Anatoly Mikhaylov and Nick Hefty from Zendesk explain their cost optimization efforts around observability data using Datadog. As Zendesk expanded its use of Datadog for metrics, logs, and application performance monitoring (APM), costs began to rise significantly. To manage these expenses while maintaining visibility for engineers, they initiated a cost-optimization project based on the 80/20 rule. They focused on identifying the most valuable traces, logs, and metrics that contributed to the majority of costs and benefits. They created customized dashboards to visualize usage and costs, linking billing figures to specific metrics, traces, and logs. This helped them pinpoint where to make cuts without sacrificing critical observability. A major change involved switching to single-span ingestion for tracing. By enriching root spans with key performance metrics, they reduced the overall volume of trace data while still providing engineers with the insights needed for troubleshooting. One practical example highlighted how they addressed a database bottleneck. By leveraging custom facets in their root spans, they could filter and isolate slow requests tied to a specific database cluster and account. This allowed them to identify expensive queries caused by large datasets, which they addressed by offloading to Elasticsearch rather than optimizing the relational database. The changes led to a notable drop in database time for those customers, confirming the effectiveness of their approach while keeping costs manageable.