Skip to content
English
Contact us
  • There are no suggestions because the search field is empty.

User Guide: AI Reporting Agent

Overview

A new AI Reporting Agent supports end users through chat — on the Employee Center, other portals, or inside the mobile app via Now Assist — when they need to report an incident or get help. As the user describes their issue in the chat, the agent assesses how urgent and important it is. If the agent detects an emergency, it advises the user to call 911 immediately. If there is no emergency, the agent creates an incident from the conversation itself, without requiring the user to fill out forms or fields, and hands it off to the existing AI Categorization Agent for categorization.

What's New

  • Chat-based reporting — users can report an incident or ask for help directly through chat on the Employee Center, other portals, or inside the mobile app (powered by Now Assist).
  • Emergency detection — the agent evaluates urgency and importance as the user types. When it recognizes an emergency (based on protocols defined in Knowledge Base articles), it immediately advises the user to call 911 instead of collecting a report.
  • Automatic incident creation — for non-emergency issues, the agent creates an incident directly from the conversation. There’s no form to fill out — the agent grabs the relevant details itself, asking brief follow-up questions only if it needs more information.
  • Natural-language understanding — custom skills let the agent interpret everyday phrasing (for example, “five minutes ago”) and convert it into structured data, such as the event time in ServiceNow date format, so users never need to type exact values.
  • Chained categorization — once an incident is created, the existing AI Categorization Agent runs automatically to categorize it.
  • Confidence-based auto-apply — the Categorization Agent applies its suggested category automatically only when its confidence is high (for example, 90% or above); otherwise, it posts the category as a suggestion in the activity/work notes for a human to confirm.

              How It Works

              1. The user opens the chat on the Employee Center, another portal, or the mobile app, and describes their issue in their own words.
              2. The agent evaluates the message for urgency and importance.
              3. If it’s an emergency, the agent advises the user to call 911 immediately — no incident is submitted for this case.
              4. If it isn’t an emergency, the agent asks brief follow-up questions in natural language if it needs more details (for example, when the issue happened).
              5. The agent creates an incident using the details gathered from the conversation, including a short description and the event time.
              6. The AI Categorization Agent runs automatically on the new incident, applying or suggesting a category and subcategory based on its confidence level.

                          Example Scenarios

                          Case 1: Emergency — Advised to Call 911

                          A user types "there’s a man in the lobby with a knife" into the chat. The agent identifies this as an emergency based on the security protocols defined in its Knowledge Base and advises the user to call 911 immediately. Because this is an emergency, no incident report is created — the situation is handled entirely through the 911 advisory.

                          Case 2: Non-Emergency — Incident Created and Categorized

                          A user reports a broken window on the second floor. Since this isn’t an emergency, the agent asks a follow-up question ("what time did this happen?") and accepts a natural-language answer like "five minutes ago." The agent creates an incident with a short description built from the conversation, sets the event type, and parses the event time into the correct ServiceNow date/time field. The AI Categorization Agent then runs on the new incident; because its confidence in the suggested category isn’t high enough (below the 90% auto-apply threshold), it posts the category as a suggestion in the activity flow rather than applying it automatically.

                          For Admins: Keeping the Agent Accurate

                          • Maintain the Knowledge Base articles that define emergency protocols, since these determine when the agent advises a user to call 911 instead of filing a report.
                          • Review and tune the confidence threshold (currently 90%) that controls when the Categorization Agent applies a category automatically versus only suggesting it.
                          • Review custom skills used for natural-language parsing (such as relative time phrases) periodically to make sure they continue to map correctly to structured fields.