Every few months a business owner sends me a screenshot of their AWS bill with the same message: "This keeps going up and nobody can tell me why." The application is not growing that fast. Traffic is roughly flat. Yet the number climbs every month like it has a mind of its own.

This cloud cost optimization guide is for that owner. Not for companies running thousands of servers with a dedicated finance-and-ops team, but for the ordinary Indonesian business whose monthly cloud bill quietly grew from Rp 8 million to Rp 25 million while everyone was busy shipping features.

The reassuring part: in almost every bill I audit, at least 30 percent of the spend is pure waste. Not clever optimization opportunities, just money burning on things nobody uses. Finding it does not require a specialist. It requires 30 focused minutes a month and a willingness to ask uncomfortable questions.

Zombie Resources: The First Thing to Kill

The biggest source of waste is not inefficiency. It is abandonment. Things get spun up and never turned off.

Go looking for these first, because they are the fastest wins:

  • Forgotten test and staging environments. A developer created a full copy of production to test something in November, and it has been running 24 hours a day ever since. This single item is often the largest line on an SME bill.
  • Unattached storage volumes. You deleted a server but the disk that was attached to it lives on, billing you monthly for data you will never read.
  • Old snapshots and backups. Hundreds of automated backups from 2021 that no policy ever cleaned up.
  • Idle load balancers and unused static IPs. Each one small, all of them adding up, none of them doing anything.

None of this shows up as a scary alert. It just sits in the bill, invisible, because nobody owns the job of looking. Killing zombie resources alone often recovers 15 to 20 percent immediately.

Right-Sizing: Stop Paying for Headroom You Never Use

The second layer is over-provisioning. When teams set up infrastructure, they guess at the size they need, guess high to be safe, and never revisit the guess.

The classic offender is the database. Someone picked a large instance "to be safe," and it runs all day at 8 percent CPU. You are paying for a machine four times bigger than your workload needs. The same happens with application servers sized for a traffic spike that comes once a year, or never.

Right-sizing is not glamorous, but it is reliable money:

  • Look at actual CPU and memory usage over the last month, not the peak, the average.
  • If a server sits under 20 percent utilization consistently, it is a candidate to drop one size.
  • For workloads that are predictable and always-on, committed pricing (reserved or savings plans) cuts the rate by a large margin for a one to three year commitment.

Do this carefully and measure after each change, but the pattern is almost always the same: you were paying for headroom you never touched.

The Question Almost Nobody Asks

Here is the uncomfortable one. For a surprising number of SMEs, the real problem is not an inefficient cloud setup. It is that they are on a complex cloud architecture at all.

Somewhere along the way, a developer built the system the way big-company blog posts describe: managed Kubernetes, multiple services, autoscaling groups, a managed database cluster. Impressive, and completely oversized for an application serving a few thousand users a day.

So the question to put on the table honestly: would a single USD 40 per month virtual private server actually do the job here?

For a lot of internal tools, company websites, and modest applications, the honest answer is yes. One well-configured VPS, or a small managed platform, can serve the load with a fraction of the cost and a fraction of the operational complexity. The elaborate architecture was solving a scale problem the business does not have. This is the infrastructure cousin of the trap I describe in Vendor Lock-In: Questions to Ask Before You Sign, where the fancy option quietly costs more than the plain one that fits.

I am not saying rip out your cloud tomorrow. I am saying the architecture should match the business, not a diagram from a company a thousand times your size.

Make It a 30-Minute Monthly Ritual

Waste comes back the moment you stop watching, so build a light habit instead of a one-time cleanup. Once a month, block 30 minutes and walk this checklist:

  1. Open the cost breakdown by service. What went up since last month, and why?
  2. Hunt zombies. Any running environment nobody used this month? Any unattached storage or stale snapshots? Kill or schedule them.
  3. Check the top three most expensive resources. Are they right-sized against real average usage?
  4. Set a billing alert. A simple threshold that emails you when the bill crosses an expected number, so a surprise becomes a same-week problem, not an end-of-month shock.
  5. Ask the big question once a quarter. Does the architecture still match the actual load, or has it outgrown the business in the wrong direction?

Thirty minutes a month is cheaper than any consultant and catches most of the damage before it compounds.

The Takeaway

A runaway cloud bill is rarely a mystery once you look. This cloud cost optimization guide comes down to three moves: kill the zombie resources nobody remembers, right-size the servers you over-provisioned out of caution, and honestly ask whether your architecture is far larger than your business needs. Most SMEs are carrying at least 30 percent waste, and a disciplined 30-minute monthly review recovers it without a specialist. The cloud does not overcharge you. It simply keeps billing for every decision you forgot to revisit, so make revisiting the habit.