This recipe shows how to safely automate the first line of support using your own knowledge base.
Many support questions are simple repeats of documented issues. A tier-1 agent answers these automatically, freeing humans for nuanced work. The key is to ground it strictly in your content and define clear escalation paths.
Configure Audiences for “public support” and one for “logged-in support” if needed. Ground responses with flat Topics like FAQ, Troubleshooting, How-to-guides, and Error-codes. Index your help center, troubleshooting guides, and process docs under these Topics. Instruct the skill to quote or summarize from here only, reducing hallucinations.
For certain sensitive intents—like complaints, legal threats, or data-rights requests—train the skill to recognize them and not answer substantively. You can add specific AI Skills for these situation, allowing for further questioning by the AI to determine the gravity and labeling of the issue. Otherwise, it sets a needs_escalation flag and creates a ticket or task with an AI summary. A notification can alert the right team (for example, legal or privacy) when these appear.
You can measure deflection and identify knowledge gaps by tracking which questions the AI cannot answer well. It can periodically send notifications summarizing low-confidence topics along with example queries, prompting content updates.
For customers in the portal, in-app notifications can promote self-service by linking to high-value FAQs or flows based on user_segment and recent actions. This segmentation helps drive traffic to the AI agent and reduces tickets on common topics.
Conclusion
A tier-1 troubleshooting and FAQ resolution agent uses your existing knowledge to provide fast, grounded answers. With AI Skills tied to flat Topics, plus tasks and notifications for escalations and content gaps, it reduces support load while keeping control where it matters.