← Kespa Consulting case studies

Case study · cloud costs

Cutting a large AWS bill by over 30%, after a year of trying

A major domain registrar and hosting provider had spent a year on cost tools and internal reviews. They saved perhaps 5%, and the growing bill soon swallowed it. Working inside its systems team, we have cut the bill by over 30% so far, by understanding the platform as a whole and rethinking how it was built.

The situation

The company runs its registrar applications and the hosting for thousands of customers on AWS, and the bill was considerable. It had already invested heavily in bringing it down: expensive external cost-optimisation tools, and a lot of its own systems team's time.

Both found the low-hanging fruit: moving storage volumes from gp2 to gp3, right-sizing instances, moving to more modern instance types (complicated by a lot of legacy systems), and the other changes a tool, or AWS's own built-in advice, will point out.

Together those efforts saved perhaps 5%, and the bill kept growing, so even that was soon swallowed.

How we approached it

We started by understanding the existing infrastructure completely: every system and application, and what each one actually does. We mapped the whole estate, then re-evaluated each part against what the business needs it to do, and worked out the architecture really needed to provide that.

Much of the work was component by component. We looked at individual parts of the platform and found alternative ways of implementing them. Often the alternative was more resilient and more scalable, as well as much cheaper to run.

One small example: customer hosting data

This is just one small part of the work, but it shows how the approach works in practice.

Storing customer hosting data on the company's legacy platform was a huge cost. We didn't just look at what the storage cost and search for something cheaper. We needed to understand how the data was used: who accesses it, how often, and how quickly they need it.

We looked at how IOPS and throughput were provisioned and what they cost, and at the free space every file system had to keep in reserve, across many servers. A tool can tell you your drives are bigger than the data on them, or that your IOPS and throughput are higher than your metrics suggest. What it can't tell you, without understanding your systems, is why those decisions were made.

Understanding how the hosting data was used let us see it as distinct kinds of data. Some is used all the time: mailboxes in use, websites people visit. Much of it is almost never touched: old spam, sites on expired domains that nobody can reach any more, media files and backups someone uploaded by FTP years ago and forgot about.

That pointed to Amazon EFS, network storage that automatically separates recently used data from rarely used data and charges a much lower rate for the rarely used part.

To be sure it was viable, we had to know what proportion of the data would sit in the much cheaper infrequent-access tier. We worked that out by examining web and mail traffic patterns, and identifying hosting packages that were unused or rarely used.

EFS is the wrong answer for most problems like this. Its performance grows with the amount of data stored, so on a typical volume, metadata-heavy work such as scanning a mailbox or listing the files in a web directory becomes very slow. Knowing the platform in detail is what showed it would work here.

The result wasn't only a large saving on storage volumes and their high IOPS and throughput charges. The company could retire its bulky storage servers altogether, cutting maintenance effort as well as cost, and was left with a far more scalable and resilient hosting platform.

The result

The AWS bill is already down by more than 30%, and our target is to take the saving past 40%. The platform is more resilient and scalable than before, not just cheaper.

How long it takes

There's no fixed timescale. Results are directly proportionate to the level of engagement. Here, cost was never the main job. We have been part of the company's systems team since 2023, and the savings came over about two years alongside everything else that team does.

With an engagement focused on cost, results like these typically take six to twelve months. Where the cloud bill is large enough to justify the work, the savings usually repay its cost within six months. For smaller bills, we'll tell you up front if the numbers don't add up.

More than savings

The process itself paid off. Mapping everything exposed architectural weaknesses, single points of failure and gaps in disaster recovery, which could then be fixed.

It also produced something the company didn't have before: one complete understanding of its systems as a whole. Until then, each team understood its own part of the platform, but nobody had the full picture.

Why it worked

Tools see usage metrics; they can't see what a system is for. The real savings came from understanding what each part of the platform does and why, and that understanding came from working inside the team that runs it. It's why we approach cost as part of operating and designing a system rather than as a one-off exercise.

Saving so far
30%+
Target
40%+
Typical timescale
6 to 12 months of focused work
Return
Savings found are usually a multiple of the time invested
Before us
About 5% saved by tools and internal effort, then lost to growth
Also found
Single points of failure, disaster recovery gaps, architectural weaknesses
Client
A major domain registrar and hosting provider
Engagement
Embedded in the systems team, ongoing
Practice
Kespa Consulting

Discretion by default

We never name our clients: not on this site, not in sales conversations, not on social media. Your use of Kespa stays between us unless you choose otherwise. That's why you'll find descriptions here instead of logos.

Talk to us

Is your cloud bill still higher than it should be?

We reply within one working day, and we'll tell you quickly if we're not the right fit.