A GCP data review works through where data spend concentrates: BigQuery query cost and the expensive recurring queries, the BigQuery pricing model choice between on demand and capacity, table partitioning and clustering, Cloud Storage class and lifecycle rules, idle and orphaned datasets, and egress. You run it from the billing export, give every finding an owner and a decision, then automate the recurring parts so it holds. On most GCP estates the data platform, led by BigQuery, is both the biggest line and the one with the most reversible waste.
This is a GCP data and BigQuery discipline. Here is the order to work the review and the numbers to capture.
Start from the billing export and INFORMATION_SCHEMA
Pull data cost from the billing export in the FOCUS format, then use BigQuery's INFORMATION_SCHEMA views to rank queries and jobs by bytes scanned and slot usage. The billing export tells you how much you spent; INFORMATION_SCHEMA tells you which queries and which users drove it. Together they let you assign every expensive query to an owner. If datasets carry labels for team and product, the review is fast; if not, labelling is the first step, covered in the GCP cost optimization guide.
Where does GCP data spend concentrate?
Work the review in priority order, biggest reversible cost first.
| Check | What to look for | The decision |
|---|---|---|
| Top queries | Recurring queries scanning huge byte volumes | Partition, cluster, or rewrite to scan less. |
| Pricing model | Steady heavy load on on demand pricing | Test capacity pricing with autoscaling slots. |
| Table design | Unpartitioned, unclustered large tables | Add partitioning and clustering to cut scans. |
| Storage class | Cold data in Standard storage | Lifecycle rules to Nearline, Coldline, Archive. |
| Idle datasets | Datasets and tables nobody queries | Confirm ownership, then archive or delete. |
| Egress | Cross region and internet egress | Co locate compute and data; review patterns. |
How do you cut BigQuery cost?
On demand BigQuery bills on bytes scanned, so the cheapest query is the one that reads less. Partitioning and clustering tables, selecting only the columns a query needs instead of scanning everything, and eliminating accidental full table scans cut cost directly and immediately. The single most expensive pattern is a recurring scheduled query that scans a giant unpartitioned table every run; fixing one of those can move the bill on its own.
The second lever is the pricing model. On demand suits spiky, unpredictable query volume. For steady, high volume workloads, capacity pricing with reserved or autoscaling slots can be materially cheaper, because you pay for processing capacity rather than per byte. Many estates run a mix, and the review is where you decide the split. We go deeper in BigQuery cost optimization end to end.
A Fortune 500 retailer assumed its GCP data bill was simply the cost of analytics at scale. INFORMATION_SCHEMA told a sharper story: a handful of scheduled queries on unpartitioned tables drove a large share of bytes scanned, the warehouse ran entirely on demand despite a steady daily load, and a multi petabyte bucket kept years of cold exports in Standard. Partitioning the heavy tables, moving the steady load to capacity pricing, and adding Cloud Storage lifecycle rules cut the data bill substantially. Figures are verified against billing data and anonymised.
Tier storage and clear idle data
Cloud Storage is the second front. Move data by access pattern: frequently read data stays in Standard, monthly access belongs in Nearline, quarterly in Coldline, and compliance archives in Archive, all driven by lifecycle rules so it happens automatically. Then clear the idle datasets and orphaned tables that accumulate when projects end and nobody deletes the data. Decommissioning data you no longer need is its own habit, covered in decommissioning GCP data you no longer need. Watch egress and retrieval costs when tiering, since cold data read more than expected can erode the saving.
Keep the data review standing
A one time data cleanup reverts as new queries and datasets appear. Automate the recurring parts: lifecycle rules as the default for new buckets, a top queries by bytes scanned report in the monthly review, and a quarterly pricing model check as load grows. Run the full playbook quarterly. This is the discipline that holds the 31 percent median reduction we see in the first 90 days. GCP Recommender will flag some of this, but it recommends, it does not decide, so confirm each change against real query patterns.
Frequently asked questions
Run the data review with us
We run the GCP data review for you, rank the expensive queries from INFORMATION_SCHEMA, choose the right BigQuery pricing model, and automate the storage tiering that keeps the bill down. Our guarantee: we reduce your cloud spend or we reimburse our service fee.
Put a defensible number on your cloud spend.
No provider in the room, no published price list. Tell us your footprint and we will scope the savings against your billing data — we reduce your cloud spend or we reimburse our service fee.
The Cloud Spend Navigator: what changed in cloud pricing, commitments, and FinOps — no vendor spin.