Unravel Data

Overview

Unravel optimises data platform workloads rather than cloud bills. Its subject is jobs, queries, clusters and pipelines on Databricks, Snowflake, BigQuery, EMR and Cloudera, which makes it a different proposition from a general cloud cost tool.

That scope is the first thing to check against your need. It will tell you why a Spark job is expensive and fix it; it does not see your wider infrastructure bill, your SaaS spend or your AI provider spend.

Deployment is deliberately light. A software-as-a-service control plane does the thinking and a small connector per platform collects telemetry, with nothing installed in production and data staying in your own infrastructure.

Access is read-first by default, collecting metadata and telemetry rather than query contents, with write privileges granted separately and explicitly where you want it to act.

Acting is the differentiator. AutoApply executes query rewrites and cluster right-sizing, at three autonomy levels from recommend-and-approve, through auto-approving specific categories, to full automation with guardrails.

The May 2026 release, Arvix AI, is the current story and older descriptions miss it entirely. It is pitched as an agentic engine pre-trained on billions of workloads, with a context graph across compute, workload, data, code, platform and business.

Its safety model is worth noting because it is unusual: every proposed change is tested against thirty days of real workload behaviour before deployment, then monitored and reverted automatically if performance degrades.

Published outcomes are specific rather than vague, claiming 25 to 35% sustained improvement on Snowflake and citing an airline applying 1,500 optimisations in three days for $340,000 of savings.

Its gaps against a general FinOps tool are real: no spend forecasting, no statistical anomaly detection on spend, and no commitment or reservation purchasing advice are published.

Key features and capabilities

The same headings are used for every cost & finops entry, so two tools can be read side by side.

What spend it covers

• Databricks, Snowflake, BigQuery, Amazon EMR, Cloudera and Google Dataproc
• Open-source engines including Spark, Kafka, Hadoop and HBase
• Snowflake coverage reaches warehouses, queries, Snowpark, dynamic tables, Snowpipe, Cortex AI and storage
• No coverage of general cloud infrastructure spend, SaaS spend or external AI provider spend is published
• Optimisation for Iceberg and other open table formats is not published

Attribution and allocation

• Attribution is at workload level, across jobs, queries, clusters, pipelines, datasets and users
• Cloud cost management is a published use case, with a FinOps scorecard in the free health check
• Budget tracking exists as a named capability aimed at preventing overruns
• Showback and chargeback are not published as named features

Finding savings

• Query rewrite with a review gate, showing a side-by-side diff validated against thirty days of workload
• Warehouse and cluster right-sizing, idle resource detection, code and configuration optimisation
• AutoApply executes changes, at three autonomy levels from approval-required to full automation
• Automatic rollback if a deployed change degrades performance
• Published outcomes of 25 to 35% on Snowflake and 35 to 45% in mature environments
• Commitment and reservation purchasing advice is not published

Budgets, forecasts and alerts

• Automated budget tracking to prevent overruns
• Three-level autonomy governance over which actions may apply without approval
• Alerting to Slack, Microsoft Teams, PagerDuty and email
• Spend forecasting and statistical spend anomaly detection are not published

Reporting and sharing

• A FinOps scorecard and cost optimisation recommendation reports
• Delivery into messaging apps, BI dashboards and development environments
• A REST API exposing cost data, efficiency scores, optimisation queues and action history
• API documentation is provided only after onboarding
• A public demo environment exists; no open dataset or price index is published

Integrations

• Messaging and incident tooling through Slack, Teams, PagerDuty and email
• Development and delivery through GitHub, Azure DevOps and continuous integration plugins
• Data quality through Great Expectations
• dbt is not published as a named integration
• No named orchestrator or ticketing integration is published

Where it runs and what it costs

• A hosted control plane with a light connector per platform, nothing installed in production
• The control plane runs on AWS, defaulting to US East with European options available
• An on-premises option exists for Cloudera only, communicating outbound for interface access
• Read-only by default, collecting metadata and telemetry rather than query data, SOC 2 Type II
• Priced on consumption of the underlying platform, in Databricks units, warehouse consumption or slots

Pricing

Price on requestQuote only; free health check

Quote-only, with no figures published. Pricing is billed annually or pay as you go against consumption of the platform underneath, meaning Databricks units, Snowflake warehouse consumption or BigQuery slots, with EMR and Cloudera priced on request. The free entry point is a health check for Databricks or Snowflake, with no card required and a report in two to three business days. The pricing page also says it is free to get started without defining what that covers, and no trial length or free-tier limits are published.

Vendor pricing page →

Demos and videos

• Data pipeline optimisation, the easy way Product demo · YouTube
• Automated budget tracking to prevent overruns Tutorial · YouTube

About Unravel Data

Unravel Data was founded by Kunal Agarwal, its chief executive, and Shivnath Babu, its president and chief technology officer, and is based in San Jose with offices in Bengaluru and Hyderabad; no founding year is published on its own pages. It is private, having raised $107m in total across a $35m Series C in 2019 led by Point72 Ventures and a $50m Series D in September 2022 led by Third Point Ventures, with nothing published since. Named customers include Novartis, Mastercard, Citi, Equifax and NXP. A newsroom check found no funding, acquisition or ownership change in 2025 or 2026.

San Jose, California · unraveldata.com