BigQuery Cost Optimization & Editions Advisory

BigQuery cost is the question every buyer asks and the one the pricing page answers least clearly. Since flat-rate and flex slots were retired on July 5, 2023, there are two ways to pay for compute — on-demand (per byte scanned) and BigQuery Editions (Standard, Enterprise, Enterprise Plus, billed per slot-hour with autoscaling) — and two ways to pay for storage (logical or physical). Pick the wrong combination and a workload can cost three times what it should without a single slow query to show for it. This service exists to get the combination right and keep it right.

On-Demand vs. Editions: A Benchmarked Decision

We do not guess. For each project we pull thirty to ninety days of INFORMATION_SCHEMA.JOBS history, compute total bytes billed and slot-milliseconds per workload, and price the same history three ways: on-demand, Standard edition, and Enterprise edition with autoscaling. The output is a table, not an opinion. Typical patterns we find: bursty analyst workloads that are cheapest on-demand, steady ELT workloads that are cheapest on Standard, and mixed estates where the right answer is a reservation for scheduled pipelines and on-demand for everything else — which BigQuery supports through per-project assignments.

Autoscaling Reservations That Do Not Overspend

Editions reservations scale in 50-slot increments between a baseline you commit to and a maximum you cap. Set the baseline too high and you pay for idle capacity; set the maximum too low and the nightly load misses its SLA. We size both from observed slot-usage percentiles, assign workloads to reservations by project and job type, and configure one-year or three-year commitments only for the portion of demand that is genuinely constant. We also review whether Enterprise-only features you are paying for — column-level security, materialised view rewrites, advanced BI Engine capacity — are actually in use.

Storage Billing and Lifecycle

Storage is often a third of the bill and the most neglected lever. Per dataset we evaluate switching from logical to physical billing (a large win for highly compressible data, a loss for datasets with long time-travel windows), shorten the time-travel window where seven days is not required, set partition and table expiration on staging data, and convert long-term cold data to cheaper long-term storage simply by ensuring it is not touched. None of this changes a query; all of it changes the invoice.

FinOps Instrumentation

Savings erode without visibility. We deliver:

• A scheduled export of INFORMATION_SCHEMA.JOBS_BY_ORGANIZATION into a cost dataset with per-user, per-team and per-label attribution
• A Looker Studio chargeback dashboard showing spend by team, top queries by bytes billed, cache-hit rate and reservation utilisation
• Budget alerts at the project and billing-account level, with an escalation path that reaches the team responsible rather than a shared inbox
• Custom quotas on bytes billed per user per day so one runaway query cannot consume a month's budget

Query-Level Waste

Alongside the pricing model we fix the queries that make any model expensive: SELECT * over wide tables, missing partition filters (enforced with require_partition_filter), scheduled queries that recompute unchanged data, and dashboards that bypass the result cache because of non-deterministic functions. These are covered in depth by our performance optimization service and usually pay for the engagement on their own.

Quick-Win Cost Audit

Our fixed-scope cost audit takes two weeks: one week of analysis, one week to implement the no-regret changes and hand over the dashboards. Engagements on estates that have never been through this exercise typically cut 30 to 60 percent of BigQuery spend without slowing a single workload; the result depends on your starting point, and we show you the numbers before you commit to any change. Contact us to book an audit.

Back to services