Are you paying BigQuery on-demand rates for steady workloads, or for slots that sit idle?
The right BigQuery pricing model for every project, configured and tuned
On-demand bills you for bytes scanned and rewards bursty, scan-light workloads. Slots bill for capacity over time and reward steady, slot-heavy workloads. Most estates have both, and pay for the wrong one somewhere. We work out which is which from your own job history, configure the reservations with you, then monitor and tune baseline, autoscaling, queueing and commitments until the bill settles. Fixed price, nothing paid upfront.
Your team's time: about 3 hours in total, for access, applying the changes with us and the readouts.
- Price
- From $1,500, fixed
- Timeline
- 2 weeks to configure, then tuning
- Control
- You apply every change
- Reads
- Metadata, never table data
Is this right for you?
A good fit if
- You spend $5,000 or more a month on BigQuery and have never tested whether on-demand or editions is cheaper for each project.
- You bought reservations or commitments and are not sure they are the right size, or you suspect slots sit idle for much of the day.
- Queries queue at peak hours and the answer so far has been to raise max slots and hope.
- You are on on-demand and one or two projects scan so much that a reservation would plainly be cheaper, but nobody has done the sums.
- Finance wants a predictable BigQuery bill without the data team losing performance.
Probably not for you if
- You spend under about $2,000 a month on BigQuery. On-demand is almost always right there, and the free Slot vs On-Demand Advisor at tools.finitizer.com will tell you.
- You want someone to change reservations without your team reviewing it. Every change is applied by you, or with you on a call.
- Your problem is query cost rather than capacity pricing. Start with the BigQuery and Databricks Cost Audit.
What you walk away with
A pricing model per project, with the numbers
Each project's real 30 to 90 days of usage priced under on-demand and under each edition, side by side, with a clear recommendation and the monthly difference.
Reservations that fit the work
Baseline and autoscale ceilings set from your usage percentiles, assignments per project and job type, idle slot sharing decided deliberately, and queue times inside a target you agree.
A bill that stays settled
Monitoring and tuning after the first configuration, so peak changes, new workloads and expiring commitments are caught and adjusted rather than discovered on the invoice.
What is included
Every package includes all of this. The tiers below change the size of the estate we cover and how much hands-on help you get.
Workload analysis from your own history
Per-minute slot usage from INFORMATION_SCHEMA jobs and reservation timelines, bytes scanned per project, queue times, peak hours and how variable each workload is. We classify every project as steady, business-hours or spiky.
The on-demand versus slots decision, per project
Scan-heavy, steady projects favor editions; small or bursty projects favor on-demand's zero idle cost. The crossover sits near 10 GB scanned per slot-hour at Enterprise rates. We price both ways for each project and recommend the cheaper one that meets your performance needs.
Reservation design and configuration
Edition choice, baseline slots, autoscale maximum, per-project and per-job-type assignments and whether idle slots are shared, written as the exact bq commands or DDL. We apply them with your engineer on a call, or you apply them.
Commitments sized from the distribution
Where a steady baseline justifies it, a 1-year or 3-year slot commitment sized to a conservative percentile of consumption, laddered so renewals never land on one day. You buy in your own console.
Queueing and concurrency tuning
Where jobs wait, we find the reservation and hour responsible and adjust baseline, ceiling or assignments so queue times stay inside the target, typically a few seconds at the 95th percentile.
Monitoring and tuning after go-live
Weekly checks of utilization, idle capacity, queueing and spend against the plan, with adjustments proposed as the workload changes. Sixty days on Starter, ninety on Standard and Scale.
Guardrails
Maximum bytes billed on on-demand projects, custom quotas, budgets and alerts, so a runaway query cannot undo the savings.
What we hand over
- Pricing model decision per project: on-demand or edition, with both costs side by side
- Reservation plan: edition, baseline, autoscale maximum, assignments and idle sharing per reservation
- The exact bq commands or DDL for every change, with rollback
- Commitment recommendation with term, size, break-even and a renewal ladder
- Weekly tuning notes during the monitoring period, and a monthly summary
- Guardrails checklist: quotas, maximum bytes billed, budgets, alerts
- Final readout with before-and-after spend, utilization and queue times
How it works
Day 0
Free scoping call
Thirty minutes on your projects, current pricing model, reservations and pain points. If you are on on-demand, we ask you to run the free Slot vs On-Demand Advisor first; its result often sets the scope.
Day 1
Grant read access
Viewer roles on the projects in scope and read access to reservation history in the administration project, granted to a dedicated Finitizer reader identity. Five minutes in Cloud Shell. Or export the history yourself and upload it.
Days 2 to 7
Analysis and plan
We price every project both ways from its own history, design the reservations and size any commitment. You get the plan with the exact commands and the expected monthly change.
Days 8 to 14
Configure with you
On a call with your engineer, in your console, we apply the reservation changes together, starting with the lowest-risk ones. Commitments are bought by you, only when you are satisfied with the reservation data.
Days 15 to 90
Monitor and tune
Weekly checks and adjustments as the workload settles: baseline and ceiling nudges, assignment changes, queue fixes, and a watch on idle capacity. A monthly summary shows spend against the plan.
Day 90
Final readout
Before-and-after spend, utilization and queue times, the renewal calendar, and what to watch next. Keep tuning through FinOps-as-a-Service if you want us to stay.
Pricing
One fixed fee per engagement, agreed before we start, with no upfront payment: 50% is invoiced once the reservations are configured, about day 14, and 50% after the final readout. Pay by card or bank transfer. Prices are in US dollars and exclude applicable taxes.
Starter
Up to 5 projects, one reservation administration project
Typical BigQuery spend: $5,000 to $25,000 a month
$1,500USD, one-time
- Pricing model decision for every project
- Reservation plan with the exact commands
- Configuration call with your engineer
- 60 days of monitoring and tuning
- Final readout
Standard
Up to 20 projects, several reservations
Typical BigQuery spend: $25,000 to $100,000 a month
$3,000USD, one-time
- Everything in Starter
- 90 days of monitoring and tuning
- Commitment sizing with a renewal ladder
- Queueing and concurrency tuning
- Guardrails configured with you
- Monthly summary for finance
Scale
Up to 50 projects, several administration projects or regions
Typical BigQuery spend: $100,000 a month and up
$5,000USD, one-time
- Everything in Standard
- A working session with your data platform team every month
- Showback of slot cost by reservation, project and team
- A full re-plan at day 90 as workloads change
- Advice on Enterprise Plus and negotiated pricing
- Want us to keep monitoring and tuning after the readout? Add it at $375 a month, or include it in FinOps-as-a-Service.
- More than 50 projects, or several organizations? Book a call for a quote.
- Start FinOps-as-a-Service within 60 days and the full fee is credited to your retainer.
Twice the fee, or your money back
If the engagement does not identify at least twice its fee in annual savings, estimated conservatively from your own usage data and priced at list rates, we refund the fee in full. It is written into your order, and you keep the plan either way.
The data we need from you
Everything is sized from your own job and reservation history, so we need at least 30 days of it, and 90 is better. Grant read access and we collect it, or export it and upload it.
- If you grant access: BigQuery Resource Viewer and Metadata Viewer on each project in scope, and BigQuery Resource Admin, used read-only, on the reservation administration project for reservation, assignment and commitment history.
- If you export instead: INFORMATION_SCHEMA.JOBS_TIMELINE_BY_ORGANIZATION or JOBS_TIMELINE_BY_PROJECT (per-minute slot usage), JOBS_BY_PROJECT (bytes billed, queue time, job type) for each project, and RESERVATIONS_TIMELINE, RESERVATIONS, ASSIGNMENTS and CAPACITY_COMMITMENTS from the administration project, for the last 30 to 90 days, as CSV or Parquet.
- The BigQuery lines of your Cloud Billing export for the same period, so both pricing models are costed at what you actually pay, including any negotiated rates.
How it reaches us: Exports go through a secure, customer-specific upload link we send you at kickoff. Everything is stored encrypted in Finitizer's Google Cloud environment in the US, used only for your engagement, and deleted on request. Query text, where present, is used only to attribute cost.
Access and security
We read job, reservation and commitment metadata. We never read the data in your tables, and we never hold permission to change a reservation or buy a commitment.
- Roles are granted per project and on the administration project to a dedicated Finitizer reader identity, using short-lived impersonation tokens. No keys are created and no organization-level role is needed.
- Resource Admin on the administration project is used read-only, for the reservation and commitment timelines. Without it the analysis is shallower, and we say so in the readout.
- Reservation changes are run by your engineer, or by us together on a call in your console. Commitments are purchased by you.
- Job history includes query text and the user who ran each job; we use them only to attribute cost, keep them in a dedicated dataset for your company, and delete them on request.
Cannot grant access yet? Run the free Slot vs On-Demand Advisor at tools.finitizer.com. You run one read-only query yourself, the report is computed in your browser, and we can start the conversation from its result.
Who does the work
Nitin GandhiFounder, Finitizer
Every engagement is led by Finitizer's founder, a FinOps practitioner who has run cost programs inside large enterprises, not just advised them. The Finitizer platform does the scanning; the judgment is human.
- 25 years in enterprise IT
- FinOps programs for IT budgets of more than $220 million
- Five FinOps certifications
- MBA, Cornell University
Frequently asked questions
How do you decide between on-demand and slots?
From each project's own history, not a rule of thumb. On-demand bills bytes scanned, so a workload that scans a lot per unit of compute is expensive there and cheap on slots. Editions bill slot time, so a bursty workload that is idle most of the day is expensive there and cheap on-demand. We price both for every project over the same 30 to 90 days and recommend the cheaper one that still meets your performance needs. The crossover is roughly 10 GB scanned per slot-hour at Enterprise rates.
What do baseline, autoscale maximum and queueing mean, and what do you tune?
A reservation has a baseline of slots you pay for continuously and an autoscale ceiling it can grow to in 50-slot steps when demand rises. If both are too low, jobs queue; if either is too high, you pay for idle capacity. We set the baseline near steady usage, the ceiling near the 95th percentile, assign projects and job types deliberately, and then adjust as the real queue times and utilization come in.
Will this slow our queries down?
That is the constraint we tune against. You agree a queue-time target up front, typically a few seconds at the 95th percentile, and no change that breaches it stays in place. Many estates get faster, because the biggest workloads get dedicated capacity.
Do you change our reservations yourselves?
No. We hold read-only access. Every change is written as an exact command with its rollback, and your engineer runs it, or we run it together on a call in your console. Commitments are always bought by you.
Should we buy a slot commitment?
Only where a steady baseline justifies it. A 1-year commitment is about 20% below pay-as-you-go edition rates and a 3-year about 40%, but a commitment to capacity you do not use costs more than it saves. We size any commitment to a conservative percentile of real consumption and ladder the terms.
Which editions do you cover?
Standard, Enterprise and Enterprise Plus, and on-demand. Standard cannot hold commitments and has no autoscale baseline, which matters for the decision, so it is part of the comparison rather than ignored.
What happens after the monitoring period?
You keep the plan, the commands, the renewal calendar and the monitoring queries. If you want us to keep watching and tuning, add monitoring at $375 a month or include it in FinOps-as-a-Service.
Will your analysis increase our BigQuery bill?
Barely. We read INFORMATION_SCHEMA views, which process a small amount of metadata; for most estates the whole engagement costs a few dollars of query time.
How is this different from the BigQuery and Databricks Cost Audit?
The audit looks at everything that drives the bill: queries, tables, storage, slots and clusters, and hands you a ranked fix list. This service goes deep on one lever, capacity pricing, and does not stop at a recommendation: we configure the reservations with you and tune them for 60 to 90 days. If you are not sure which you need, book a call.
How do we pay?
Nothing is paid upfront. Half the fee is invoiced once the reservations are configured, about day 14, and the other half after the final readout. Invoices can be paid by card or bank transfer, and we can work with a purchase order.
Ready to start?
Pick a package above, or book a free 30-minute call and we will tell you which one fits, or whether you need one at all. From $1,500, one-time fee.