← Blog
Case study

How we found ₹1,16,387/month in real Azure waste — traced to the rupee

CloudOptimizer AI Team · Pilot deployment, live Azure subscription

Most cloud cost tools will tell you that you're "typically" wasting 30–40% of your bill. That number comes from an industry benchmark, not your account. We wanted to know what a real customer's real bill actually showed — so we ran CloudOptimizer AI against a live, production Azure subscription with a real, invoiced monthly bill, and only counted what the data actually supported.

₹1,16,387
Real monthly bill
₹14k–36k
Identified savings (varied by scan)
3
Real data sources used

That bill figure isn't an estimate — it's the subscription's actual last-month invoice, pulled directly from Azure's own Cost Management API, the same number that shows up in the Azure portal's billing page. Every recommendation CloudOptimizer AI produced against that account was priced against one of two things: the resource's own real observed cost in that same billing data, or Azure's live retail pricing API when no billing history existed yet for a given resource. Nothing was estimated from a lookup table.

What it actually found

The scan surfaced waste across a handful of familiar categories — none of it exotic, all of it the kind of thing that accumulates quietly in any account that's been running for a while:

CategoryWhat was happeningBasis for the number
Dev/staging VMs & AKS clustersRunning 24/7 despite only being used during business hoursReal cost × exact off-hours fraction of the recommended schedule
Unattached managed disksLeftover from deleted or resized VMs, still billing monthlyReal observed cost — deleting removes it entirely
Public IPs / load balancersReserved but not attached to anythingReal observed cost or live retail price for that SKU
Premium-tier disks on low-utilization workloadsProvisioned above what the workload's actual IOPS/throughput needed14-day real Azure Monitor utilization data

Why the number is a range, not a single figure

Not every recommendation had enough real data behind it to justify one exact number. When a resource's real cost existed but the correct percentage to apply wasn't precisely computable — no exact schedule, no direct utilization signal — we showed a labeled low–high range instead of collapsing it into a single point figure that would have been more precise-looking than it actually was.

And where neither a real cost nor a computable real price existed at all — a handful of advisory-only findings, like "this cluster should be tagged with an environment so we can classify it automatically" — we showed no ₹ figure whatsoever. Not a small "~estimated" number next to it. Nothing. If the data doesn't support a figure, the honest answer is no figure, not a smaller guess.

The range itself

Across repeated scans against the same account (re-run as new schedules were approved and some recommendations were acted on), identified monthly savings landed between roughly ₹14,000 and ₹36,000 — a real spread, not noise, because the number legitimately changes as recommendations get approved, disks get resized, and the account's actual usage shifts month to month. A tool that always reports the same tidy percentage regardless of what changed in the account isn't measuring anything real.

That's the whole point of building it this way: a savings number a customer can hold up against their own Azure portal and have it match, rather than a number that just sounds plausible.

See what's real in your own account

Free scan, no card required — connect with Microsoft and get the same kind of breakdown in a few minutes.

Run a free scan →