FinOps Program That Cut a SaaS Cloud Bill 38% ($860K a Year)

The Company

ClearBench Analytics is a 250-employee B2B SaaS company whose platform runs large-scale analytical workloads for enterprise customers: ingestion pipelines, query processing, reporting and the always-on infrastructure that serves dashboards to thousands of business users. Its customers buy analytics the way they buy insurance — continuously, and with expectations of constant availability and fast performance — which means ClearBench's engineering organization runs a substantial cloud estate: compute fleets for query processing, object storage for historical data, managed databases, and the development and staging environments every healthy engineering team needs.

As a venture-backed company, ClearBench lives under a constraint every SaaS leadership team will recognize: gross margin. The margin commitments made in its most recent funding round are reviewed by the board every quarter, and infrastructure is one of the largest cost lines against revenue. That made the cloud bill not merely an engineering topic but a company-level financial instrument — and in the quarter this story begins, that instrument was out of tune. Flexera's State of the Cloud reports that organizations self-estimate around 28% of their cloud spend goes to waste, a figure that has stayed stubbornly stable for years, and Gartner has forecast worldwide public cloud end-user spending to reportedly surpass $1 trillion by 2027, elevating cost discipline to a board-level topic. ClearBench was about to become a case study in why.

Where Things Stood

ClearBench's annual cloud bill had climbed to $2.26M, growing three times faster than revenue, while gross margin drifted toward the red line the funding round had promised investors. The deeper problem was that nobody owned the number. Fewer than 40% of resources carried cost-allocation tags, so finance could see the total invoice but could not attribute spend to teams, products or environments. Development and staging environments ran around the clock whether anyone was using them or not. Engineers provisioned for worst-case peaks — a habit left over from capacity scares early in the company's history — and never looked back at utilization afterward, so oversized instances accumulated like sediment across the estate.

The organizational response made things worse before it made them better. Finance, seeing the trend line, proposed blanket budget freezes, which engineering correctly predicted would slow the product roadmap: capacity requests would queue behind approval workflows, and every new feature would inherit a procurement delay. The two functions were heading toward a structural standoff — finance holding the invoice it could not explain, engineering holding the systems that generated it. Into this stalemate, the CFO issued a challenge to the platform team: one quarter to prove that costs could fall without freezing anyone's roadmap, or spending controls would move to finance permanently. It was, in effect, a deadline with an org chart attached.

The Challenge

The challenge was to cut a seven-figure cloud bill materially, within one quarter, while the product shipped faster — not slower — than before. The FinOps Foundation's State of FinOps surveys report that reducing waste or unused resources is practitioners' top priority, ahead of rate reduction and forecasting accuracy, and ClearBench's diagnosis matched that pattern precisely: the estate had a waste problem first and a rate problem second. But knowing waste exists and attributing it are different problems, and with fewer than 40% of resources tagged, the company could not yet even answer the question "whose spending is this?"

Three constraints shaped the solution. First, accountability could not be centralized: a central gatekeeper approving every capacity request would recreate the budget-freeze outcome by another name, so the cost controls had to live with the teams generating the costs. Second, the savings had to be fast enough to matter within the CFO's one-quarter deadline, which meant the program needed early, visible wins rather than a six-month architectural crusade. Third, the savings had to hold — one-time cleanups fade, and bills creep back the moment attention moves on — so the program needed a durable operating rhythm, not a sprint. The platform team accepted the deadline and, with Emerson Technology's help, framed the work explicitly as FinOps: a practice, not a purge.

Discovery and Diagnosis

Emerson Technology began with a two-week cost and utilization diagnostic built on the billing data itself. Cost and Usage Report data was landed in Athena and visualized in QuickSight, then sliced by service, account, environment and team. The untagged majority was reconstructed retroactively from resource metadata so the team could see, at least approximately, where the money actually went. The findings were clarifying. A substantial share of compute spend was oversized instances running below 20% utilization. Non-production environments consumed meaningful money during nights and weekends when no engineer was working. Cold analytics data — query history, old exports, replicated backups — sat in standard storage tiers years after anyone had last read it. Orphaned snapshots and load balancers, residue from decommissioned projects, billed quietly every month.

The diagnostic also mapped the human system, because the numbers alone could not explain why waste persisted. Engineers had no visibility into their own team's spend — the invoice went to finance, the dashboards went nowhere — so there was no feedback loop connecting a provisioning decision to its monthly consequence. Architecture reviews evaluated performance and reliability but never cost, which meant expensive decisions were made implicitly at design time and discovered later in the invoice. The diagnosis reframed the problem from "engineers are wasteful" to "engineers are blind" — and that reframing determined the solution's shape: give every team its own spend picture, give every design a cost dimension, and let accountability do the work that mandates never could.

The Solution

We ran the effort as a FinOps program rather than a cost-cutting sprint, standing up a weekly cost review with engineering leads, finance and product in the same room. The operating principle throughout: visibility first, quick wins second, commitments last.

Visibility before cuts

Cost and Usage Report data in Athena and QuickSight gave each team a tagged, per-service view of its own spend, and a lightweight showback report landed in Slack every Monday with week-over-week deltas and drill-down links. Tagging was pushed hard in the first weeks — every resource mapped to a team, product and environment — because you cannot hold anyone accountable for costs nobody can attribute. The weekly cross-functional review put live spend data on screen with engineering, finance and product together, which ended the finance-versus-engineering blame cycle faster than any policy could: both sides were finally arguing from the same numbers.

The engineering quick wins

The first wave of cuts came entirely from hygiene, not architecture. Oversized instances were rightsized using utilization data collected over several weeks, with a safe-list for workloads whose peaks justified their size. Dev and staging environments were scheduled to shut down nights and weekends, instantly eliminating the largest slice of non-production waste. Cold analytics data was tiered to S3 Infrequent Access and Glacier according to access patterns, and a cleanup pass deleted orphaned snapshots, unattached volumes and forgotten load balancers. Each change was small and reversible; together they bent the curve within the first month, which bought the program the credibility and patience it needed for the slower work ahead.

Commitments and unit cost

Savings plans and reserved instances were layered on only after rightsizing stabilized the usage baseline — committing early would have locked in today's waste at a discount for one to three years. With the footprint stabilized, commitments covered the predictable core of the estate at materially better rates. Two durable mechanisms then made the savings structural: every proposed architectural change now includes a cost estimate alongside its performance analysis, making cost a first-class engineering decision at design time; and the team adopted unit-cost metrics, tracking spend per active customer rather than totals alone, so efficiency improvements remain visible even while absolute spend grows with the business. The weekly FinOps review is now how ClearBench evaluates every new service it adopts.

The First 90 Days

Week one established the instrumentation: billing data in Athena, the QuickSight views built, and the first retrospective tagging pass completed so the baseline could be attributed. Weeks two through four drove tagging past the point of usefulness — tag coverage climbed toward 90% of billable resources — and launched the Monday showback report into Slack. The first weekly cross-functional review convened in week three and set the ground rules that held for the rest of the program: no team's spend was policed centrally, but every team's trend was visible to everyone.

Weeks five through eight were the quick-win harvest. Rightsizing changes shipped behind careful observation windows, the dev and staging shutdown schedules went live, storage tiering moved petabytes of cold history to cheaper classes, and the orphaned-resource cleanup ran to completion. The invoice that arrived at the end of week eight showed the first unmistakable bend in the curve, and the showback report that Monday was, by engineering's own accounts, the most-opened message the platform team had ever sent — teams could finally see their own names on the graph. Weeks nine through twelve added the durable machinery: the commitment strategy was executed once utilization stabilized, unit-cost dashboards went live, cost estimates became a required section in design reviews, and the CFO's one-quarter deadline was met with room to spare. The release-frequency data from those same twelve weeks carried the counterintuitive headline — deployments were up, not down, because engineers stopped waiting on capacity approvals once they could self-serve and see the consequences.

The Results

Cloud spend fell 38% — $860K annualized — within two quarters, and the savings have held because accountability lives with the teams generating the costs rather than with a central gatekeeper. Development velocity actually improved: release frequency rose 20% as engineers stopped waiting on procurement for capacity and finance stopped drafting budget freezes. Tag coverage reached 98% of billable resources, unit cost per active customer fell for the first time in two years, and the weekly FinOps review is now how ClearBench evaluates every new service it adopts. The CFO's deadline was met early, and the spending controls stayed with engineering — where the people making the decisions can see their consequences.

The relationship between finance and engineering changed as much as the invoice. Budget conversations are now grounded in per-team data rather than totals and suspicion, and the annual planning cycle inherits a cost model built from observed unit economics instead of negotiation. Engineers approach capacity with the same professionalism they bring to performance — provisioning what the workload needs, closing down what it does not, and costing new designs before they are built. Waste, which Flexera's surveys suggest runs near 28% across the industry even as organizations believe they are efficient, is now something ClearBench measures weekly rather than estimates annually.

What Your Business Should Do

ClearBench's program is deliberately portable, and its order of operations is the whole trick. Tag resources before cutting anything: push allocation tagging past 90%, mapping every resource to a team, product and environment, because you cannot hold anyone accountable for costs nobody can attribute. Run one weekly cross-functional review with engineering leads, finance and product in the same room with live spend data on screen — shared numbers end the finance-versus-engineering blame cycle faster than any policy. Harvest quick wins first: rightsize oversized instances, schedule idle dev environments to shut down nights and weekends, and delete orphaned volumes and load balancers; quick wins buy patience for the slower architectural changes. Tier storage by access pattern, moving cold logs, backups and analytics history to infrequent-access and archive classes — storage often hides 20-30% of a bill in data nobody has read for a year.

Then make the savings permanent rather than episodic. Buy commitments after usage stabilizes, layering savings plans and reserved instances only once rightsizing is done, because committing early locks in today's waste at a discount for one to three years. Adopt unit-cost metrics — spend per customer, per transaction or per query — since unit-cost trends prove efficiency is improving even while absolute spend grows with the business. Add cost estimates to design reviews, requiring a rough monthly figure alongside every significant architecture or feature proposal so expensive decisions are caught before they reach production. And show costs back to owning teams weekly, with deltas and drill-downs, because visible numbers change engineering behavior without a single budget mandate. As you plan, interrogate your own estate with the questions this program raised: can finance attribute last month's spend to specific teams today, what percentage of your environments run nights and weekends unused, whether your cost per customer is trending up or down, and which architectural decision from two years ago now costs you the most each month. Work through these eight steps with your team using the action checklist in this practice book, and revisit the reflection questions each quarter — cloud costs are a practice, and practices survive only on a review cadence.