- Market
- Saudi Arabia
- Written for
- COO, CTO
- Decision
- Understand why the cloud bill keeps rising and what actually brings it down
- Languages
- EN · AR
Where the money goes
The five line items that quietly grow an AWS bill
Almost every runaway bill we open is made of the same five things. None of them is dramatic; each grows a little every month, and together they compound because nobody is looking.
| Variable | Effect on price | Why |
|---|---|---|
| Idle and unused capacity | Paid for around the clock, used a fraction of it | Instances that run at low utilisation all day, volumes detached from anything, load balancers in front of nothing, addresses reserved and forgotten. The cloud bills for what is provisioned, not for what is used, and the difference is pure waste. |
| Over-provisioned instances | Sized for a peak that never comes, or by guesswork at launch | Sizing decisions get made once, under uncertainty, and never revisited. A service that needed headroom for launch keeps that headroom for years, because reducing it feels risky and nobody is measured on it. |
| Data transfer surprises | Traffic priced by path, growing with usage nobody maps | Moving data between zones, regions and out to the internet each carries a different price, and an architecture drawn without those paths in mind pays for them invisibly. Cross-zone chatter between services is the classic case: correct, working, and expensive. |
| Forgotten environments | Test, staging and demo estates that outlived their project | An environment spun up for a proof of concept, a migration or a departed contractor keeps billing after the reason for it is gone. Nothing marks it as disposable, so nothing disposes of it. |
| Spend without an owner | The driver behind the other four | When the bill arrives as one number to one budget, no engineer sees the cost of their own decisions, and no team answers for its share. Everything above persists because of this one: waste that nobody owns is waste that nobody removes. |
Ask any supplier quoting this
- Which teams and workloads does each line of our bill belong to, and can you show the attribution rather than assert it?
- What utilisation data is the rightsizing recommendation based on, and over how long a window was it measured?
The real work
Rightsizing is engineering work, not a spreadsheet exercise
The tempting version of cost control is a spreadsheet: list the instances, sort by price, cut the big ones. It fails because the spreadsheet cannot see what an instance is for. The correct version starts from measurement — utilisation over weeks, not a screenshot of one afternoon — and treats every downsizing as a change to a production system: made deliberately, observed afterwards, reversible if the measurement was wrong. Some of the biggest reductions are not size changes at all but architectural ones — a workload that only runs during business hours scheduled to stop at night, a batch job moved to capacity the provider sells cheaply because it can be interrupted.
This is why handing the bill to finance alone does not work. Finance can see that the number is too big; only engineering can see which parts of it are load-bearing. The productive arrangement is both together: finance sets the pressure and owns the baseline, engineering makes the changes and stands behind their safety, and a shared report keeps each honest about the other’s half.
The path down
From one opaque invoice to spend somebody answers for
The order matters. Committing to capacity before measuring it locks the waste in; cutting before attributing breaks something and discredits the whole effort.
Tag and attribute
Every resource gets an owner, an environment and a workload tag, enforced going forward so the untagged pile stops growing. The bill becomes a set of smaller bills each team can recognise as its own. A bill broken down by team and workload.
Show each team its share
Showback puts each team’s spend in front of it monthly, beside the trend. No enforcement yet — visibility alone changes behaviour, because engineers fix waste they can see and are named against. Monthly spend per team, trending and visible.
Rightsize by measurement
With utilisation measured over a real window, the estate is resized workload by workload: idle resources retired, over-provisioned ones stepped down, schedulable ones scheduled — each change observed and reversible. An estate sized to what it actually does.
Commit where usage is steady
Only now does reserved or committed capacity make sense — as a category, it trades flexibility for a lower rate, so it belongs on the baseline that rightsizing has proven steady, never on the estate as it was before the clean-up. Commitments on the proven baseline only.
Our role
What we change in the estate and what your teams keep owning
- 01
The diagnosis first
We break the bill down into the five drivers, attribute each to a workload and a team, and put the result in front of you before proposing any change — because a cut made without attribution is a gamble with production.
- 02
Changes made as engineering
Rightsizing, scheduling and architectural cost fixes are made as production changes — measured before, observed after, reversible throughout — so the bill comes down without the estate becoming fragile.
- 03
Ownership that outlasts us
We leave the tagging discipline, the showback report and the guardrails that stop untagged or oversized resources appearing again, because a one-off clean-up without ownership is a bill that grows back.
How much pressure to put on cost against speed of delivery is your call; we give you a bill you can read, an estate sized by evidence, and the ownership structure that keeps both true after we leave.
Common questions
What buyers ask
Why does our AWS bill keep growing when our traffic has not?+
Because most of a growing bill is not traffic. It is provisioned capacity, and provisioned capacity grows through decisions, not usage: a new environment for a project, an instance sized generously to be safe, a proof of concept nobody tore down, replication and transfer paths added along the way. Each is small and reasonable at the time; none is ever revisited, because the bill arrives as one number that no single team answers for. That is why the first fix is attribution rather than cutting — until each line has an owner, the bill grows by default and shrinks only by campaign.
When does reserved capacity make sense, and when is it a trap?+
Reserved and committed capacity, as a category, trades flexibility for a lower rate: you promise the provider a baseline of usage and pay less for it. That is a good trade on workloads whose demand is genuinely steady and already rightsized — the always-on core of the estate after the clean-up. It becomes a trap in two ways: committing before rightsizing, which locks today’s waste in at a discount, and committing to shapes the estate is about to move away from, which leaves you paying for capacity the architecture no longer uses. The sequencing rule is simple — measure, rightsize, and only then commit to the baseline that survived.
Can finance cut the cloud bill without engineering being involved?+
Only at the edges, and with risk. Finance can negotiate rates, question line items and set budgets, and all of that is worth doing. But the substance of a cloud bill is engineering decisions — instance sizes, environments, data paths, schedules — and only the people who understand what each resource does can change it safely. A cost programme run by finance alone either stalls at the negotiation layer or forces cuts that take something down and discredit the effort. The arrangement that works pairs them: finance owns the target and the baseline, engineering owns the changes, and a shared monthly view of spend per team keeps the pressure where the decisions are made.
In this cluster
Diagnose it with us
A cost review that names which of the five drivers your bill is made of and who should own each, so the next month is measured against a baseline rather than guessed at.
Start the conversation
