$378,000 a Year for Databases Nobody Was Using

A US FinTech was spending almost $3 million a year on AWS. Not unusual, not especially alarming - just the number they'd grown used to.

In my first week looking at the account, I found a handful of RDS databases that had no reason to exist. They'd been built for a client of this company - a customer that had since cancelled their service. The business relationship ended. But the databases kept running. And the former customer kept hammering them with traffic they weren't supposed to be handling anymore.

Nobody turned them off because nobody noticed them.

That's $30,000 a month. $378,000 a year. Running in the background, serving a customer who no longer paid for the service, billing to a company whose engineers had long since moved on to other problems.

The engineers at this company are smart. This isn't a knowledge problem. Everyone on the team knows how to decommission a database. It's not a skill issue or a tooling gap. They know what RDS is, they know what it costs, they know how to shut one down.

It's an attention problem.

The same week, same account: a pile of EBS volumes with no servers attached. Most of them weren't even formatted - they'd literally never been used. Just allocated, never mounted, never written to, billing quietly every month. Another $27,000 a year.

This is the thing about cost leaks that makes them different from other engineering problems. They don't set off alarms. They don't cause outages. They don't show up in sprint retrospectives. They just sit there, in the monthly bill, in a line item nobody looks at closely enough to question.

I spent three years on the AWS EC2 team helping customers find exactly these kinds of problems before going independent. I know where things hide. And most of the time, the things I find aren't sophisticated. They're just invisible.

The zombie databases were decommissioned shortly after I flagged them. The unattached volumes were deleted. That's $405,000 a year in cost leaks plugged in the first week.

No architectural changes. No code deploys. No migration plans. Just someone paying attention to what was already there - and what shouldn't have been.

I built a tool for the unattached volumes problem. It spins up EC2 instances in each Availability Zone, mounts the orphaned volumes, and inspects them. The vast majority - some over 1TB each - weren't even formatted. The remainder were empty or contained only OS files. Nothing worth keeping. Just storage, allocated and abandoned.

Nobody's job is to look at what's already running and ask whether it should still be running. That question never makes it into a sprint. It never gets prioritized over new features. It just sits there, unanswered, at $30,000 a month.