SoftTech AI Registry
Atlassian · Jira Service Management

Virtual Service Agent

Atlassian's customer-facing conversational agent in Jira Service Management. It answers customer questions from linked knowledge base spaces using generative AI (AI answers), recognises configured intents with a machine learning model trained on the space's own intents and training phrases, and runs turn-by-turn conversation flows whose steps can create Jira work items, override request type and field values, send web requests to external servers, run Jira automation flows and trigger Workato recipes or Workday actions.

Recorded characteristics

Function
Atlassian documents the virtual service agent as operating in customer help channels for a Jira Service Management service space. It uses machine learning to recognise questions and requests and can automate repetitive tasks using turn-by-turn conversation flows, AI answers connected to a knowledge base, or a combination of both. AI answers uses generative artificial intelligence to search across linked knowledge base spaces and summarise that information in response to customer questions without an intent being created. When a customer sends their first message, the agent first tries to match an intent; if no intent matches with high confidence it attempts an AI answer. The match intent standard flow may confirm a detected intent with the customer, offer several possible intents, answer through AI answers, ask the customer to rephrase, or escalate. Conversation flows are built from documented step types: offer choices (1-10 options, each creating a branch), send message, ask for information (collects Date, Paragraph, Short text or Single-select custom field values), change request type and fields (invisible to the customer, overrides the default request type and can set field values for work items created later in the conversation), send web request (sends a request to a server and evaluates success conditions such as status ~ 2??), run automation flow (triggers a Jira automation flow containing the Conversation flow step trigger, in the background), trigger Workato recipe (background) and trigger Workday action (background, for example submitting a leave request built from collected fields). The agent creates work items in Jira Service Management on behalf of customers using the configured virtual service agent default request type unless a change request type and fields step overrides it, and offers escalation to a human agent when a conversation is unresolved. Standard flows include auto-close after five minutes of abandonment and a resolve-with-CSAT ending.
Data access
Atlassian establishes that the agent reads customer messages in its channels and searches linked knowledge base spaces to generate AI answers, and that the articles used reflect the customer's own knowledge base permissions, except in Slack where only articles set to All logged-in users are used. Its machine learning model is trained on intents and training phrases created by the project admin, generated from the project's historical data, or suggested by Atlassian and adopted by the project admin. Conversation flows read the customer's own responses collected in ask for information steps, and send web request response variables may be reused in later steps. Conversation records (including the source articles used for AI answers) are stored for 30 days for analytics and may be reviewed by human administrators; Atlassian warns customers not to enter confidential information. In email, the AI answer and the customer's feedback are added as comments on the created work item.
Actions
Can take actions
External actions
Yes
Human confirmation
Conditional
Permission basis
Mixed
Administrative control
Documented controls: a space admin sets up the virtual service agent and activates or deactivates AI answers (Space settings > Channels & self service > Virtual service agent > Manage in Studio > Knowledge), with activation and deactivation taking effect immediately across all connected channels; Rovo must be enabled and a knowledge base space linked and set to All logged-in users; an organisation admin can turn the virtual service agent off; a project admin sets or changes the default request type (Studio > Identity > Jira Service Management settings), which must not be hidden from the portal and must have no required fields because the agent cannot fill required fields; project admins create, train, activate and deactivate intents and build conversation flows; a project admin sets up Slack agent and request channels and can switch a Slack request channel between virtual service agent, automatic work item creation and manual work item creation; a space administrator activates the agent in Microsoft Teams and in email (which additionally requires Outgoing Mail enabled); a site or organisation admin activates it in the help center; Atlassian recommends testing the agent before activating it for customers; a Conversations page in Studio logs conversations from the portal, help center and Slack with Action (Matched, AI answered, Unassisted), Resolution (Escalated, Resolved, Closed) and CSAT columns and filters; automation flows triggered from a conversation have their own audit log recording the trigger, result and actions performed.
Default state
Disabled
Availability
Atlassian documents the virtual service agent as included with Jira Service Management Premium and Enterprise plans. It is stated as not available in the Atlassian Government environment. Channels documented are the portal, help center, widget, email, Slack and Microsoft Teams, with the note that the agent can be used in Microsoft Teams or Slack but not both; Atlassian distinguishes generally available channels from channels in beta for usage counting. In email the agent responds using AI answers only. Additional languages can be added to a service space so customers are served in their preferred language.
Licensing
Atlassian states the virtual service agent is included as part of all Premium and Enterprise plans, with usage measured in assisted conversations: 1,000 per month on monthly subscriptions or 12,000 per year on annual subscriptions included, and an Extra assisted conversations add-on billed on usage (per-conversation for monthly, tiered for annual) beyond that allowance. Only generally available channels count towards assisted conversations. Usage limits for the virtual service agent took effect on 16 October 2024. AI answers additionally requires Rovo to be enabled.
External model or provider
Not established
Limitations and uncertainty
Recorded boundaries and open questions: (1) Atlassian does not identify the model or provider behind AI answers; the intent model is described only as a machine learning model tailored to the organisation. (2) The action classification rests on documented persistent effects - work item creation using the default or overridden request type, field values set by the change request type and fields step, and background execution of Jira automation flows - not on the agent resolving or transitioning existing work items; no Jira status transition authority is documented, and the "Resolved" state in the Conversations page describes a conversation outcome, not a work item transition. (3) External action capability is recorded as yes because the send web request step sends requests to an external server, and Workato recipe and Workday action steps execute in external systems; Atlassian documents these as configured by administrators, not chosen by the agent. (4) Human confirmation is conditional: the match intent standard flow can ask the customer to confirm a detected intent, and ask for information steps wait for a customer response, but Atlassian states customers will not notice when an automation flow, Workato recipe or Workday action is run - these happen in the background with no confirmation step documented. (5) Permission basis is mixed: AI answers reflect the customer's knowledge base permissions (with the Slack restriction to All logged-in users), while web requests, Workato recipes and Workday actions run through administrator-configured connections whose runtime identity Atlassian does not specify, and the identity under which work items are created on the customer's behalf is not stated beyond "on behalf of your customers". (6) Default state is recorded as disabled because the agent requires setup and either AI answers activation or at least one active intent before it operates in any channel. (7) Atlassian does not document any autonomous action outside an administrator-authored conversation flow: the agent selects a branch, not an action set. (8) The run automation flow step is described on a page marked as referring to features currently rolling out. (9) The Conversations page does not yet show conversations from all channels. (10) Reuse of the same custom field in multiple ask for information steps in one branch may cause the wrong information to be used in later send web request steps. (11) Conversations are stored for 30 days and may be reviewed by human administrators; if organisation data contribution settings are enabled, Atlassian may use metadata or in-app data to improve apps for all customers, and the trained model is not shared with other customers.

Evidence