1. Logging nobody reads
Log ingestion and retention add up quietly. In one estate we reviewed, CloudWatch was a large share of the monthly AWS bill. Review what each service logs, cut debug-level logging in production and set retention periods that match what you actually need for operations and compliance.
2. NAT gateways in every network
Each NAT gateway carries an hourly charge plus data processing fees. Estates that grew one project at a time often have several where one or two would do. Consolidating them is a common saving.
3. Leftovers from old projects
Look for client VPN endpoints nobody connects to, site-to-site VPNs to servers that no longer exist, load balancers with no healthy targets, unattached volumes and forgotten environments. Confirm each with its owner before removing it, keep a final snapshot where data is involved, and record what you removed.
4. Usage-based tools on the same card
Developer tools and AI APIs increasingly bill by usage. When one changed its billing model in an estate we worked on, the usage charges pushed a company card over its limit and put the AWS account, paid from the same card, at risk of suspension. Set hard spending caps on usage-based tools, and consider paying AWS separately from other subscriptions.
5. No budgets or alerts
AWS Budgets can alert you when spend crosses a threshold. Set budgets per account or per system, and send alerts to someone who will act on them, not to an unread inbox. Track spend monthly against approved budget lines so that variances are explained before finance asks.
6. Region choice made by accident
Prices differ by region. The Cape Town region is closest to Zimbabwe and can matter for data protection, but it usually costs more than the largest US regions. Make the region decision deliberately, with latency, cost and the Cyber and Data Protection Act all on the table. See our guide to hosting personal data on AWS.
What not to cut
Do not save money by switching off WAF, backups or the logging you need for security investigations. Make them cheaper instead: shorter retention for noisy logs, lifecycle rules for old backups and sampling for high-volume traffic.
This guide is general information from an engineering perspective, not legal or financial advice.
