Serverless Cost Audit: The Half-Year Check-In

People ask if serverless stays cheap once you have real traffic. The short answer is yes, but only if you watch the right metrics. My serverless bill has stayed under $1/month for half a year. Here is how I audit it.

The Monthly Ritual

First of every month I open Cost Explorer and filter by service. DynamoDB, Lambda, SES, CloudWatch, S3. For each one I note: usage this month, usage trend, and whether any item is trending toward paid tier.

June 2025 breakdown:
- DynamoDB: 0.2 GB stored / 25 GB free = 0.8% used
- Lambda: 48k invocations / 1M free = 4.8%
- SES: 1,200 emails / 3,000 free = 40%
- CloudWatch: 0.3 GB logs / 5 GB free = 6%
Total charges: $0.12 (Route53 zone)

The SES line is the one I watch most. At 40% I have headroom, but the newsletter list is growing. When it hits 75% I will move to a paid-SES tier or rethink cadence.

The Traps

CloudWatch logs are the sneakiest cost. Default log retention is forever. I set 30 days on every log group and cut the 'you will grow into this bill' risk to zero.

DynamoDB on-demand is also a trap if traffic spikes. I run my tables on provisioned mode with the free tier allocation (25 WCU + 25 RCU). If I ever need more I switch, but predictable costs beat 'surprise spike' costs.

The Philosophy

Serverless is cheap because you pay for what you use. Which means someone needs to watch what you use. Monthly audits are that watch. Fifteen minutes prevents hundreds of dollars of lazy spend.

See the AWS Free Tier page for current limits. They change. The discipline of checking is what matters more than the numbers themselves.