Engagement at a glance
- Client profile: Series B payments FinTech, Dubai; AWS spend growing 3x faster than revenue for two consecutive quarters.
- Shape: 6-week fixed-scope FinOps project, read-only access first, change windows agreed with compliance.
- Result: 47% monthly AWS bill reduction, sustained 6+ months; $312K annualized; zero production incidents during right-sizing.
- Handover: tagging policy, teardown automation, dashboards and runbooks owned by the client's platform team on exit.
The situation: spend growing 3x faster than revenue
The company processed card and wallet payments across the Gulf, ran everything on AWS in me-south-1 and eu-west-1, and had just closed a Series B — which is exactly when boards start reading the infrastructure line. Finance saw a bill compounding quarter over quarter; engineering saw nothing wrong, because nothing was wrong operationally. Everything worked. It just cost 3x what it should have.
The first week's bill audit told the usual story, with three headline offenders. Production RDS instances sized for a traffic projection two years stale, running multi-AZ at roughly 20% utilization. NAT gateway data processing charges nearly the size of the compute line under them, because every service call to S3, SQS and DynamoDB from private subnets was being routed through NAT. And 40+ "temporary" dev and staging environments, some untouched for months, each a full copy of production networking, each billing 24/7.
The constraint: a payments company can't wing it
A payments platform under PCI-DSS obligations and central-bank oversight doesn't get to "just resize things and see." Every change needed an evidence trail: load data justifying the new size, a rollback plan, and a change window signed off in advance. That constraint shaped the whole engagement — read-only analysis first, then changes in ascending order of blast radius, cheapest-to-reverse first. It's also why the result line says zero production incidents: not luck, sequencing.
What we did, week by week
The ordering is the method: commitments came last, because a Reserved Instance bought against an oversized baseline locks the waste in for a year and pays you a discount on your own mistake. Right-size first, commit second — the same sequence we lay out in AWS cost optimization, ranked by what pays.
Where a 47% reduction typically comes from
In engagements of this shape — a payments-grade AWS estate that grew fast and was never pruned — the savings split lands in a recognizable pattern. Right-sizing and Graviton migration on the database tier is usually the single biggest slice, because databases are sized once, under deadline, from optimistic projections, and then nobody wants to touch them. Environment teardown is the second slice, and the one finance never believes until they see it: non-production estates commonly bill a third of production's rate for workloads that are used forty hours a week. NAT and egress redesign is the quiet third — a pure architecture fix with zero performance trade-off. Commitments come last and smallest, precisely because they're bought after the baseline shrank; buying them first would have made every other lever less valuable.
Just as important is what we deliberately didn't do. No Spot instances anywhere near the payment path — interruption risk has no place in an authorization flow. No multi-year commitments on a baseline that was still moving. No cost cuts to the observability stack in the middle of a change program, which is the classic way teams save $2K a month and then spend $200K on one blind incident. And nothing was resized on projection alone: every change carried observed load data, because "it should be fine" is not a change-control artifact a regulated FinTech can file.
The 6+ month sustainability is the part most cost projects skip. One-off cleanups decay — new environments accumulate, tags rot, the next quarter's hires re-create the same waste. The two mechanisms that hold the line are boring by design: the tagging policy is enforced in CI (untagged infrastructure fails the pipeline, not a spreadsheet review), and the teardown automation removes the human decision from environment hygiene entirely. Cost review became a 30-minute monthly ritual the client's platform team runs from their own dashboards.
The results, six months later
- 47% monthly AWS bill reduction, still holding 6+ months after our exit — the tagging policy and teardown automation keep the two fastest-regrowing waste classes from returning
- $312K annualized savings across compute, storage and egress
- Zero production incidents across every resize, migration and endpoint change
- A cost practice, not just a cheaper bill: the client's platform team runs the monthly review with their own dashboards; no retainer dependency
Engagement type: Project Delivery · 6 weeks · fixed-scope SoW. What a scoped audit like this covers — and what "free audits" usually don't — is documented in what a cloud bill audit actually covers.
A note on this write-up: this is a representative engagement — the shape, sequencing, controls and outcome metrics are exactly how InfraZen runs this work, published this way while client-cleared named case studies are prepared. Where competitors show logo walls, we currently show mechanics; we think the mechanics are the more useful half.
Related: Cloud billing & FinOps consulting · AWS cost levers, ranked · Reserved Instance calculator · Next case: 97.1→99.98% uptime