NordShop, an Oslo-based e-commerce platform on AWS, was overspending by roughly 40%. In 90 days RioCloud rightsized EC2, switched steady workloads to Savings Plans, fixed an over-provisioned RDS cluster, rebuilt the CI/CD pipeline, and added a FinOps review loop. Result: 40% lower monthly bill, deploys down from 45 to 12 minutes, zero downtime.
What was the situation at NordShop before the audit?
NordShop is a Norwegian direct-to-consumer e-commerce platform serving Nordic shoppers across Norway, Sweden and Denmark. By late 2024 the engineering team had scaled the AWS estate organically for three years: every new feature shipped with a fresh service, every Black Friday spike justified one more reserved buffer, and nothing was ever switched off. The monthly AWS invoice had crossed a number their CFO was no longer willing to sign without a fight.
When NordShop's CTO reached out to RioCloud Solutions in early 2025, they were not asking for a generic cost review. They wanted a fixed-scope, 90-day engagement with a hard target: cut the bill at least 30%, do not break checkout, and leave the team a playbook they could run themselves afterwards. We agreed to a 40% target instead, because the early data already pointed there.
The pre-audit baseline looked like this:
- ~64% of EC2 spend on on-demand instances, no Savings Plans, two stale 3-year RIs from 2022.
- A production RDS Aurora cluster running two r6g.4xlarge writers — one of them effectively idle.
- CI/CD on a self-hosted Jenkins fleet of six m5.xlarge runners, average utilisation 11%.
- Deploy time from merged PR to production: 45 minutes.
- An S3 estate of 47 TB with no lifecycle policy — every product image since 2021 in Standard.
- No tagging discipline; 38% of cost untagged in Cost Explorer.
How did we audit the AWS account in week one?
The first five working days were pure discovery. We did not touch a single resource. The goal was a defensible map of where every dollar was going, and a ranked list of waste by severity. Cost work without that map fails — you cut something visible, miss the real bleed, and lose the team's trust.
The week-one audit covered six layers in this order:
- Cost Explorer + CUR export to Athena — 13 months of line-item data into a queryable warehouse. This is the only honest source of truth; the AWS console aggregates lie about reality.
- Compute Optimizer for EC2, EBS, Lambda and RDS rightsizing recommendations, exported and cross-referenced against actual CloudWatch metrics for 14 days.
- Trusted Advisor sweep for idle load balancers, unattached EIPs, orphaned snapshots and underutilised RIs.
- Tag coverage report per service, per account — anything below 90% tag coverage was flagged for remediation before we cut spend on it.
- Architecture interviews with the four engineers who knew the system — what is safe to touch, what looks unused but is critical, what runs only at month-end.
- FinOps maturity scoring using the FinOps Foundation framework — NordShop scored "Crawl" on every capability, which is normal and fine.
By Friday of week one we had a 22-page audit report with a prioritised savings backlog worth an estimated 41% of the monthly run-rate. Roughly half was no-risk (delete this, switch that commitment), the other half required code or architecture changes.
What changes actually moved the AWS bill?
Cost optimisation reads like dozens of tiny knobs in a blog post. In practice, two or three decisions usually account for most of the savings. At NordShop the impact concentrated in five moves. The order matters — we did the reversible, no-risk work first, banked the savings, then earned the trust to do the harder architectural changes.
| Change | Risk | Monthly impact | Week shipped |
|---|---|---|---|
| Delete idle load balancers, EIPs, snapshots, dev EC2 left running | None | ~4% | Week 2 |
| 3-year Compute Savings Plan for steady baseline | Low | ~14% | Week 3 |
| RDS Aurora — drop second writer, rightsize reader, switch to Graviton | Medium | ~9% | Week 5 |
| S3 Intelligent-Tiering + lifecycle to Glacier IR for legacy product imagery | Low | ~5% | Week 6 |
| CI/CD rebuild — Jenkins to GitHub Actions on ephemeral runners with layer caching | Medium | ~8% | Week 9 |
Total: 40% reduction in monthly AWS spend by day 90, sustained for the following two quarters. The cumulative annualised saving is enough to fund two senior engineering hires.
Why did the deploy time drop from 45 to 12 minutes?
The CI/CD rebuild deserves its own section because it was the change the engineers cared about most. AWS bill reduction is a CFO win; faster deploys is a quality-of-life win that the team feels every working day.
The old Jenkins setup had three problems compounding: cold runners that took four minutes to boot, no Docker layer cache so every build pulled npm and Composer from scratch, and a sequential test stage that ran integration tests against a single shared staging database. The fix was structural, not incremental.
- Moved to GitHub Actions with ephemeral self-hosted runners on EC2 spot — saves money, scales horizontally, no idle fleet.
- Added a persistent Docker layer cache in S3 + BuildKit — node_modules and vendor builds went from 7 minutes to 40 seconds.
- Parallelised the test matrix into five shards, each against its own ephemeral RDS snapshot.
- Replaced the bespoke shell deploy script with a typed pipeline using OIDC for AWS auth — no more long-lived access keys.
- Added pipeline gates that fail fast: lint and unit tests in the first 90 seconds, integration after.
The same patterns apply to any team running an unloved CI pipeline. We cover the underlying playbook in our step-by-step AWS cost reduction guide.
What did the FinOps governance loop look like after launch?
The hardest part of cost optimisation is not the cuts. It is making sure the savings stay cut. Six months after most one-off audits, spend creeps back to within 10% of the starting baseline because nothing structurally changed in how the team makes decisions. We left NordShop with three lightweight rituals that stop the creep without slowing engineering down.
- Weekly anomaly digest — automated email every Monday flagging any service whose week-on-week spend moved more than 15%. Investigated, not acted on, by default.
- Monthly FinOps stand-up — 30 minutes, engineering plus finance, looking at one dashboard. The only agenda is "what surprised us and what do we do about it".
- Quarterly Savings Plan review — coverage and utilisation rates, plus a forward look at any planned architecture changes that will move the baseline.
NordShop's team now runs all three without us. That, more than the 40%, is what makes the engagement repeatable. We do the same handover on every cloud and DevSecOps engagement RioCloud takes on.
What does this case study mean for your AWS bill?
If your AWS estate looks anything like NordShop's — grown organically over a few years, no formal FinOps, deploys that take more than 20 minutes, no Savings Plans coverage you can quote off the top of your head — there is almost certainly a 25–40% saving sitting in the account, plus deploy-time wins you would not have to argue for. The work is well-understood, the risk is manageable, and the payback is measured in weeks, not quarters.
Our generic playbook for any team that wants to attempt this in-house is documented in how to reduce AWS costs by 40% without sacrificing performance. For an e-commerce-specific framework, see our e-commerce cloud cost audit guide. For broader platform decisions, our AWS vs GCP vs Azure comparison for 2026 covers when AWS is the right choice in the first place.
Frequently asked questions
- How long did the NordShop AWS optimisation take?
- 90 days end to end. Week 1 was audit only, weeks 2–9 delivered the five biggest cost changes, weeks 10–12 hardened the CI/CD rebuild and handed over the FinOps governance loop.
- What exactly added up to the 40% saving?
- Roughly: 14% from a 3-year Compute Savings Plan, 9% from RDS rightsizing and Graviton, 8% from the CI/CD rebuild on spot, 5% from S3 lifecycle, 4% from deleting idle resources. The exact split varies per account.
- Was there any downtime during the work?
- Zero customer-facing downtime. The RDS writer change was a blue-green failover during a maintenance window; the CI/CD migration ran the old and new pipelines in parallel for two weeks.
- Why a 3-year Savings Plan and not 1-year?
- NordShop's baseline compute had been stable for 18 months and the business roadmap pointed to growth, not contraction. The extra discount over a 1-year commitment was worth roughly 4% of the monthly bill.
- Can the same approach work on GCP or Azure?
- Yes. The names change (Committed Use Discounts on GCP, Reservations on Azure), but the framework — audit, rightsizing, commitment coverage, lifecycle policies, CI/CD efficiency, governance loop — is identical. RioCloud delivers the same engagement on all three.
- How much does an audit like this cost?
- The 90-day NordShop engagement was fixed-scope. For most mid-market AWS accounts we offer a paid 5-day audit that produces the same ranked backlog; if the projected savings do not cover the audit fee 5× over, we refund it.
- Do you work with companies outside the Nordics?
- Yes. RioCloud Solutions has delivered cloud and DevSecOps work across 12 countries including the UK, USA, UAE, Singapore, Germany and India. See recent client work.
Next steps
If your monthly AWS invoice has stopped making sense, the cheapest first move is a one-week diagnostic. Book a no-obligation AWS cost call with RioCloud and we will tell you within 48 hours whether there is a 25%+ saving in your account, what it would take to capture, and whether you should do it in-house or with us. While you wait, the companion read is our AWS cost reduction playbook and the Kubernetes for small teams guide if container spend is part of the story.
