Where the AWS bill leaks, and how to plug it
A Noida SaaS company found ₹2.1 lakh a month going to servers nobody had logged into since February. The account was not hacked. It was just never cleaned up.
I got called into a finance review at a 120-person SaaS firm in Noida. The CFO, Meera, had one question. Why was the AWS invoice up 40 percent on a headcount that had barely moved. Their CTO, Rohan, thought the product was just busier. The numbers said something duller and more fixable.
We put Cost Explorer on the screen, grouped by service, then by usage type. That single view ended the argument in about ten minutes. The growth was not traffic. It was stuff left running.
Where the AWS bill leaks first
Cloud waste is rarely one big mistake. It is a hundred small ones that compound while everyone is busy shipping. Here is what we found, in the order it hurt.
Idle dev and staging instances, running 24 by 7. Their three test environments cost the same overnight and on weekends as they did at noon on a Tuesday. Nobody turns a physical server off, so nobody thought to turn the cloud one off either. Old habit, new bill.
An oversized database. The main RDS instance was a db.r5.2xlarge picked during launch month, when everyone over-provisions out of fear. CPU had not crossed 18 percent in ninety days. They were paying for a hall and using a meeting room.
Forgotten storage. A 2023 project left 4 TB of EBS snapshots and an S3 bucket full of logs with no lifecycle rule. Storage is cheap per GB, which is exactly why it grows quietly until it is not cheap at all.
Data transfer and NAT gateways. The line item Meera could not read at all. Cross-AZ chatter and a busy NAT gateway were adding close to ₹35,000 a month, invisible until you group the bill by usage type.
| Leak | What it looked like | Roughly per month |
|---|---|---|
| Idle dev/staging running 24×7 | 3 environments never paused | ₹68,000 |
| Oversized RDS | db.r5.2xlarge at 18% CPU | ₹71,000 |
| Orphaned EBS snapshots + S3 logs | 4 TB, no lifecycle rule | ₹26,000 |
| NAT gateway + cross-AZ transfer | unread usage-type line | ₹35,000 |
Indicative figures from their account, not a price list. Yours will sit in different boxes. The pattern holds though. In the reviews we run, roughly a third of an untuned bill is going to things nobody would defend out loud.
Plugging the holes, in order of effort
We did the cheap, reversible things first. No architecture changes, no risk to production.
The dev environments got a schedule. Instance Scheduler turns them off at 9 pm and back on at 9 am on weekdays, off entirely on Sunday. That alone took 65 percent off their non-production compute. A two-hour job.
The database we right-sized after watching AWS Compute Optimizer for a week. It recommended a db.r5.large, two sizes down. We changed it during a Sunday maintenance window. Nothing broke. The graph stayed flat and the cost halved.
Storage got lifecycle rules. Snapshots older than 60 days roll off automatically now. S3 logs move to a cheaper tier after 30 days and delete after a year. Set once, forgotten safely.
Then the part that needed Meera in the room. Their steady production baseline, the compute that genuinely runs every hour of every day, went onto a one-year Savings Plan. You commit to a baseline spend, AWS discounts it. For workloads that are not going anywhere, paying on-demand prices is just a tax on indecision.
The guardrail that stops the next leak
Finding the waste once is satisfying. Useless, though, if the account drifts back in six months. So the last hour was the most important one.
We set AWS Budgets with a hard monthly number and an alert at 80 percent of it. The day spend trends over, Meera and Rohan both get an email, not a quarter-end surprise. We turned on AWS Trusted Advisor cost checks so idle resources get flagged on their own. And we tagged everything by environment and team, so the next time the bill moves, you can see whose it is in one click.
That is the whole jugaad of cloud cost. Not a clever trick. Just visibility, a schedule, and someone who gets an email before the card does.
What it added up to
Their bill went from about ₹6.2 lakh a month to ₹4.1 lakh, and the product ran exactly as before. The migration question that started the meeting, whether to leave AWS for something cheaper, quietly answered itself. AWS was never the problem. The way it was run was.
If you want the same pass on your own account, we do a read-only review from our office in Vashi, Navi Mumbai. We look, we tell you where the money goes, you decide what to fix. The full version of this story sits in our surprise dollar-bill writeup, and if your invoice is still in USD, start with AISPL INR billing so finance can read it at all. The bigger picture lives on our AWS in India page. If part of your estate is still on-prem, the same math we ran on the cloud applies to a server consolidation too.
We have done this for 200+ Indian businesses out of Vashi, Navi Mumbai. Free review, no card, no contract, no sales call. Reach us on WhatsApp at +91 91375 93228, 10 to 7 IST, or start below.
P.S. Sudeep here. We ran this exact cleanup for a Noida SaaS team last quarter. The CFO asked the same thing you might be asking right now, which is whether the bill going up means the business is growing. Sometimes it does. Often it just means three test servers have been awake since February. Bas, switch them off and look at the rest with fresh eyes. We will do the looking for free if you want a second pair.
Key takeaways
- Group your bill by service, then by usage type. The leaks show themselves.
- Schedule non-production compute to switch off nights and weekends. Biggest fast win.
- Right-size with Compute Optimizer data, not a launch-month guess.
- Put a budget and an 80 percent alert on the account before you do anything else.
FAQ
How much can a cost review actually save on an AWS bill?
It depends on how tuned the account already is. A first serious review of an untuned estate commonly finds 25 to 35 percent in idle and oversized resources. A well-run account might only have a few percent left. The only way to know your number is to look.
Will right-sizing or scheduling break my production setup?
Scheduling only touches what you tell it to, usually dev and staging, never production. Right-sizing a production database is done in a maintenance window with the change reversible in minutes. We watch a week of real usage before recommending any size change.
Is the review really free, and what do you get out of it?
Yes, the read-only review is free. We do it because most teams who see where their money goes ask us to help run the account afterwards. If they do not, they keep the findings. No card and no contract to get the report.
