treating it as finance's problem. Finance can see the bill, but engineers are the ones who can change it.
When to bring in outside help
A capable in-house team can do many of these steps. Consider outside help if nobody has time to own the work, if costs are rising faster than your business and you cannot explain why, or if the fix involves architecture changes. It can also help right after a migration, because systems moved as they were often keep the sizing of the old environment. If a move is still ahead of you, our article on cloud migration cost factors will help you plan the budget.
Get a second opinion on your cloud bill
Scalovanta's cloud and DevOps team works on migrations and cost optimization for businesses in the USA and UK, on AWS, Azure and Google Cloud. If your bill keeps growing and you want to know why, book a free consultation and we will talk through where to look first.