Telemetry Insights

Value Page

Tursio provides a value page with metrics demonstrating how the platform delivers measurable cost and time savings and translates efficiency into tangible business value. The Value tab appears only when ROI tracking is enabled for your deployment. This dashboard allows users to adjust the date range (the last 30 days by default, or a custom period), providing flexibility in analyzing performance across different reporting windows.

Cost savings. The percentage of token cost saved with Tursio compared to the default, where the default assumes each query is answered at the baseline token cost per query. Queries answered from cache make no LLM call and cost nothing. This value can be negative if Tursio's actual token spend exceeds the baseline.

Time savings. The percentage of query time saved with Tursio compared to the default, where the default assumes each query is authored by hand at the average time per query.

Business impact. The total savings: token cost saved plus time saved, valued at the average human expert hourly rate.

Each card also shows the change from the previous period.

Customize metrics. Users with update permission can click Customize to open the Customize metrics modal and set the inputs used in these calculations:

• Avg time/query(min) — the average time to author a query by hand.
• Avg. human expert cost ($/hour) — the hourly rate used to value time saved.
• Baseline token cost ($/query) — the default token cost per query.

Trends. The supporting charts compare Default with With Tursio for each day in the reporting period:

• Cost savings: Token cost per day.
• Time savings: Time taken in hours per day.

Hover over a bar to see its values.

The combination of metrics and trend charts underscores a clear story: as adoption grows, productivity gains and business impact scale proportionally, delivering measurable ROI for enterprises.

Metrics Page

The metrics dashboard provides visibility into the system's workload, performance, and accuracy.

Users can select a date range (the last 30 days by default, or a custom period) for flexible performance analysis across different reporting windows.

Workload. The total number of queries in the selected period.

Performance. The average time taken to process a request end to end.

Accuracy. The percentage of queries with positive or no feedback — that is, not marked negative by users. Unrated queries count as accepted, so accuracy reads 100% until someone marks an answer negative.

Each card also shows the change from the previous period.

Trends. The charts provide deeper visibility into system workload and performance:

• Workload: Shows the volume of queries handled daily, split into unique and recurring queries. Spikes indicate peak demand periods.
• Latency: Displays the average processing time per day. Variations highlight how workload or optimization measures impact speed.

Hover over a bar to see its values.

Query History

The Query history tab in Insights provides full visibility into previously executed queries, enabling accurate tracking, dataset-level filtering, feedback monitoring, and seamless navigation back to search workflows.

By default, Query history displays all queries executed within the last 30 days, including the query text, time, user, and associated feedback (positive, negative, or none). When All Databases is selected, a Database column is added, and each multi-database run appears as one row that expands to show the individual searches it ran. The view is dataset-aware — when a user switches datasets, the history automatically refreshes to show only queries relevant to the selected dataset, with no stale data carried over.

Users can refine results using a date range filter, which reloads data when changes are applied and strictly limits results to the selected timeframe. Each query entry has a ⋮ menu: View query opens the query on Search Home in a new tab and runs it to display results, and Inspect plan shows the query's plan when one is available.

For users who can edit the data model, an Edit data model option is also available on queries from datasets other than Cassandra and DynamoDB. Selecting this option opens the query on Search Home, auto-runs it, scrolls to the Data Model section, and opens it directly in edit mode.

The system also supports structured feedback management. Negative feedback entries display alert indicators only when valid and not previously ignored. Users can ignore feedback individually or in bulk within a selected date range. Once ignored, alert indicators are removed and the interface refreshes to reflect the updated state.

Together, these capabilities ensure reliable query tracking, precise filtering, structured feedback governance, and tight integration with search and data model workflows — supporting operational transparency, streamlined analysis, and scalable enterprise usage.