Claire_Conant
ServiceNow Employee
Options
- Post History
- Subscribe to RSS Feed
- Mark as New
- Mark as Read
- Bookmark
- Subscribe
- Printer Friendly Page
- Report Inappropriate Content
4 hours ago
You built the topic, tested it in the designer, and it worked. Then it hit production and broke. Maybe the topic won't fire, the panel won't load, the Natural Language Understanding (NLU) model keeps guessing wrong, or every test ends in "I'm having technical issues." Virtual Agent runs on four configuration layers: topic setup, the NLU model, the portal embed, and live agent routing. A failure in any one of them tends to surface as the same handful of symptoms, which is what makes these so hard to pin down. Find your symptom below and start with the layer most likely behind it.
If your Virtual Agent is running but producing wrong or low-quality answers, that's a topic design problem, not a setup problem. See Five Virtual Agent configuration steps to verify before go-live.
If you're troubleshooting Now Assist (generative AI) responses, the configuration layers differ from classic Virtual Agent.
See: Now Assist in Virtual Agent doesn't fire topics. It runs plans.
See: Where to look when your Now Assist in Virtual Agent answers are off
1. Topic doesn't trigger or the wrong topic fires
This issue comes down to whether your topic is using keyword matching or Natural Language Understanding (NLU). Check the topic Properties tab in Virtual Agent Designer. If an NLU model and intent are assigned, the topic uses NLU to interpret what users type. If the topic uses keyword triggers instead, there's no NLU model listed.
If your topic uses keyword matching
Keyword trigger configuration on the topic controls which input maps to which topic. If keywords overlap between topics, the wrong one fires. If none match what the user typed, no topic fires at all.
If your topic uses NLU
The NLU model, the intent, and the topic must stay bound for prediction to work. Several configuration points can break that binding with no obvious error. Model publication state, topic language records, application scope alignment, and the discoverability flag are all part of the same chain.
See: Virtual Agent does not predict the mapped topic when NLU is enabled
See: Checklist for Virtual Agent NLU discovery troubleshooting
See: Checklist for Virtual Agent NLU discovery troubleshooting
If your custom intent isn't appearing in the Associated Intent list
Each intent belongs to exactly one NLU model; you can't share an intent across models. If you created the intent in a different model than the one assigned to your topic, it won't appear in the list.
2. NLU predicts the wrong intent or no intent at all
These two failures look similar on the surface but come from different places.
When Virtual Agent picks the wrong topic
The issue is usually competing intents. If two or more intents have overlapping utterances, or if the confidence threshold is set low enough to let marginal matches through, Virtual Agent may fire on an intent that barely qualifies instead of waiting for a stronger one.
When Virtual Agent doesn't match any topic at all
Even though NLU Workbench shows a correct prediction, the root causes overlap with Scenario 1. The model binding, application scope, discoverability flag, and role restrictions are the same configuration points to check.
3. Virtual Agent panel doesn't appear in your portal or external embed
Where you're embedding Virtual Agent determines where to look. Two deployment contexts can produce the same symptom but trace to different places.
Embedded in an external website via iframe
Topics need the Public role assigned to display in an external embed. This is a per-topic setting, so you may see some topics appearing while others are missing. If that's what you're seeing, check the role assignment on the topics that aren't showing up.
In a scoped portal (such as the Employee Service Center [ESC] portal)
Topics created in the Global application scope won't appear in a portal scoped to a specific application (for example, Human Resources: Virtual Agent Conversations). If your topics are missing from a scoped portal, the path forward is creating them in the matching application scope.
Note: If you upgraded to Zurich and the Discoverable and Visible options are no longer at the topic level, this removal was intentional and is not a bug.
4. Live agent handoff starts but nobody connects
In this scenario, Virtual Agent acknowledges the request and appears to be working, but the conversation either hangs at "Routing you to a live agent" or returns "No Chat Agents Currently Available" even though agents show available in Workspace.
There are several possible root causes here, so it helps to work through them from simplest to most involved.
The Live Agent Support topic may be inactive or unpublished
If this base system topic isn't both active and published, the handoff has nowhere to go. Go to sys_cs_topic.list, find the Live Agent Support topic, and confirm that both the Active and Published fields are checked.
The Fulfiller UI may be set incorrectly on the chat setup record
If the Fulfiller UI is set to Connect instead of Agent Workspace, live agent routing can fail with an invalid queue message in the logs. Go to sys_cs_live_agent_setup.list and verify the Fulfiller UI is set to Agent Workspace.
The Advanced Work Assignment (AWA) queue routing condition may be blocking work items
AWA queues use a Work item routing condition to determine which interactions get routed. If that condition filters on the State field, it can silently block routing. Interactions are created in a Work in Progress state, but if the condition expects New, no work items are ever created.
Go to Advanced Work Assignment > Settings > Queues, open the queue, and check whether the Work item routing condition includes a State filter. If it does, remove the State filter. Other fields can be used in the condition if needed.
The interaction is auto-assigned instead of routed to AWA
If the assigned_to field on the Interaction table has a default value set in the sys_dictionary, interactions get assigned directly to the user instead of creating a work item. AWA never picks it up, so the agent never sees the request.
Handoff breaks without an obvious error message
This can happen when the chat experience uses a custom context topic instead of the base system Live Agent Support, when a queue has an incorrect context condition, or when a schedule configuration doesn't match agent availability.
Routing to specific AWA queues dynamically
The Live Agent queue variable only works with Connect Support queues (chat_queue), not AWA queues (awa_queue). If you're trying to route dynamically to AWA queues, you'll need to use a context variable with the queue's routing condition instead.
See: Work item queues
5. Conversations end with "I'm having technical issues"
This error appears across several unrelated root causes and doesn't indicate what actually failed, which is what makes it harder to trace. You'll need to work backward through the conversation records to find the source.
A setup topic may be missing or inactive on the chat experience
If the Closing topic or another topic configured as a setup topic on the chat experience record is inactive, Virtual Agent surfaces this error at the end of every conversation. Check that all topics referenced in your chat experience configuration are active and published.
A duplicate Virtual Agent widget on the page
If the Virtual Agent widget loads twice on the same page (for example, once from the page itself and again from the portal header, footer, or Agent Chat configuration), the duplicate creates a conflict that specifically breaks live agent handoffs. Check your portal page, theme header and footer, and any Agent Chat definitions for duplicate instances.
A script error in a topic flow
If a Bot Response, Bot Input, or custom script node hits an unhandled exception, Virtual Agent surfaces this error instead of the actual exception details.
Where to go from here
Most Virtual Agent problems after setup trace back to one of these four layers, and once you know which layer you're in, the path forward is usually clear. If you'd rather understand how the layers fit together than chase a single fix, the Virtual Agent learning path covers the full setup chain end to end. Keep it handy for your next rollout so you know exactly where to look.
Note: Licensing, usage, and entitlements can vary by release, setup, and contract. Always confirm the specifics for your environment.
