Cloud bills almost never spiral out of control overnight. They creep up little by little, month after month, until one day someone in finance asks why you’re paying the equivalent of a part-time salary. And the honest answer is usually uncomfortable: we don’t know exactly. That’s where the real problem starts, and it isn’t the cloud itself, but the lack of control over what gets switched on, what gets forgotten and what nobody ever switches off.
The good news is that cutting cloud costs doesn’t mean sacrificing power or accepting that your website will run slower. Almost all the savings sit in money you’re already throwing away without anyone noticing: oversized machines, test environments left running all weekend, orphaned disks from servers you deleted six months ago. This is what’s known as FinOps, and applied sensibly to a small business it hands you back between 20% and 40% of the bill without touching performance. Let’s break it down.
Why the cloud bill spirals out of control
The cloud is sold with a perfect promise: you only pay for what you use. The small print says something else: you pay for what you have switched on, whether you use it or not. And switching on is trivial —two clicks— while switching off requires someone to remember, to know that resource is theirs and to have the time. In a small business without a clear cloud owner, that “someone” doesn’t exist, so nothing ever gets switched off.
On top of that comes the “just in case” culture. A developer spins up a machine bigger than necessary because they don’t want to fall short. More storage than you’ll need in two years gets bought. Environments are duplicated for a demo and left there. Each decision, on its own, looks reasonable. Added together, they’re a constant leak that the bill hides behind a single global figure nobody breaks down.
The third factor is the complexity of the pricing model. AWS, Azure and Google Cloud have catalogues with thousands of items, charges for data transfer, and rates that vary by region. It’s almost impossible for anyone without dedicated time to understand where the money goes. And what isn’t measured can’t be controlled. That’s why the first step in any serious optimisation isn’t switching anything off, but actually seeing the bill: by service, by project and by owner.
Switched-off, oversized and forgotten resources
This is where the bulk of the quick savings lives, the kind that requires no redesign at all. There are three types of waste that show up in almost every account we audit, and they’re fixed in days, not months.
- Orphaned resources: disks that outlived the machine you deleted, unused static IPs, backups of servers that no longer exist, load balancers with nothing behind them. They serve no one and you pay for them in full every month.
- Oversized resources: the 16 GB RAM machine that always runs at 8%. The database engine provisioned for peaks that never arrive. Adjusting to the real size (rightsizing) usually cuts that resource’s cost in half without the user noticing any difference.
- Resources running unused: development, test and staging environments running 24/7 when they’re only used during office hours. Switching them off at night and on weekends removes 70% of their bill in one go.
That last point is the easiest win and the one most people ignore. A test environment only needs to be up 40 hours a week, not 168. Scheduling an automatic shutdown outside working hours is half an hour of work and saves money every week for years. The same goes for lifecycle policies: moving old data to cheaper tiers and automatically deleting whatever expires.
The prerequisite for all of this is tagging. If every resource carries its project, its environment and its owner, you know what you can switch off without fear. Without tags, every cleanup is Russian roulette where the prize is taking down production. It’s what turns optimisation into something repeatable instead of a one-off feat of heroics.
Reservations, autoscaling and well-tuned licences
Once you’ve cleaned up what’s surplus, it’s time to optimise what you actually need. And here the most expensive mistake is paying full list price for workloads that are perfectly predictable. If you know a server will be up all year long —and your production database is— paying for it on demand is throwing money away.
Reservations and savings plans for the stable stuff
Reserved instances and savings plans give discounts of between 30% and 70% in exchange for committing to a certain usage for one or three years. For the foundation of your infrastructure —the part you know will still be there— it’s free money you’re turning down. The key is to reserve only what’s genuinely stable and leave the rest on demand; over-committing ties you to resources you might stop using.
Autoscaling for the variable stuff
For workloads that rise and fall —your website during a campaign, a process that spikes at month-end— the answer isn’t a huge machine just in case, but autoscaling: letting the infrastructure grow when needed and shrink once the peak passes. That way you pay for high capacity only during the hours you use it, and performance doesn’t suffer because the system reacts before the user notices. This dismantles the false dilemma between saving and performing: set up properly, you do both at once.
Licences you pay for twice
And then there’s the silent chapter: licences. Windows Server, SQL Server or commercial databases billed inside the cloud machine when you may already own licences you could bring, or when an open source alternative would do the same job for free. Reviewing this once a year recovers surprising amounts, because it’s money paid on autopilot.
How to keep an eye on it every month without going mad
Optimisation isn’t a one-off project. If you abandon it, within six months the bill inflates again because the team keeps creating resources and nobody reviews them. But that doesn’t mean living glued to dashboards: it means setting up three or four automatic controls that work for you and only alert you when something drifts.
- Budgets with alerts: you set a monthly cap per project and get a warning when projected spend is about to exceed it. You find out mid-month, not when the bill arrives.
- Anomaly detection: the three big providers alert you when spend spikes in an unusual way. A runaway process burning resources can cost hundreds of euros over a weekend if no one spots it.
- A 30-minute monthly review: a look at the breakdown by service compared with the previous month. What went up and why? With tagging done properly, the answer is one click away.
- Recommendations report: the platforms suggest what to rightsize and what to reserve. Don’t accept it blindly, but it’s your starting point every month.
The key is to lean this oversight on your usual monitoring, so you can cross-reference cost with real performance. There’s no point trimming a machine if it degrades the service: the goal is to spend just enough for the performance your business needs. Cost and performance are always looked at together.
The cloud isn’t expensive. What’s expensive is the cloud nobody watches.
How MagicBoxDesk helps you optimise your cloud
At MagicBoxDesk this is exactly what we do: we audit your cloud account thoroughly and show you, in euros, where the money goes and how much we can recover without touching your performance. First the quick cleanup, then the fine-tuning of reservations, autoscaling and licences, and finally we leave the monitoring system in place so the bill doesn’t spiral out of control again. All part of our managed cloud infrastructure services, with support across Spain and on-site visits to your offices when needed.
The best part is that cloud optimisation almost always pays for itself: the first month’s savings usually cover the work, and from there it’s margin for your business. We take care of everything so you can focus on what you do best. Ask for a no-obligation quote and, with your real bill in front of us, we’ll tell you how much you can genuinely save.
