All insights
DevOpsSeptember 30, 2026

Cutting cloud spend without cutting reliability

Addicnet Cloud Practice· Cloud operations and optimisation· 7 min read
Cutting cloud spend without cutting reliability
#FinOps#Cost Optimisation#Cloud Operations

Cloud bills rarely grow because someone chose expensive services. They grow because nothing forces a review. A workload is sized for a launch week, the launch passes, and the sizing stays. A storage bucket accumulates years of logs nobody reads. A commitment is signed to capture a discount and then the usage profile changes.

Cost optimisation is a governance discipline with a technical component, not a procurement exercise.

Find the money before cutting anything

The first pass is measurement, not action. Four views answer most of the question:

  • Spend by service and account. Where the money actually is, not where it feels like it is.
  • Spend by tag or owner. Anything untagged is unattributed, and unattributed cost cannot be managed.
  • Utilisation against provision. CPU, memory, storage throughput and database connections over thirty days. Peaks matter, but so does the median.
  • Growth rate by line item. A small line growing twenty percent a month is a bigger problem than a large stable one.

Most organisations we assess find that a handful of line items account for the majority of avoidable spend: idle development environments, oversized databases, unattached storage volumes, and data that has never been lifecycle-managed.

The levers, in order of risk

Remove what is not in use. Unattached volumes, unused load balancers, orphaned snapshots, idle development environments outside working hours. This is the safest saving available and usually funds the rest of the programme.

Right-size against evidence. Not against a hunch, and not against a single peak. Use thirty days of utilisation data and size for the ninetieth percentile, then monitor after the change. Resize in steps, and keep the rollback option until the workload has run through a full business cycle.

Apply storage lifecycle policies. Logs, backups and analytics data have a natural temperature curve. Moving cold data automatically to infrequent-access tiers is a configuration change with a recurring return.

Use commitments for the stable baseline. Reserved capacity and savings plans are genuinely valuable for predictable workload, and genuinely expensive when they overshoot. Commit only to the floor of your usage, not the average, and only after right-sizing has settled.

Fix the architecture where it is cheaper to be correct. Serving the same object repeatedly through an application layer when a cache would do, keeping chatty cross-region traffic, or using managed services that duplicate capability you already pay for.

The cheapest instance is not always the cheapest outcome. A workload that fails under load costs more in an afternoon than it saves in a year.

What not to cut

Some savings are borrowed from the future at a punitive rate:

  • Redundancy you have not tested. Removing a standby node because nothing has failed lately is not optimisation, it is a wager.
  • Backup retention. Shortening it reduces the invoice and shortens the recovery window at the same time.
  • Observability. Cutting logging and monitoring reduces spend now and lengthens every future incident.
  • Security controls. The cheapest way to pay for a security tool is once, before the incident.

Making it stick

Optimisation that is not operationalised reverts within two quarters. The minimum viable governance loop is small:

  1. Tagging standard enforced at provisioning, with an owner on every resource.
  2. Budgets and alerts per account or product, reviewed monthly by someone with authority to act.
  3. A monthly cost review with a short agenda: what changed, what grew, what was removed, what is next.
  4. Cost awareness at design time. A ten-minute conversation during architecture review is cheaper than a migration later.

Cloud cost is an operational output, like uptime. Treat it with the same cadence and you stop having the annual emergency.

Start a conversation

Let's simplify your cloud infrastructure.

Tell us about your business and infrastructure requirements. Our team will assess your needs, recommend an approach and prepare a proposal.