--- sourceDocument: Australia Enable AI sourceDocumentLink: https://www.servicenow.com/docs/r/australia/intelligent-experiences Release : - australia ft:locale : - en-US ft:publication_title : - Australia Enable AI ft:clusterId : - platai bundleId : - platai workflow : - Platform --- # Session details # Session details {#ariaid-title1} Release version: Australia Updated September 10, 2026 ![](https://www.servicenow.com/docs/portal-asset/ico-clock) 4 minutes to read Summarize ![AI sparkle icon](https://servicenow.com/docs/portal-asset/ai-sparkle-icon) Summarized using AI This content was generated using new OpenAI-powered functionality. Results are provided on an as is basis and are not guaranteed to be accurate or complete. ## Summary of Session details The Session details feature provides ServiceNow customers with deep insights into AI system interactions by showing session duration, token usage, latency, and span counts. It enables you to analyze quality and safety scores at the session, trace, and span levels to identify issues and understand their root causes. Show full answer Show less ## Key Features * **Session Information:** Displays AI system type, session duration (from first interaction to close or timeout), total runtime tokens (cost of running the interaction), latency (model response time), and span count (number of spans in all traces). * **Quality and Safety Scorecards:** Shows overall session scores with detailed breakdowns by relevant metrics, including metric weights, scores, and explanatory notes. You can identify which metrics lower the session's quality or safety score. * **Trace-Level Analysis:** Allows sorting and filtering traces by lowest quality, safety, or highest latency to pinpoint problematic traces. Traces may show N/A scores if no applicable metrics exist at the trace level, indicating the need to add trace-level metrics or review span-level scores. * **Trace Details and Span Timeline:** Visualizes each trace as a waterfall timeline with nested spans. Quality and safety indicators highlight spans with the best or worst scores to focus investigation efficiently. * **Evaluation Scores and Judge's Reasoning:** Provides per-metric scores and detailed judge explanations for each trace and span, helping understand why scores were assigned. * **Input and Output Details:** Shows exactly what data the AI system received and produced at each trace and span, enabling you to determine if issues stem from incoming data or the AI's response. * **Span Details:** Displays execution duration, token consumption (input, output, total), quality, and safety scores to identify performance bottlenecks, excessive context usage, or inappropriate content generation at a granular level. ## Key Outcomes By leveraging Session details, ServiceNow customers can: * Quickly identify which factors most impact the AI system's quality and safety during sessions. * Trace problems down to individual spans or traces with detailed scoring and judge commentary. * Understand whether issues originate from the input data or the system's output to target troubleshooting effectively. * Optimize AI system performance and reliability by addressing slow spans, incorrect tool choices, or inappropriate content generation. * Manage evaluation costs by differentiating between tokens used for running interactions and those used for scoring evaluations. View metric breakdowns, the judge's reasoning, and trace and span data for a specific evaluated session. ## Key benefits {#mon-ai-session-details__key-benefits} * Understand which metrics are pulling a session's quality or safety score up or down by reviewing the scoring breakdown. * Read the judge's reasoning for each metric to understand why a score was assigned. * Identify the exact trace and span where an issue occurred by following the score indicators in the span timeline. * Compare the AI system's input and output at each step to determine whether the issue originated in the data the system received or the response it produced. {#mon-ai-session-details__ul_benefits} Figure 1. Session details ## Viewing session information {#mon-ai-session-details__accessing} The session details page shows the AI system, its type, session duration, and token usage. Duration is measured from the first interaction to session close or timeout (10 minutes of inactivity). Total runtime tokens reflect the cost of running the interaction, not the cost of evaluating it. The header also shows latency and span count for the session. Latency reflects the AI system's language model response time across the session's traces, which is typically much shorter than the session duration since duration also includes think time and tool-call execution. Span count is the number of spans across all traces in the session. ## Quality and Safety score cards {#mon-ai-session-details__score-cards} Review the overall quality and safety of the session by reviewing the score cards. View a breakdown of the score by selecting the side panel icon. The side panel is filtered to show only the metrics that apply to this session, along with each metric's weight, score, and explanatory notes about how the score is calculated. Learn which metrics are impacting the overall score by viewing the lowest scoring metrics list. To learn more about any single score, point to the information icon. The reason notes that the score is an average and points you to open an individual span or trace to read the judge's reasoning for each one. ## Details {#mon-ai-session-details__traces} Identify which trace in the session is responsible for the low score by reviewing the traces list. Sort the Traces list to focus your investigation by selecting Lowest quality score, Lowest safety score, Highest latency, or other criteria from the list. A trace shows N/A for its Quality or Safety score when no metrics in your corresponding metric template apply to that trace. This happens when all metrics in the template are evaluated at the session or span level only, or when the metrics that do apply to the trace level weren't evaluated for that particular trace. An N/A score doesn't indicate a problem with the trace. Rather, it indicates that your template's formula has nothing to calculate at the trace level. To see scores for that trace, either: * Add a trace-level metric to your template. See [Configure an evaluation metric template](https://www.servicenow.com/docs/~6Wl1cgX0wLnQujKladjaw "Adjust the scoring formula for a metric template to change which metrics contribute to quality or safety scores, or to change how much weight each metric carries."). * Review span-level scores by expanding the trace and selecting individual spans in the timeline. {#mon-ai-session-details__ul_ivs_yz5_y3c} ## Trace detail {#mon-ai-session-details__trace-detail} View the span timeline and evaluation data by selecting a trace. Span timeline : Visualizes the trace as a waterfall where each span appears as a horizontal bar showing its duration. Nested spans are indented under their parent. Quality and safety indicators on individual spans highlight which spans scored highest or lowest, so you can focus your investigation on the right span without reviewing each one individually. Evaluation scores : Lists each evaluation metric scored for this trace, along with the score and the judge's reasoning. Select a reason to open the full reasoning in a modal. Use this table to understand which metrics drove the trace's score before selecting individual spans. Details : Shows what the AI system received and what it produced for this trace. Select Open details to view the full content in a modal. Select the Input tab to view the user's request or the data passed to the agent. Select the Output tab to view the agent's response. {#mon-ai-session-details__dl_trace_detail} Navigate between traces without returning to the session detail page by selecting Previous trace or Next trace. ## Span detail {#mon-ai-session-details__span-detail} Identify the root cause of an issue at the most granular level by selecting a span in the timeline and viewing the details. Duration : Time the span took to execute. Unusually high duration can indicate a slow external service call or a timeout that affected the agent's response. Tokens (input, output, total) : Number of tokens consumed by the span. Compare input and output token counts to assess whether the response is proportional to the prompt. A high input token count relative to other spans may indicate the agent is sending excessive context to an LLM call. Quality : Quality score for this span. A low quality score on a specific span helps you pinpoint which step in the agent's reasoning or tool usage produced an incorrect or incomplete result. Safety : Safety score for this span. A low safety score on a specific span helps you isolate which step in the agent's execution produced inappropriate content. {#mon-ai-session-details__dl_span_metrics} Note: The token counts reflect the cost of running the AI system's actual interaction, not the cost of evaluating it. Evaluation scoring consumes separate tokens that aren't displayed here. Learn why a span scored the way it did by reviewing the evaluation metric scores and the judge's reasoning. For example, a Tool choice accuracy score of 54% with reasoning that says "the agent selected a knowledge base search when a CMDB lookup was more appropriate" tells you exactly what decision the agent made incorrectly. Determine whether the issue originated in the input or the output by viewing the Details section. The Input tab shows the data or prompt passed to the span. The Output tab shows what the span returned. For example, a correctly formatted request paired with an incorrect tool call tells you the issue is in the agent's logic, not the incoming data. Select Open details to view the full content when the preview is truncated.