A customer noticed that one of their largest development workspaces had become significantly more expensive. Their monthly reporting showed an unexpected increase in Databricks consumption, but they had no idea what was driving it.
Using LHO, the team started at the executive cost dashboard and immediately drilled into the workspace. A weekly comparison quickly isolated the issue to a single two-week period in June, where weekly spend jumped from roughly $65,000 to more than $100,000, which was an increase of approximately $35,000 per week.
From there, LHO identified exactly which compute services were responsible, revealing that Job Compute was the primary contributor. With a single click into the Optimization Review, the team compared the two weeks side-by-side and instantly surfaced the two jobs responsible for nearly all of the increase.
The investigation continued into the job-level trend lines, where LHO showed the exact configuration changes made before costs spiked. Within minutes, the team knew which engineer had modified the jobs, what configuration had changed, when it changed, and how those changes impacted cost and performance. The owner confirmed the experimental configuration, corrected it, and costs returned to normal.
The entire investigation, from executive dashboard to root cause, took less than five minutes.
Client snapshot
Who
A regional telecommunications provider using Databricks to support large-scale engineering and development workloads.
Industries
Telecommunications
Stakeholders
Engineering leadership, platform engineering, FinOps, and Databricks administrators.
Use Cases
- Databricks cost monitoring
- Cost anomaly detection
- Root cause analysis
- FinOps optimization
- Engineering workload governance
Work summary
LHO helped the customer trace an unexpected Databricks cost increase from executive dashboards to the exact jobs, owners, and configuration changes in under five minutes, while enabling proactive monitoring to prevent future cost overruns.
Minutes to identify the root cause
Weekly cost increase identified
Jobs responsible for the increase
Visibility into affected workloads, owners, and configuration changes
The hidden value
The most important finding wasn’t simply identifying the expensive jobs. LHO had already detected the abnormal spend and generated incidents every day as costs increased. The only missing piece was that no email notifications had been configured for those incidents, so no one saw the alerts.
The team immediately updated their notification rules to send email alerts whenever similar incidents occur, ensuring future anomalies are surfaced on the first day rather than being discovered weeks later during monthly financial reviews.
Business impact
Without LHO, the customer would have continued manually searching through thousands of jobs with little indication of where to begin. Instead, LHO:
- Reduced root cause analysis from days or weeks to under five minutes.
- Identified the exact jobs, owner, and configuration changes responsible.
- Prevented similar cost overruns through automated alerting.
- Demonstrated how engineering experimentation can continue safely with governance and real-time visibility.
In this case, the unnecessary increase was approximately $35,000 per week. Had it continued for just two weeks before discovery, it represented roughly $70,000 in avoidable spend. More importantly, the customer now has proactive monitoring in place to ensure future issues are identified immediately rather than after the monthly bill arrives.
The solution
Using Lakehouse Optimizer (LHO), the customer quickly moved from a high-level view of Databricks spend to the exact workloads responsible for the increase. Starting from the executive cost dashboard, LHO isolated the affected workspace, compared weekly spending trends, and identified Job Compute as the primary driver of the anomaly.
From there, the team drilled into the Optimization Review to compare the impacted weeks side by side. LHO pinpointed the two jobs responsible for nearly all of the increased spend and revealed the configuration changes, job owner, and timeline behind the spike. Within minutes, the team confirmed the experimental configuration, reverted the changes, and restored costs to expected levels.
Because LHO had already generated incidents as the anomaly developed, the customer also enabled automated email notifications to ensure future cost spikes are detected immediately rather than during monthly financial reviews.
