How to Audit Your Google Cloud Bill: A Practical CSV Walkthrough (Feb 2026)

Google Cloud bills look deceptively clean — fewer line items than AWS, fewer SKUs than Azure. That cleanliness hides a different problem: GCP waste compounds silently in three services (BigQuery, Compute Engine, Cloud Storage), and most teams don't notice until the bill triples.

Step 1 — Export the CSV (60 seconds)

In the Google Cloud Console:

1. Billing → Reports
2. Set the time range to Last month (closed)
3. Filter by Project / Service — leave wide open for the first audit
4. Export to CSV (top right)

You'll get a file like gcp-billing-2026-01.csv.

💡 If you want line-item granularity (per-SKU, per-day), enable BigQuery billing export once. It's free, runs every day, and lets you write SQL against your own bill. For a first audit the basic CSV is fine.

Step 2 — The 4 columns to focus on

Column	What it tells you
Service	GCP service (e.g. Compute Engine, BigQuery)
SKU	The specific charge (e.g. Analysis or N1 Standard Instance Core running in Americas)
Usage amount + Usage unit	How much you used
Cost	What you paid

Sort by Cost descending. Top 30 rows = ~85% of bill.

Step 3 — The 5 patterns we see on every GCP audit

1. BigQuery on-demand scans (the $/TB tax)

Service = "BigQuery" AND SKU contains "Analysis". On-demand pricing is $6.25/TB scanned. We see teams running a 4 TB SELECT * query every hour from a Looker dashboard — that's $625/day, $19K/month, on one chart. The fix is partitioned tables + clustering, plus a hard query-bytes cap on the dataset.

2. Idle Compute Engine VMs

Service = "Compute Engine" AND SKU contains "Instance Core". Compare cost to your active workload. Most teams have dev VMs running 24/7 that should be on shutdown-on-idle (built into GCE — most teams don't know it exists). Median savings: $200–$1,200/month per orphan VM.

3. Cloud Storage egress to internet

Service = "Cloud Storage" AND SKU contains "Download". At $0.12/GB egress to internet (cheaper than AWS, but still pricey), public buckets that don't have Cloud CDN in front waste 30–70% of their egress dollars.

4. Vertex AI silent overage

Service = "Vertex AI". This one's new and dangerous. A single Gemini Pro API call costs ~$0.0001 — but a training job on a misconfigured n1-highmem-32 cluster can cost $30/hour and run for days. Look for any SKU containing Training or Custom Training and check the cost against what you authorized.

5. GKE node-pool waste

Service = "Kubernetes Engine". Most teams over-provision system node pools at the default n1-standard-1 for control-plane workloads that fit in e2-small. Sum cost per node pool and compare to actual pod resource requests — almost always a 30–50% over-spend.

Step 4 — The "cost per project" 30-second pivot

Pivot:

• Rows: Project name
• Values: sum of Cost

Any project you don't immediately recognize? Probably an old PoC. Most enterprises have 6–14 forgotten GCP projects with $50–$500/month running quietly.

Step 5 — Sustained Use Discounts and CUDs

Unlike AWS RIs, Sustained Use Discounts are applied automatically on GCE — you don't need to buy anything. But Committed Use Discounts (1- or 3-year commits) save another 25–55% on top, and most teams don't enable them until the bill is huge. If your monthly GCE spend is above ~$3K, CUDs are usually a no-brainer.

Step 6 — Audit the lazy way

CARTIE's free GCP bill audit reads your CSV in-memory, ranks your top 5 leaks by $/month, and tells you which SKU and project to look at. No service account, no IAM, no retention.

If you want CARTIE plugged into GCP for continuous monitoring (with BigQuery slot recommendations and per-query cost attribution), that's our paid product.

---

Companion read: BigQuery Slot Tuning: How to Cut Query Costs 60%.