Cloud Cost Optimization

The usual suspects: EC2 sized for a peak that happens twice a year, gp2 volumes nobody migrated to gp3, NAT gateway data processing, cross-AZ traffic showing up as an unlabeled line item. We work from your Cost and Usage Report rather than the console summary, because that's where the surprises live. Commitments get reviewed too, since a Savings Plan bought for last year's architecture is often paying for the wrong things.

Azure bills fail differently. Lift-and-shift VMs still sized like the datacenter boxes they replaced, orphaned managed disks and public IPs left behind by deleted machines, Log Analytics ingesting everything at premium rates. We review reservations and Hybrid Benefit eligibility, and untangle subscription sprawl so cost has somewhere sensible to land.

A read-only pass through your billing data and your accounts, ending in an itemized set of findings: the resource, what it costs per month, what changing it involves, and what could go wrong. Not a themes-and-recommendations deck. Any line of it can be handed to an engineer and acted on.

Sizes chosen under deadline pressure become permanent. We look at actual utilization over weeks, including the spikes, and resize with headroom for failover and growth rather than to the prettiest average. Sometimes the right recommendation is to leave a resource alone, because an undersized production database costs more than it ever saved.

Every mature account has them: load balancers routing to nothing, unattached volumes, stopped instances still paying for storage, the test cluster from an experiment that ended in March. Finding them is the easy part. The real work is tracing ownership so deletion is a decision rather than a gamble; nothing goes until someone who understands the system confirms it can.

Non-production environments commonly run 168 hours a week for teams that work 40 of them. We automate stop and start on schedules the teams control, with an override that takes one command, because a schedule people can't escape gets deleted. Environments that hold state or run overnight jobs don't fit this pattern; those get downsized instead.

Storage only ever accumulates. Snapshots of servers retired years ago, logs kept forever by default, buckets nobody can name an owner for. We set lifecycle and retention rules that match what your compliance obligations say, and we're careful with archive tiers: cheap to fill, expensive to read, and wrong for anything you still touch.

You can't ask anyone to own a cost you can't attribute to them. We design a tagging standard small enough that people follow it, and enforce it in Terraform and account policies instead of a wiki page. The awkward remainder gets handled in the open: shared networking, support fees, and the cluster four teams use will never tag cleanly, so we allocate them by an explicit rule everyone can see.

Most cost dashboards get opened twice. We build reporting that engineers and finance both trust because it comes from the same data: cost per service and per team, unit costs where your business has a meaningful unit, and anomaly alerts tuned so a notification means something. Then we help you set a review cadence with named owners, because visibility nobody acts on is just a prettier bill.