Adata Berna All articles
Data Governance

The Line Item Nobody Audits: Running a Warehouse Spend Reclamation Sprint in 2026

Adata Berna
The Line Item Nobody Audits: Running a Warehouse Spend Reclamation Sprint in 2026

Photo: Iraqi Old Man, CC BY-SA 4.0, via Wikimedia Commons

The warehouse invoice does not announce itself. It arrives as a line inside a cloud bill that is already large, gets approved by someone who is not certain what it contains, and grows by single-digit percentages each month in a way that never quite justifies a conversation.

Then a new finance lead joins, asks a reasonable question about unit economics, and discovers that nobody in the building can explain what the largest analytics cost center actually pays for.

Why This Cost Escapes Review

Most organizations audit headcount, software licenses, and vendor contracts on a fixed calendar. Consumption-based data infrastructure rarely gets the same treatment, because it does not look like a contract. It looks like weather.

The pattern is consistent across mid-market US firms. Storage is inexpensive and therefore ignored. Compute is elastic and therefore approved in advance as a capability rather than a commitment. Query volume is a byproduct of usage and therefore assumed to be good. A dashboard with more users is assumed to be more valuable than one with fewer, which is true often enough to be a dangerous default.

The result is an expense that behaves like a governance failure while being categorized as an IT line item.

Four Species of Waste

Spend forensics is less mysterious than it sounds. In practice, almost all avoidable consumption falls into four recognizable categories.

Zombie pipelines. Jobs that still run on schedule and write to tables no report reads, no analyst queries, and no model consumes. They were built for a project that concluded, or for a stakeholder who has since changed roles. Nothing turns them off, because turning them off requires knowing they exist.

Runaway schedules. A transformation that rebuilds a large table every fifteen minutes when the underlying source changes daily. An hourly job that scans three years of history to produce a snapshot that only ever needs the last ninety days. The logic is usually defensible in isolation; the cadence is not.

Duplicated tables. Five derivatives of the same customer entity, each created by a different team at a different time, each refreshed independently. Storage is the smaller cost. The compute required to maintain all five is the larger one.

Unused dashboards. Reports built for a launch, screenshotted once, and abandoned. Each may refresh on schedule, each may query live data, and each may have a last-viewed date older than the current fiscal year.

Step One: Reconstruct the Last Ninety Days

Start with query history rather than with the architecture diagram. Pull ninety days of activity and group it by user and service account, by warehouse size, by bytes scanned, and by destination table.

Three lists emerge almost immediately. The first is your top consumers by cost, which is usually far shorter than expected — a handful of queries frequently dominates. The second is your scheduled jobs, with their cadence and their output targets. The third is your dormant objects: tables written to but never read, and dashboards that have not been opened.

If the platform supports tagging, check how much of the spend carries a cost-allocation tag. Untagged spend is unowned spend, and unowned spend is where the waste concentrates.

Step Two: Attach Every Dollar to a Name

Build a single working table with five columns: object, monthly cost, technical owner, business consumer, and the decision the object supports. The last two columns do the real work. They force the question that consumption reports never ask — not whether the object exists, but whether anyone would notice if it did not.

Expect resistance here, and expect it to be informative. When no business consumer can be named, that is the finding. When three teams each claim to depend on a dashboard and none can describe the last decision it informed, that is also the finding.

The Six-Week Reclamation Sprint

Week one — instrument. Establish cost tagging by team and by data domain. Where native tagging is unavailable, maintain a mapping table manually for the duration of the sprint.

Week two — inventory. Extract the object list, refresh schedules, and usage telemetry into one place. Do not analyze yet; the goal is coverage.

Week three — triage. Review every object with its named owner and assign one of four verdicts: keep, fix, merge, or retire. Keep is a legitimate verdict and should be used freely — the exercise is not an austerity campaign.

Week four — act. Pause zombie schedules. Archive the DDL for duplicate tables before dropping them, so reversal remains possible. Retire dashboards after notifying their owners in writing and waiting a defined period.

Week five — right-size. Adjust warehouse sizing to actual workload rather than peak assumption. Set query timeouts. Cap the maximum size a scheduled job may request. Move long-running analytical work off peak hours.

Week six — publish and protect. Report the recovered spend, but also publish the guardrails that keep it recovered. Organizations that skip this step reliably rediscover the same costs within two quarters.

Guardrails Worth Keeping

Four controls prevent the pattern from reasserting itself. Show the estimated monthly cost on every dashboard in the catalog. Require a named consumer and a scheduled review date before a new pipeline is approved. Set default expiry on dashboards that go unviewed for a defined period. And put consumption on a standing thirty-minute agenda item at whatever committee already owns data governance.

The Point of the Exercise

Most organizations running this for the first time find that a meaningful share of consumption — frequently a fifth or more — attaches to objects no one is actively using. That figure is useful, but it is not the real deliverable.

The real deliverable is the ownership map. Once every table and every dashboard has a name attached to it, cost becomes a decision instead of an accident. That shift is what makes this a governance problem rather than an IT problem, and it is why the firms that get it right spend less while building more.

All Articles

Related Articles

Before the Term Sheet: Auditing the Target’s Data Estate in US Mid-Market Acquisitions

Before the Term Sheet: Auditing the Target’s Data Estate in US Mid-Market Acquisitions

Ninety Days to Cleaner Data: A Quarter-Long Sprint That Pays Off in Q1 Planning

Ninety Days to Cleaner Data: A Quarter-Long Sprint That Pays Off in Q1 Planning

Dirty Data, Drained Budgets: The Billion-Dollar Problem Most US Businesses Refuse to Acknowledge

Dirty Data, Drained Budgets: The Billion-Dollar Problem Most US Businesses Refuse to Acknowledge