The Top BigQuery Cost Management Tools in 2027
Kristóf Horváth
26 min read
This post compares the top BigQuery cost management tools on Google Cloud in 2027, including which ones automate changes and which only recommend them, starting with Google Cloud’s built-in tooling and then seven vendors in alphabetical order, for engineering and data teams looking for a platform to optimize BigQuery costs and/or performance.
Last verified: 17 November 2026. Every vendor's information comes from its own website and documentation. Spot a mistake? Email marketing@followrabbit.ai.
Disclosure: this post is published by Rabbit, which is one of the tools below. All tools are described with the same criteria.
Quick answer (not a ranking):
- Start with Google Cloud’s native controls. The slot recommender and Cost Anomaly Detection have no additional charge, but querying exported billing or metadata can incur normal BigQuery costs.
- Alvin, Masthead, Rabbit, and Unravel describe some form of automated change. DoiT, Revefi, and Ternary mainly provide visibility and recommendations.
- Alvin, Rabbit, and Unravel describe generating SQL rewrites. Of those, Alvin and Rabbit also choose between on-demand and capacity pricing for individual jobs; Table A shows the full picture.
- The tools also differ in how they access BigQuery: metadata-only ingestion, an optimizer in the query or job path, or write roles and code-level hooks for automation.
What should you look for in a BigQuery cost management tool?
A useful tool, whether you call it a BigQuery FinOps tool or a cost optimization platform, saves more than it costs, in dollars and in engineering hours. We described every tool with the same eight criteria:
- Savings and engineering effort. What the vendor claims, always attributed. Note: vendors measure savings differently, so the figures are not comparable.
- Pricing transparency. How you are charged, and whether the vendor publishes pricing on its website.
- BigQuery depth. Four levers: slots and reservations; query and table optimization; storage; attribution and alerts.
- Automation and guardrails. What each platform changes on its own (including per-job pricing routing), who approves it, and how it rolls back.
- Integration and data access. What you install, what runs in your environment or query path, and how long first value takes, as stated by the vendor.
- Security posture. Vendor-stated data handling, assurance reports, and certifications.
- Enterprise readiness. Named customers, marketplace listings, and regional support.
- Fit. Who it is for, and who it is not for.
Cells in the tables use plain statuses: Automated, Customer-configured, Recommends, Visibility, Not offered, and Not described. “Not described” means we found nothing on the vendor’s website or documentation, not that the capability does not exist. “Announced; not shown as shipped” separates a launch announcement from a documented current feature.
How do the top BigQuery cost management tools compare?
Table A: what each tool does, by lever
| Tool | Focus | Slots and reservations | Query and table optimization | Storage | Attribution and alerts | Automation model |
|---|---|---|---|---|---|---|
| Google Cloud native | Built-in recommendations and on-demand limits | Recommends | Recommends (tables); on-demand guardrails | Visibility (storage cost forecast) | Visibility and alerts; on-demand quotas | Recommendations need action; configured limits are enforced; per-job pricing: manual override only (project-level recommendation) |
| Alvin | Query-path automation | Automated | Automated, per vendor (query rewrites) | Not described | Visibility; alerts not described | Vendor-managed after setup; per-job pricing: automated |
| DoiT | Multi-cloud data cost governance | Recommends | Recommends | Recommends | Visibility and alerts | Visibility and recommendations; automation announced; per-job pricing: announced (SELECT), not shown as shipped |
| Masthead | Data observability plus reservation assignment | Recommends | Recommends | Not described | Visibility | Runs after you deploy its configuration; per-job pricing: customer-configured assignment |
| Rabbit | BigQuery automation across several levers | Recommends; Automated, opt-in | SQL rewrites: Automated, opt-in (agentic, for top-cost queries; you apply or reject); antipattern checks at pull request time; tables: Recommends | Recommends; Automated, opt-in | Visibility, alerts, and spend caps | Opt-in, per optimizer; per-job pricing: automated, opt-in, and per project with Reservation Optimizer |
| Revefi | Multi-platform cost and data monitoring | Recommends | Recommends | Not described | Visibility and alerts | Reservation script (adjusts maximums hourly); policy detail limited |
| Ternary | Finance-led allocation and commitments | Recommends | Not described | Not described | Visibility | Recommendations and reporting |
| Unravel | Controlled SQL and reservation automation | Automated, opt-in; Recommends | SQL rewrites: Automated, opt-in (AutoApply, or apply after review) | Recommends | Visibility and alerts | Opt-in, per workload; pricing model: recommended per plan only |
How to read the columns:
- Slots and reservations: sizing baselines, autoscaling limits, and commitments.
- Query and table optimization: rewriting queries or recommending changes to queries, tables, and jobs.
- Automation model: who applies the change. Where a tool routes individual jobs between on-demand and capacity pricing at runtime, this column says so; a pricing model recommended for a whole project is a different feature, and is shown as such.
Table B: cost and buying
| Tool | Claimed savings (per vendor) | Pricing model (as published) | Trial or POC | Integration | Security (as stated by vendor) |
|---|---|---|---|---|---|
| Google Cloud native | No figure | Slot recommender and anomaly detection at no additional charge; normal usage costs may apply | n/a | Built in | Google Cloud |
| Alvin | 30-70% on BigQuery | Percentage of savings | Free savings report; 2-week no-cost POC | Optimizer in the query path | SOC 2 Type II, GDPR |
| DoiT | No BigQuery figure published | Tiered platform plans (Essentials $0); SELECT by quote | Essentials plan; free trial for PerfectScale for BigQuery | Service account, metadata and billing export | SOC 2 Type II, SOC 3, ISO 27001:2022, GDPR |
| Masthead | 15-50% | Per Google Cloud project per month: $0, $549, or custom | 14-day trial | Agent in your project or Masthead’s | SOC 2 Type II |
| Rabbit | 32% average BigQuery cost reduction | Tied to your cloud bill and savings | 30-day free trial (extendable); 2-month free POC | Job and billing metadata, including SQL text with literals masked; plugins or proxy for job routing | SOC 2 Type II audited, ISO/IEC 27001:2022 certified, GDPR data processor |
| Revefi | 30-70% (across data platforms) | Percentage of data platform spend; free Starter under $50K | Free trial; free Starter plan | Metadata-only service account | SOC 2 Type II, HIPAA, ISO 27001 |
| Ternary | No BigQuery figure | Not published | Not described | Billing export and service account | SOC 2 Type II, ISO 27001 |
| Unravel | No BigQuery figure | Annual subscription or pay as you go | Not described for BigQuery | Service account | SOC 2 Type II |
Pricing and savings are as published on each vendor’s website. Savings figures use different bases, so do not rank on that column. Security lines are vendor-stated; we did not audit the reports or certifications.
Which tool fits which situation?
- You want to apply changes yourself, with no third party: Google Cloud native.
- You want SQL rewrites generated for you: Alvin says it rewrites queries automatically as they pass through its endpoint. Rabbit generates rewrites, tested on synthetic data, for your most expensive queries and checks new SQL at pull request time; Unravel generates rewrites you approve before they apply, and also offers AutoApply.
- You want each job routed between on-demand and capacity pricing: Alvin (in the query path) and Rabbit (dbt and Airflow plugins or a proxy). Masthead assigns workloads from a configuration you deploy.
- You want slots and reservations adjusted automatically: Alvin, Rabbit, and Unravel. Revefi supplies a script that adjusts maximums hourly.
- You need coverage beyond BigQuery: DoiT (multi-cloud), Revefi (Snowflake, Databricks, BigQuery), and Ternary (finance-led, across cloud, SaaS, and AI spend).
- You want data observability and cost in one product: Masthead and Revefi.
- Finance needs allocation, budgets, and commitment planning: Ternary across cloud, SaaS, and AI spend; Rabbit for BigQuery, with budgets and hard caps, commitment planning, and virtual labels for chargeback. DoiT and Unravel also offer budgets or showback and chargeback.
The tools
Google Cloud native tooling: the built-in baseline
Google Cloud’s native tools cover more than most teams assume: recommendations for reservations and tables, storage billing forecasts, plus budgets and anomaly alerts.
Best for: every BigQuery team as a starting point, especially teams with the engineering time to apply recommendations manually.
Claimed savings: Google publishes no savings figure.
Pricing model: the slot recommender and Cost Anomaly Detection have no additional charge, but queries over billing exports or INFORMATION_SCHEMA can incur normal BigQuery costs.
What it does for BigQuery:
- Slots and reservations: the slot recommender analyzes 30 days of usage and recommends commitment slots, autoscaling slots, and maximum reservation size. It covers Editions and on-demand workloads, including conversion to Enterprise edition. A single job can also be run on a chosen reservation, or on on-demand pricing, by setting its reservation when it is submitted.
- Tables: the partitioning and clustering recommender analyzes up to 30 days of workload to produce recommendations. Materialized view recommendations are available in Preview. Partitioning can only be changed on a copy of the table, so applying a partition recommendation means copying the table first; clustering recommendations can be applied to the existing table.
- Guardrails: custom daily query quotas and a maximum bytes billed per query limit on-demand spend. They apply to on-demand usage only.
- Storage: logical or physical billing is set per dataset, and
TABLE_STORAGEviews forecast cost under both. After switching, you wait 14 days to switch again. - Attribution and alerts:
INFORMATION_SCHEMA.JOBSexposes labels, reservation ID, and bytes billed. Budgets can send email or Pub/Sub alerts. Cost Anomaly Detection is enabled by default for eligible billing accounts, with root-cause detail by service, region, and SKU.
Automation and guardrails: optimization changes are recommendations that a person applies. Quotas and maximum-bytes-billed limits, once configured, are enforced by Google Cloud.
Integration and security: native. Nothing to connect, and no third party with access.
Where it is strong: built-in recommenders for reservations, partitioning, clustering, and materialized views (Preview), enforced daily quotas and maximum bytes billed on on-demand usage, and anomaly detection with root-cause detail by service, region, and SKU.
Limitations: each tool works at its own scope (a project, a billing account, a table), so nothing coordinates reservations, pricing model, query changes, and storage as one workflow. A job can be pointed at a reservation or at on-demand pricing by hand, but BigQuery does not choose between them for you, and recommendations need a person to apply them and revisit them as workloads change.
Alvin: query-path optimization and pricing-model routing
Alvin is a BigQuery-only optimizer that intercepts queries before they reach BigQuery and overrides the reservation ID at run time so they run on on-demand or capacity pricing. Its website also describes automatic query rewriting.
Best for: teams that can point workloads at an external endpoint and want query-level automation.
Claimed savings: 30-70% per Alvin’s website. The site also reports 52% average savings per customer and 15-30% shorter query duration, without stating the basis.
Pricing model: a percentage of savings; the percentage is not published. Alvin states there is no charge for the POC, no onboarding cost, and no minimum commitment, and it offers a free savings report and a 2-week no-cost POC.
What it does for BigQuery: query rewriting (cache hits, partition pruning, anti-patterns), per-query routing between on-demand and capacity pricing, reservation configuration, and cost reporting by query source. Storage optimization and anomaly alerts are not publicly described.
Automation and guardrails: described as autonomous once set up. We did not find public detail on rollback, approval workflows, or safety limits.
Integration and security: you point each query source (dbt, Airflow, Looker, and others) at Alvin’s endpoint through BIGQUERY_EMULATOR_HOST, which Alvin says takes about 15 minutes per source. For billing-model optimization, Alvin’s docs say it needs permission to manage your reservations and ask you to grant the BigQuery Resource Editor role. Alvin states that it holds a SOC 2 Type II attestation, is GDPR compliant, and does not store or process underlying BigQuery data.
Where it is strong: per-query routing between on-demand and capacity pricing and, per Alvin’s website, query rewriting (cache hits, partition pruning, anti-patterns), applied automatically once set up.
Limitations: Alvin sits in the execution path of production queries, so its endpoint becomes a dependency. We did not find public detail on failure behavior, rollback, or how query text, literals, and logs are handled. Billing-model optimization needs write permission on reservations. Storage optimization and alerts are not described.
DoiT: a multi-cloud platform with BigQuery visibility and a new automation product
DoiT covers BigQuery through Cloud Intelligence (including BigQuery Intelligence), PerfectScale for Data Platforms, and SELECT, which it acquired in January 2026.
Best for: teams that want multi-cloud governance, reseller procurement, and engineering support alongside BigQuery analysis.
Claimed savings: we did not find a BigQuery savings percentage on DoiT’s pages.
Pricing model: Cloud Intelligence has Essentials ($0), Enhanced, and Enterprise tiers. BigQuery Intelligence is included on every plan, and PerfectScale for BigQuery is priced separately, by quote. SELECT is quote-only.
What it does for BigQuery:
- BigQuery Intelligence gives recommendations for commitment slots, partitioning and clustering, materialized views, and physical storage billing.
- PerfectScale for BigQuery covers cost by reservation, project, and query, slot commitments sized to demand, usage groups with budgets, and anomaly monitors. Its page says automated optimization and insights “find and capture savings continuously” but does not say which BigQuery settings it changes.
- SELECT has a live BigQuery page and a public setup guide. Its 20 July 2026 changelog announced BigQuery in Private Preview, with cost by project, user, job, and query, quantified savings insights, and a view of each reservation’s baseline, autoscale, and maximum slots, spend, and utilization. Its BigQuery page is labeled Early Access. An April 2026 announcement described an automated Storage Optimizer, Reservations Manager, and Query Router, which are not shown as shipped.
Automation and guardrails: published material supports visibility and recommendations. PerfectScale uses broad automation wording, and SELECT’s automated layers are announced or in preview. We did not find shipped per-job routing.
Integration and security: DoiT connects with a service account that reads the billing export; its BigQuery features also query INFORMATION_SCHEMA views. BigQuery Intelligence creates a dataset and audit-log sink in your billing project and runs its processing there. SELECT uses an organization-level custom role of metadata permissions plus project-level job and read-session roles, through a service account you create. DoiT lists SOC 2 Type II, SOC 3, ISO 27001:2022, and GDPR.
Where it is strong: recommendations for commitment slots, partitioning and clustering, materialized views, and physical storage billing; cost by reservation, project, and query; slot commitments sized to demand; and usage budgets and anomaly monitors.
Limitations: BigQuery automation is announced rather than shown as shipped, and SELECT’s own changelog describes BigQuery as Private Preview. The public pages do not map each lever to a specific one of the three products.
Masthead: reservation control inside a data reliability platform
Masthead Data describes itself as an end-to-end data reliability solution for Google Cloud, with BigQuery cost and reservation control as a second pillar next to data observability.
Best for: Google Cloud teams that want data observability and reservation assignment in the same product.
Claimed savings: 15-50% potential per Masthead’s website, and cost per TiB reductions of 20% or more. The site also describes 15%+ savings as guaranteed (the pricing page says Business customers get “at least 10% guaranteed”).
Pricing model: per Google Cloud project per month. The pricing page lists Test Runner at $0 (up to 200 tables), Scale-up at $549 (up to 1,999 tables), and a custom Business plan beyond that, with a 14-day trial.
What it does for BigQuery: it recommends reservation setups and, once you deploy its configuration, assigns workloads to reservations or on-demand through open-source packages for Airflow, dbt, and Dataform. Query suggestions remain recommendations. Its anomaly detection covers data health (freshness, volume, schema); we did not find cost-spike alerting described on the website.
Automation and guardrails: the Airflow configuration is generated from Masthead’s recommendations, and you are responsible for pulling it into your environment. Assignment is automated after that customer-side deployment.
Integration and security: Masthead’s agent runs in your Google Cloud project (VPC) or in Masthead’s project (SaaS), collecting aggregate metrics from logs and metadata. Masthead states it is SOC 2 Type II compliant and works from logs and metadata only, and also lists GDPR and HIPAA.
Where it is strong: reservation assignment as code that fits existing orchestration, alongside data observability.
Limitations: storage optimization is not described on its website. Changes go through a customer-side configuration deployment rather than a closed loop. Per-project pricing scales with the number of projects.
Rabbit: BigQuery optimization across reservations, jobs, and storage
Rabbit is a BigQuery cost and performance optimization platform that tunes slots, reservations, pricing, storage, and SQL together, with recommendations by default and opt-in automation.
Best for: enterprise data, platform, and FinOps teams running BigQuery at scale.
Claimed savings: per Rabbit’s website, the average BigQuery cost reduction is 32%, and customers have saved over $100M combined. Reported enterprise BigQuery savings include 47% at Nordstrom and 52% at Lufthansa Group.
Pricing model: tied to your cloud bill and the automated savings Rabbit helps realize, with a 30-day free trial (extendable) or a free 2-month POC that includes all features. Rabbit is available through Google Cloud Marketplace.
What it does for BigQuery:
- Slots, reservations, and commitments: Reservation Planner and Commitment Planner recommend baselines, maximums, and commitments. When enabled, Max Slot Optimizer adjusts maximum slots, Reservation Optimizer creates Rabbit-managed reservations and moves projects between them and on-demand, and CUD Automation buys commitments in weekly batches you approve.
- Queries and tables: agentic SQL optimization that works through the queries that cost the most and proposes rewrites tested on synthetic data for identical results, plus partitioning, clustering, and materialized view recommendations.
- Storage: sets logical or physical billing per dataset and applies your time-travel retention limits.
- Attribution, alerts, and caps: per-job cost by query, project, label, and reservation; cost and performance alerts; Rabbit Labels for showback and chargeback; and budgets and hard caps by user or label.
- Development workflow: cost-aware code review that catches SQL antipatterns before merge and estimates the cost impact of changed queries; pull requests generated from Rabbit recommendations; a CLI and a coding agent plugin that give coding agents Rabbit’s context on your BigQuery environment.
Automation and guardrails: recommendations are the default, and automation is opt-in, per optimizer, including job-level routing between on-demand and capacity pricing through Airflow and dbt plugins or a BigQuery reverse proxy you deploy yourself. Max Slot Optimizer keeps a safety buffer and can be disabled in the UI, and code changes arrive as pull requests.
Integration and security: Rabbit integrates through Google Cloud APIs, with nothing to install for the base platform, which reads metadata (including SQL text, with literal values masked before storage) and never table rows or query results. Opt-in automation needs broader access: write roles for the optimizers you enable, and plugins or a reverse proxy for job-level routing. Rabbit is SOC 2 Type II audited and ISO/IEC 27001:2022 certified, and acts as a GDPR data processor.
Where it is strong: one platform that treats slots, reservations, per-job pricing, storage, and SQL as interacting levers, so tuning one does not shift waste to another, with enterprise governance (spend caps, chargeback labels, per-job attribution) in the same product.
Limitations: while the base platform needs only get and list permissions to read metadata, automation needs broader access, including write permissions for the optimizers you enable and, for job-level routing, sending the full SQL text and job configuration to Rabbit at job time (not saved, per Rabbit).
See what Rabbit would find in your BigQuery environment:
Book a demo
Revefi: cost optimization and data observability across data platforms
Revefi combines cost and data-quality monitoring across Snowflake, Databricks, and BigQuery, with a free entry plan.
Best for: data teams that want cost and data-quality monitoring across several platforms.
Claimed savings: per Revefi’s website, “Optimize costs by 30-70%”. This figure covers its data platforms, not BigQuery alone.
Pricing model: a free Starter plan for combined annual spend under $50K with self-serve support. The FinOps, DataOps, and All plans are “Based on % of Spend”, by quote, with a free trial and no credit card.
What it does for BigQuery: it monitors slot utilization, flags underused reservations and contention, and recommends rightsizing commitments and autoscaling thresholds. Revefi supplies a script that adjusts maximum thresholds hourly. Usage views break cost down by project, user, table, and query, with anomaly alerts through Slack or email. Per-job routing and storage optimization are not publicly described.
Automation and guardrails: Revefi’s docs describe automated policies and guardrails to prevent overages; detail on the policy engine is limited.
Integration and security: Revefi describes a read-only service account with “metadata access only” and a custom role, and recommends granting reservation permissions on each admin (billing) project. Revefi states that it is SOC 2 Type II, HIPAA, and ISO 27001 certified.
Where it is strong: slot utilization and contention monitoring, rightsizing recommendations for commitments and autoscaling thresholds, and a script that adjusts maximum thresholds hourly.
Limitations: public detail on per-job routing, storage, and the policy engine is limited, and the savings range on its pricing page is not BigQuery-specific.
Ternary: finance-side system of record for technology spend
Ternary describes itself as “Technology Investment Intelligence for the CFO”, a finance-owned ledger across AI, SaaS, cloud, and on-premises spend.
Best for: finance-led governance (CFO, FP&A, FinOps) of technology spend across clouds, with BigQuery as one cost center.
Claimed savings: we did not find a BigQuery-specific figure on Ternary’s website.
Pricing model: not published on Ternary’s website.
What it does for BigQuery: Ternary’s website covers BigQuery slot commitment recommendations, which it says are typically generated only for baseline reserved slot capacity, and lists permissions for reservation, assignment, and commitment visibility. Its Google Cloud integration ingests Cloud Billing export data from BigQuery. Query-level optimization, per-job routing, and storage optimization are not described on its website.
Automation and guardrails: visibility and recommendations. We did not find automated BigQuery changes described.
Integration and security: per the integration guide, Ternary reads the billing export dataset through a tenant service account with BigQuery Data Viewer, plus an organization-level custom role. Its homepage states no custom code, no local agents, and no API changes, SOC 2 Type II and ISO 27001, and availability as SaaS or self-hosted.
Where it is strong: allocation, forecasting, and budgeting across cloud, SaaS, and AI spend; Cloud Billing export ingestion; slot commitment recommendations; and a self-hosted deployment option.
Limitations: BigQuery capabilities beyond slot commitment recommendations and billing export ingestion are not described on its website, and pricing is not published.
Unravel: AI-generated query rewrites with controllable autonomy
Unravel Data’s Arvix AI generates BigQuery query rewrites, validates them against a 30-day workload, and either applies them or sends them for review.
Best for: large data estates that want validated query rewrites with control over autonomy.
Claimed savings: we did not find a BigQuery savings percentage on Unravel’s BigQuery page. Its documentation says most customers see 5-10x return relative to subscription cost, across platforms.
Pricing model: Unravel’s pricing page offers “Subscribe Annually or Pay as You Go” and bases BigQuery pricing on slot consumption per month. Its documentation describes an annual subscription based on compute footprint, with unlimited users and actions. The wording differs between the two.
What it does for BigQuery: it supports on-demand and Editions pricing, recommends a different pricing plan and minimum and maximum slot settings, makes storage recommendations, and provides showback and chargeback reports for compute and storage by project and user.
Automation and guardrails: automation is set per workload: AutoApply (no human step, logged) or Human-in-Loop approval, with a side-by-side diff for each rewrite.
Integration and security: it connects with a service account. Write roles (Resource Editor or Admin) are required only for AutoApply, and a BigQuery User role lets Unravel run queries. Unravel states it collects metadata and never customer data, table contents, or query results, and states SOC 2 Type II, with the report available under NDA.
Where it is strong: validated rewrites with per-workload autonomy and a diff review step, plus coverage across slots, queries, storage, and chargeback.
Limitations: BigQuery outcomes are not published (testimonials are anonymized). AutoApply needs write IAM roles. Its documentation says optimization actions, including query rewrites, are applied through the service account, but does not describe how a rewrite reaches your code (your repository or orchestrator).
How much do BigQuery cost management tools cost?
The models range from no additional charge for named Google controls to percentages of savings or platform spend, per-project plans, subscriptions, and custom quotes. Flat per-project pricing grows with project count; outcome-based pricing depends on how savings are measured. For context, see what a BigQuery slot hour costs and on-demand versus capacity pricing.
How did we choose these tools, and what did we leave out?
To be included, a tool had to cover BigQuery specifically, with a publicly documented capability beyond reading the billing export, target engineering or data teams at enterprise scale, and publish enough primary-source material for us to describe it.
Tools centered on AWS or commitment purchasing, Kubernetes-first tools, Snowflake-first tools, and general FinOps suites without BigQuery depth are not covered here. Open-source tools for BigQuery cost analysis exist, with varied feature sets, but they do not meet the criteria above, because we required publicly documented support, security, and procurement information.
How are these different from general Google Cloud cost management tools?
General Google Cloud cost tools, such as Cloud Billing reports, budgets, and Recommender, look at the whole bill across services: Compute Engine, GKE, storage, networking, and more.
The tools here work inside BigQuery: slots and reservations, query and job behavior, pricing models, and storage billing. They answer a narrower question, and they go deeper on it.
If you are looking for Google Cloud-wide tooling, start with Rabbit’s guide to Google Cloud’s native cost tools and best practices.
What about building it yourself?
Plenty of teams start with INFORMATION_SCHEMA queries, billing export dashboards, and scripts. Our BigQuery cost optimization guide walks through those manual steps, from partitioning to reservations. That approach can become difficult to maintain as projects, reservations, and teams multiply. We cover the trade-offs in Build vs. Buy: the real cost of DIY BigQuery optimization.
What should you ask a vendor before a proof of concept?
- What savings do you claim for BigQuery specifically, and how do you measure them against a baseline?
- How are you paid, and what is the percentage or price at my spend level?
- What do you read (billing export,
INFORMATION_SCHEMA, query text), and what do you write? - Do you sit in my query path? What happens to my queries if you are unavailable?
- How do I roll back a change, and how fast?
- Can I review a recommendation before it applies, and set cost ceilings or approvals?
- How do your recommendations and automation work with the reservations I already have (baseline slots, autoscaling limits, commitments)?
- Do you tune reservations, pricing model, queries, and storage separately or together?
- What does setup require from my team, and who supports it?
- Which certifications apply, with what scope, and where is the report?
How do these tools handle security and access to your data?
The tools differ in what access they need and where they run. Three patterns recur:
- Metadata ingestion. These tools describe reading metadata, with different roles and footprints:
- DoiT: a service account with billing export and
INFORMATION_SCHEMAmetadata. BigQuery Intelligence creates a dataset and audit-log sink in your billing project, and SELECT adds an organization-level custom role plus job-user and read-session roles. - Masthead: an agent in your Google Cloud project (VPC) or in Masthead’s project (SaaS), collecting aggregate metrics from logs and metadata.
- Rabbit: ingests
INFORMATION_SCHEMA(including SQL text, with literal values masked before storage), billing export, Cloud Asset Inventory, and monitoring metrics through Google Cloud APIs, using a role with get and list permissions only. - Revefi: a custom role and service account, with reservation data from permissions it recommends on the admin (billing) project.
- Ternary: the billing export dataset shared with a tenant service account (BigQuery Data Viewer), plus an organization-level custom role.
- Unravel: a service account, with a BigQuery User role so it can run queries.
- DoiT: a service account with billing export and
- In the query or job path. Alvin routes queries from each connected source through its endpoint before they reach BigQuery, and states it does not store or process your underlying BigQuery data. Rabbit’s job-level routing runs through Airflow and dbt plugins or a BigQuery reverse proxy; per its documentation, the plugins and the proxy send the full SQL text and job configuration to the Rabbit API at job time (not saved, per Rabbit), and the dbt adapter falls back to the original configuration if optimization fails.
- Write roles and code-level hooks. Alvin’s billing-model optimization needs the BigQuery Resource Editor role. Rabbit’s Max Slot Optimizer needs
bigquery.reservations.update, and its Reservation Optimizer and Storage Model Optimizer each have their own opt-in roles. Unravel’s AutoApply needs Resource Editor or Admin. Masthead’s open-source packages inject reservation assignment into dbt, Airflow, and Dataform from a configuration you deploy.
All security statements in this post are as stated by each vendor on their websites (for Rabbit, also its documentation and compliance documents).
If you want to see what Rabbit would find in your BigQuery environment, book a demo, or estimate your starting point with the BigQuery Savings Calculator.


