FinOps has matured fast. According to the State of FinOps 2026, 78% of teams now report to the CTO or CIO. The discipline has moved up the org chart, embedded into engineering workflows, and stretched its scope well beyond compute into SaaS, licensing, and AI spend.
None of that solved the data platform problem.
The line item keeps growing. 30 to 50% of total cloud spend at most enterprises, with no real lever to pull. Not because FinOps teams aren't doing their jobs. Because the tools they have weren't designed to reach inside Databricks, Snowflake, or BigQuery. Arvix AI by Unravel is how that changes.
Two structural problems explain the gap. Unravel solves both.
- 30–50% of the cloud bill — and growing. The data platform line item.
- 78% report to CTO / CIO — FinOps has shifted up, left, and out.
- <10% of native recommendations ever get implemented — the action was always the bottleneck.
Problem One: You Can't See Inside the Data Platform
FinOps platforms see infrastructure beautifully. Accounts, services, resources, reserved instances, account-level spend. That is what they were built to do.
The job, query, warehouse, and cluster level? That was never in scope. Which is a problem, because that is exactly where the cost gets made.
A misconfigured autoscaling policy on one Databricks pipeline. A Snowflake query that runs 30 minutes when it should run five, ten times an hour. A BigQuery storage tier that's been hot for two years and should have moved to cold last quarter. None of that shows up in a FinOps dashboard with any actionable detail. It just shows up as a bigger bill.
"Dashboards are table stakes of yesterday. Reactive. You have to move to proactive, real-time, automation."— Practitioner, State of FinOps 2026
Problem Two: Even If You Could See It, You Still Can't Optimize or Govern It
This one gets less airtime. It's the harder problem.
The standard FinOps operating model is a small central team (fewer than 20 people, even at organizations spending hundreds of millions on cloud) plus federated champions scattered across business units. That team is supposed to govern what's happening on the data platform side.
What are they actually governing? Hundreds of engineers and data scientists. Thousands of queries and pipelines every week. Code that gets written, pushed, refactored, deprecated, and redeployed at a pace no human review process keeps up with. The math doesn't work.
You can't tap every shoulder. You can't review every PR. You cannot email a data engineer in São Paulo about a $400 query that runs nightly when 11,000 other queries are running tonight. Manual policing at data platform scale isn't a strategy. It's a fantasy.
The FinOps Foundation sees this clearly. Jonathan Morley's March 2026 piece on agentic use cases describes the coming shift as moving "from doing the work to orchestrating the workers." That framing is exactly right. Human-paced optimization doesn't work here.
Granularity: How Arvix AI Shows Where Cost Is Actually Generated
Unravel was built to see what FinOps platforms can't.
Every query, job, pipeline, cluster, and warehouse across Databricks, Snowflake, and BigQuery mapped into one picture, attributed to the team, project, and budget that owns it. The questions that have always been just out of reach suddenly have answers:
- Which 20 queries drove last week's spend anomaly?
- Which pipeline is behind the 3.4× cost spike on the analytics cluster?
- Which Snowflake warehouse is auto-scaling into next month's overage?
- Which BigQuery tables are sitting hot when they should be cold?
Attribution is the prerequisite for action. The Context Graph that powers these views is also what makes safe autonomous optimization possible. You can't touch what you can't accurately account for. Unravel does both in the same platform, across all three data platforms.
That solves problem one.
Optimization and Governance: Automatic, Validated, Continuous
Visibility is necessary. Not sufficient.
Most FinOps leaders have seen plenty of tools that surface the problem and stop there. Arvix AI does not generate a recommendation and wait for a data engineer to get around to it. The sequence looks like this:
- It rewrites the query.
- It tunes the cluster configuration.
- It validates the change against real workload behavior before anything touches production.
- It applies the fix.
- It monitors post-apply and reverts automatically if anything drifts.
Recommendation engine versus optimization engine. One creates tickets. The other closes them.
Governance shifts from manual policing to continuous autonomous enforcement. Full audit trails. Graduated controls based on workload criticality. Native recommendations from Databricks, Snowflake, and BigQuery exist, but the data is clear: fewer than 10% ever get implemented. The recommendation was never what was missing. The action was.
"But I'm Not Ready to Let an AI Touch Production"
You shouldn't be. Not without the right controls in place.
The FinOps Foundation has been explicit about the trust gap. Most organizations aren't comfortable with agents making production changes without human sign-off, and that caution is warranted. Arvix AI was designed for exactly that reality. Every workload carries its own autonomy setting, set by your team.
Where autoapply makes sense:
- Non-critical workloads
- Dev and staging
- Cost optimizations with no logic change
Where human-in-the-loop is required:
- Production
- Customer-facing pipelines
- Regulated data
- ML training
Every change gets validated against real execution before it ships. Every action is logged. Post-apply watchdog monitoring runs continuously. If behavior drifts from expected, it reverts.
- Zero production incidents — from Arvix-AI-applied optimizations, across every enterprise customer running it today.
- 70%+ of actions on autoapply — trust gets earned per workload, not granted up front.
The Bigger Payoff for FinOps Leaders
One line from the State of FinOps 2026 captures the pressure most FinOps leaders are operating under right now:
"Many organizations report being asked to self-fund AI investments through optimization savings."— State of FinOps 2026
The CFO wants AI funded. The CTO wants it funded fast. Optimization is supposed to create the headroom. But as one practitioner put it, "we have hit the 'big rocks' of waste and now face a high volume of smaller opportunities that require more effort to capture." The easy savings are gone. What's left requires a different kind of tool.
The largest remaining pool sits inside the data platform layer. It's also the fastest-growing slice of the bill.
- 58% Databricks spend cut — top-3 global airline · 30+ recommendations auto-applied per week.
- $340K realized in 3 days — same airline. Savings, not projections.
- 70% cloud data costs cut — Fortune 100 logistics company · in six months.
That is the headroom your AI roadmap needs. It has been sitting in the one part of the cloud bill FinOps could not reach.
Until Now
FinOps was built to govern cloud spend. Across infrastructure, SaaS, licensing, and increasingly AI, the discipline has done that well. The data platform layer was the exception: the fastest-growing slice that was also the least reachable.
That ends with Arvix AI by Unravel.
Find Us at FinOps X
If you're at FinOps X in San Diego, June 8–11, find the Unravel team for a live demo of Arvix AI running against real data platform workloads. We'll show you the answers, and then we'll show you the fix.
Autonomous data platform optimization for Databricks, Snowflake, and BigQuery across AWS, Azure, and Google Cloud. SOC 2 Type II compliant.