Service guide
How to reduce AWS Lambda costs
Lambda bills honestly — you pay for what runs. The waste hides in configuration: architecture, memory size and provisioned concurrency are all set once by guess and never revisited, and every invocation quietly writes paid CloudWatch logs.
The free scan checks architectures, provisioned concurrency and the log-ingestion side-cost per function, across every region.
Run the free scanThe Lambda configuration checklist
- →x86 → arm64: flat 20%. Graviton pricing is ~20% below x86 per GB-second, and most runtimes (Node, Python, Java) switch with a one-line configuration change. Unless a function ships native x86 binaries, this is free money.
- →Guessed memory sizes. Memory sets both RAM and CPU, so it is a real tuning knob: oversized wastes directly, undersized runs longer and can cost more overall. AWS Lambda Power Tuning (free, open source) finds the cost-optimal setting per function empirically.
- →Provisioned concurrency on quiet functions. It bills continuously whether invoked or not. Right for latency-critical hot paths; wrong for the internal API that gets 40 calls an hour. Review which functions still justify it — traffic patterns change, the setting does not.
- →The CloudWatch side-cost. Every invocation logs at $0.50/GB ingested. Chatty functions at high volume can cost more in logs than in compute — check the log group sizes next to the function costs, and set retention on all of them.
- →Zombie event sources. Functions still wired to schedules and queues for features that shipped their last release long ago — invoking, logging and billing on autopilot. Zero-value invocations are findable from invocation metrics vs. downstream usage.
Put numbers on it
Frequently asked questions
Is arm64 migration safe for production functions?
For interpreted runtimes it is usually a configuration change plus a redeploy through your normal pipeline — same code, same tests. The exceptions are functions bundling compiled x86 dependencies (native modules, binary layers), which need their dependencies rebuilt for arm64 first. Migrate function by function, not big-bang.
Our Lambda bill is small. Worth optimising?
Check the log groups before deciding — the pattern where Lambda compute is £40/month and its CloudWatch ingestion is £150 is common enough that we check it on every scan. The function list is also where zombie schedules surface, and those cost attention as well as money.
£499 fixed. Free scan first. 20%+ found or it’s free.
The scan is read-only — a role you create and delete, no keys shared — and shows your estimated monthly saving before anyone talks about money.