Home / Services / Cloud, cost and accessibility

Cloud cost optimisation that engineers own

By Mirav Labs, led by Deepak MittalUpdated
Short answer

Cloud cost optimisation means paying only for the capacity you actually use, and keeping it that way. We audit your accounts for idle and oversized resources, rightsize instances, storage and Kubernetes requests, tag and allocate spend so every team sees its own bill, buy commitments only for steady load, and set budgets and alerts with named owners.

Key takeaways

  • Most cloud waste is created by engineering defaults, so engineers have to be part of the fix.
  • An audit looks for idle resources, oversized instances and Kubernetes requests, storage, egress, old logs and non-production running all night.
  • Tagging and allocation come first: a team that cannot see its spend cannot manage it.
  • Rightsize before you commit, so you do not lock in discounts on capacity you do not need.
  • Guardrails and a monthly review with named owners keep savings from drifting back.

Your cloud bill is a record of engineering decisions. Every instance size, every replica count, every log stream with no expiry and every test environment left on over the weekend shows up as a line item. That is why cloud cost optimisation in India works best when it is treated as an engineering job, not only a finance review at the end of the quarter.

We work with your engineers and your finance team together. We find the waste, fix what is safe to fix, make sure every team can see what it spends, and put guardrails in place so the bill does not creep back up six months later.

Cost is an engineering problem as much as a finance one

Finance sees the invoice. Engineering creates it. When the two teams look at different numbers, nobody can answer the simple questions: which product costs what to run, why did spend rise last month, and who should act on it.

The FinOps Foundation describes FinOps as an operational framework and cultural practice for getting the most business value from technology, built on engineering, finance and business teams working together. Its framework runs in three repeating phases: Inform (understand cost and usage), Optimize (find ways to improve efficiency) and Operate (carry out the changes, then loop back). Our five steps below follow that same cycle, with the engineering detail filled in.

The practical point is simple. Most waste comes from sensible defaults that nobody revisited: a generous instance size picked at launch, Kubernetes requests copied from another service, storage that grows because deletion was never designed in. Fixing these needs someone who can read a deployment manifest as well as a billing export.

What a cloud cost audit finds

A cloud cost audit reads your billing data, usage metrics and resource inventory side by side. Whichever cloud provider you use, the same kinds of waste tend to show up. This is the checklist we work through.

Where waste hidesWhat we look forUsual fix
Idle resourcesInstances, databases and load balancers with near-zero traffic for weeksConfirm the owner, then stop or delete
Oversized instancesCPU and memory use far below what was provisioned, even at peakMove to a smaller size or a different family
Kubernetes requestsPods requesting far more CPU and memory than they use, leaving nodes half emptySet requests from measured use, tune autoscaling
Unattached volumes and old snapshotsDisks with no machine attached, snapshots of systems that no longer existDelete after checking backup policy
Storage in the wrong tierRarely read data sitting in the most expensive storage classLifecycle rules that move or expire it
Logs kept foreverLog and metric streams with no retention limit, debug logging left onRetention periods per log type, lower verbosity
Data transfer and egressTraffic crossing zones or regions without need, large downloads served from originKeep chatty services together, cache at the edge
Non-production running 24x7Development, test and staging environments running nights and weekendsSchedules that shut them down outside working hours
Unused commitmentsDiscounts bought for capacity that has since moved or shrunkRebalance at renewal, size new ones to the baseline

Each finding goes into a register with the resource, its owner, its monthly cost, the proposed fix and the risk of making it. Nothing is deleted on the strength of a dashboard alone. Some "idle" systems are disaster recovery standbys or backups, and those belong to your cyber resilience plan, not the bin.

How we do the work: five steps

The audit tells you where you stand. These five steps turn it into a lower bill that stays low.

How we bring cloud spend down and keep it downSee to keep watching
Step 1See

Tag and allocate every rupee

Step 2Rightsize

Match capacity to real use

Step 3Commit

Discounts for steady load

Step 4Control

Budgets, alerts, guardrails

Step 5Keep watching

Monthly reviews, named owners

Now:

1. See: tag and allocate every rupee of spend

We start with cloud cost allocation, because a team that cannot see its spend cannot manage it. We agree a short tagging standard with you, usually owner, team, product, environment and cost centre, and enforce it at creation time through policy and your infrastructure-as-code templates. For shared costs that cannot be tagged, such as a shared cluster, networking or support charges, we agree a fair split rule. The result is a monthly view in which every team sees its own number and how it moved.

2. Rightsize: match instances, storage and Kubernetes requests to real use

Rightsizing means choosing the size that fits measured demand, with headroom, rather than the size someone guessed at launch. We look at several weeks of usage, including month-end and festive-season peaks where they matter to you. Changes go out in small batches, through your change process, with performance watched before and after. Storage gets lifecycle rules, logs get retention limits, and non-production environments get schedules.

3. Commit: discounted commitments for steady load

Once usage has settled after rightsizing, we look at the baseline that runs all day, every day. That steady load is the right place for your cloud provider's commitment discounts. Spiky or short-lived work stays on demand, and batch jobs that can tolerate interruption can move to interruptible capacity. We size commitments to what you will still be running, not to last quarter's peak.

4. Control: budgets, alerts and guardrails

Savings drift back when nothing stops them. We set budgets per team and environment with alerts that reach a named person, anomaly alerts for sudden jumps, and policies that block the common mistakes: untagged resources, oversized instance types in development, public storage, logs with no retention. Where it fits your pipeline, a cost estimate appears in the pull request, so engineers see the effect of a change before it ships.

5. Keep watching: monthly reviews with named owners

Each month, each team's owner reviews their spend against budget, explains the big movements and picks the next fixes from the register. We run the first few reviews with you and hand over a format your team can run alone. The aim is a habit, not a one-time clean-up.

Kubernetes cost optimisation

Kubernetes hides cost well. The bill shows nodes, but the decisions that fill those nodes are made in each team's manifests. A service that requests four CPUs and uses half of one still reserves four, and the scheduler adds nodes to fit requests, not real use.

Our Kubernetes cost optimisation work typically covers:

  • Requests and limits. Set from measured use with headroom, so pods pack onto fewer nodes without being starved.
  • Autoscaling. Horizontal scaling for traffic, cluster autoscaling that actually removes nodes when load falls.
  • Node pools. Machine types that match the workload shape, and separate pools for batch work that can use cheaper, interruptible capacity.
  • Namespace cost reports. Spend broken down by namespace and label, so each team sees what its services cost on the shared cluster.
  • Forgotten workloads. Old preview environments, abandoned jobs and persistent volumes left behind after a service was removed.

GPU workloads deserve their own look. Training and inference clusters are expensive to leave idle, and if you are building AI systems, sizing them sits close to the work in our AI data readiness practice.

Deleting data you no longer need

Storage is one of the few cost lines that only grows on its own. Every backup, log and export you keep costs money each month, and every copy of personal data is one more place it can leak from.

India's DPDP Act points the same way. When its obligations take effect on 13 May 2027, a Data Fiduciary must erase personal data once the purpose is no longer served, unless a law requires it to be kept. The DPDP Rules also set a floor: personal data, traffic data and processing logs must be kept for at least one year for specified purposes. So the right retention schedule is neither "keep everything" nor "delete everything": it is a defined period per data type, enforced by lifecycle rules. Our colleagues at DPDP Ready explain this in their guide to DPDP data retention.

Done well, one piece of engineering cuts your storage bill and your privacy risk at the same time. If you also need data kept in specific regions, our sovereign cloud and data residency work covers where it lives; this work covers how long.

What you get

DeliverableWhat it contains
Cost audit reportFindings across the waste checklist, each with resource, owner, monthly cost, fix and risk
Prioritised fix planQuick, safe changes first, then the larger engineering work, sized for your sprints
Tagging standard and policyRequired tags, enforcement rules and shared-cost split rules
Allocation dashboardMonthly spend by team, product and environment, with trend and budget
Kubernetes sizing changesMeasured requests, limits and autoscaling settings, raised as pull requests for your review
Commitment planThe steady baseline and the commitments that fit it, with renewal dates
GuardrailsBudgets, alerts, anomaly detection and preventive policies, each with a named owner
Monthly review packThe format, agenda and runbook your team uses after we step back

We work with read-only access for the audit. Changes to your environment are made by your team or by us through your normal change process, with your approval.

How to start

Begin with a short call about your accounts, clusters and where the bill hurts most. We will ask for read-only billing and usage access, agree the scope, and come back with the audit findings and a prioritised plan. From there you choose how much of the fixing and the ongoing review you want us to run with you.

Talk to us about a cloud cost audit.

Frequently asked questions

What is FinOps and how is it different from cost cutting?

Cost cutting is usually a one-time push to shrink the bill. FinOps is an ongoing practice in which engineering, finance and business teams share data about cloud spend and make decisions together, so cost becomes one more thing teams manage alongside speed and reliability. The FinOps Foundation describes it as an operational framework and cultural practice.

How long does a cloud cost audit take?

For a typical set of accounts we aim to deliver the findings and a prioritised plan within two to four weeks. The time depends on how many accounts and clusters you run and how much of your spend is already tagged. Fixes then run in sprints with your team.

What access do you need for a cloud cost audit?

Read-only access to billing exports, usage metrics and resource inventories is enough for the audit. We do not need write access to production. Any change we suggest goes through your normal change process, and your team approves it.

How do you reduce Kubernetes costs without hurting performance?

We compare each workload's CPU and memory requests with what it actually uses over several weeks, including peaks, then set requests with headroom and let autoscaling handle bursts. Changes go out a few services at a time, with your latency and error dashboards watched before and after each one.

Should we buy commitment discounts before rightsizing?

Usually not. If you commit first, you may lock in a discount on capacity you were about to remove. We rightsize first, wait for usage to settle, then size commitments to the steady baseline that remains.

Does deleting old data help with DPDP compliance as well as cost?

It can. Under the DPDP Act a Data Fiduciary must erase personal data once the purpose is no longer served, unless a law requires it to be kept, and personal data you do not hold cannot leak. A retention schedule that deletes on time lowers your storage bill and your exposure together, as long as it respects the minimum periods the law sets.

Scope it with us

A 30-minute call to understand your systems and what good looks like. We aim to send a written, fixed-scope proposal within three working days.

Talk to us

Related services

This page describes engineering and security services. Where it mentions laws, rules or standards, it is general information, not legal advice.

Talk to us