Techbanq

Cloud cost · a working method

The bill grew 40% and the traffic did not

Almost every estate in this position has the same underlying issue: spend cannot be attributed to anything, so nobody can tell required capacity from waste. That is the first problem to solve, and it is not the one most cost exercises start with.

Serving Saudi Arabia from our Riyadh office and our engineering hubs in Lahore and Karachi

Market
Saudi Arabia
Written for
CTO, CIO, Head of Digital
Decision
Find out where cloud spend is going and what can safely be removed
Languages
EN · AR

Information security for cloud delivery

  • ISO

    ISO/IEC 27001

    ISO/IEC 27001 certified

Certifications & registrations

The problem

A bill you cannot attribute is a bill you cannot argue with

We do not have an office in Saudi Arabia

We serve organisations there from our Riyadh office and our engineering hubs in Lahore and Karachi, travelling for the parts of an engagement that need to happen in person.

The conversation usually starts in finance. Spend is up materially year on year, the business has not grown at the same rate, and the engineering answer is that everything running is needed. Both sides are arguing from position rather than from data, because the bill arrives organised by service — compute, storage, data transfer — and the business is organised by product and team. Nothing in the invoice maps to anything anyone is accountable for.

In the estates we look at, the share of spend that cannot be attributed to a team or a product is routinely the largest single finding — larger than any individual instance-sizing problem. Until that is fixed, every optimisation is a negotiation about whether a resource is needed, and the person who shouts loudest wins. After it is fixed, the conversation becomes arithmetic.

The second thing worth saying early: not every high bill is a problem. Some estates are running exactly the capacity the workload requires, and the growth reflects real demand. A review that cannot return that answer is not a review, it is a sales process. Roughly speaking, the useful outcome is a number for how much is genuinely recoverable — and sometimes that number is small.

Recognise the shape

Six things an overgrown bill usually turns out to be

Most estates have several of these at once. They are listed in roughly the order of how often they are the largest contributor.

  1. 01

    Environments that outlived their purpose

    A performance-test environment cloned from production for a launch two years ago and never removed. Non-production running at production scale, and running at 3am on a Friday. This is the cheapest category to fix and the most common single line item.

  2. 02

    Storage nobody has looked at

    Snapshots with no lifecycle policy, log buckets in standard storage class holding years of data, and orphaned volumes from instances terminated long ago. Individually small, cumulatively substantial, and almost entirely safe to remediate.

  3. 03

    Data transfer nobody designed

    Cross-AZ chatter between services that could be co-located, NAT gateway charges from workloads that should use a VPC endpoint, and egress from a pattern that made sense when the architecture was smaller. Often invisible because it is not attached to any resource anyone recognises.

  4. 04

    Capacity provisioned for a peak that never comes

    Instances sized from an estimate made before launch and never revisited against measured demand. The distinction that matters is between low average utilisation with real spikes — which needs elasticity, not downsizing — and low utilisation throughout, which is simply oversized.

  5. 05

    A database doing work the application should do

    Frequently the largest line and the least examined. An unindexed query pattern or an N+1 in a hot path drives instance size upward, and the response is usually to buy a bigger instance because that fixes it in an afternoon. The query is cheaper to fix and stays fixed.

  6. 06

    Commitments bought against the wrong shape

    Reserved capacity or savings plans purchased for an architecture that has since changed, or bought before the waste was removed — which locks in the waste for one to three years. Commitment strategy belongs at the end of a cost exercise, not the start, and this is the most expensive sequencing error in the list.

The diagnosis

The order matters more than the tooling

Running these out of order is how estates end up with a three-year commitment on capacity they were about to delete. Each stage answers a question the next one depends on.

STAGE 01

1 · Attribute the spend

Tag coverage first, then a cost-allocation model that also handles what cannot be tagged — shared clusters, data transfer, support charges. Nothing downstream is trustworthy until this is done. Spend by team and product, and an explicit unallocated figure.

STAGE 02

2 · Find the concentration

Cost is almost never evenly spread. Identify the small number of accounts, services and resources that carry most of the bill, and work only on those — optimising the long tail is effort spent for arithmetic that does not move. A ranked list of where the money is, with an owner against each entry.

STAGE 03

3 · Separate waste from required capacity

The judgement step, and the one that cannot be automated. Unused is not the same as underused, and underused is not the same as oversized — a system that idles for twenty hours and carries the month-end run is doing its job. Two lists: safe to remove, and required — with the reason recorded for each.

STAGE 04

4 · Inspect the architecture

Data-transfer paths, service placement, storage classes and lifecycle, and whether the compute model matches the workload shape. This is where the durable savings are, and where the change carries real risk. Architecture findings with the saving, the effort and the risk stated against each.

STAGE 05

5 · Inspect the data layer

Query patterns, indexes, connection handling and retention. Databases are usually the largest single service and the least likely to have been examined as a cost problem rather than a performance one. The queries and retention policies driving instance size, with a fix for each.

STAGE 06

6 · Remediate, measured

Changes in order of saving against risk, each with a stated performance expectation and a rollback. Anything that trades latency for cost goes to the service owner as a decision, not as an optimisation. A change log with the before-and-after cost and the performance effect.

STAGE 07

7 · Commit, then govern

Only now: commitment strategy sized to the demand that remains after remediation. Then the operating model — budgets, anomaly alerting, unit-cost reporting and cost as a standing item in architecture review. Commitment plan, budget alerts, and a unit-cost series the business recognises.

Scope

What a cost review tells you

Cost optimisation is sold with percentages more than almost any other engagement in this market. We do not quote one, and the reasons are worth stating plainly.

What we do

  • Produce a figure for spend that cannot currently be attributed to any team or product, which is usually the largest single finding.
  • Separate what is genuinely unused from what is required capacity, with the reasoning recorded so the classification can be challenged.
  • State the performance implication of every recommended change, including the ones that trade latency for cost.
  • Leave you the findings in a usable form whether or not the remediation is ours.

AWS publishes its own pricing, billing and cost-management documentation, including how data-transfer and commitment pricing work. Where a figure on this page matters to a decision, check it against that source for your region and account type rather than taking it from a consultancy page — pricing changes and regional availability differs.

Primary source

Common questions

What buyers ask

How much will we save?+

We will not put a number on it before looking, and a supplier who does is quoting a marketing figure. What the review returns is the actual arithmetic for your estate: how much spend is unattributed, how much is genuinely unused, how much is required capacity, and what each remediation is worth against the effort and risk of doing it. Some estates turn out to be running close to what the workload needs — that is a legitimate finding and we will report it rather than manufacture a programme.

Will this slow anything down?+

Some changes are free — deleting an unattached volume affects nothing. Others are a genuine trade: moving objects to a colder storage class lowers cost and raises retrieval time; reducing an instance size removes headroom. We state the trade against each recommendation and route anything that touches a live service to its owner as a decision rather than applying it as an optimisation. Cost work that quietly degrades a service produces a saving in one report and an incident in another.

We already use AWS Cost Explorer and a third-party tool. What is different?+

Those tools tell you what you spent and flag resources with low utilisation. They cannot tell you whether a resource is required, because that depends on business context they do not have — the quarterly reconciliation job, the standby, the retention obligation. The work that produces a defensible remediation list is the classification step, and it needs someone who has asked the service owners. Tooling is necessary and it is not the constraint.

Does any of this apply to Azure?+

The sequence does, almost entirely — attribute, concentrate, classify, then architecture, data, remediation and commitments. The specifics change: reservations and savings plans behave differently, the storage tiers are named differently, and the data-transfer model is not the same. The two mistakes that cost the most are identical on both platforms: buying commitments before removing waste, and treating low utilisation as proof that something is not needed.

Get a scoped cost review

A fixed-scope cost review with written findings: where spend concentrates, what is genuinely unused, what is required capacity, and a remediation list ordered by saving against effort and risk. You keep the findings regardless of who does the work.

Start the conversation