Self-Learning Knowledge Base: Auto-Generated Articles from Resolved Tickets
Self-Learning Knowledge Base: Auto-Generated Articles from Resolved Tickets
Version introduced: v1.0.39
Overview
Starting in v1.0.39, the SaaS Factory support system can automatically draft knowledge base articles from tickets that were successfully resolved by the AI agent and confirmed as satisfactory by the customer. This closes the feedback loop between individual support interactions and the shared knowledge base, allowing the support system to improve itself over time without manual authoring effort.
How It Works
1. Triggering Conditions
An article draft is generated when both of the following are true:
- The support ticket was resolved by the AI agent with high confidence.
- The customer submitted a CSAT score of 4 or 5 via the post-resolution follow-up (
supportCsatFollowUp).
Neither condition alone is sufficient. A high CSAT score on a human-escalated ticket will not trigger a draft, and a high-confidence AI resolution with a low or missing CSAT score will not either.
2. Article Generation
Once both conditions are met, Claude analyzes the full ticket thread — the customer's original issue, any diagnostic steps, and the final resolution — and produces a structured KB article draft containing:
- A descriptive title derived from the problem type.
- A clear body explaining the issue and its resolution in general terms, stripped of customer-specific details.
- Suggested tags for categorization and AI retrieval.
3. Review Queue
The generated draft is saved with active: false, meaning it is invisible to customers and not yet used by the AI agent. It appears in a dedicated Suggested Articles section within the KnowledgeBaseTab in the Customer Management dashboard.
4. One-Click Approval
From the Suggested Articles section, reviewers can:
- Preview the full draft before approving.
- Edit the title, body, or tags if adjustments are needed.
- Approve with a single click — this sets
active: trueand immediately makes the article available to the AI support agent and any customer-facing search. - Dismiss drafts that are too narrow, inaccurate, or duplicative.
Knowledge Base Tab Changes
The KnowledgeBaseTab now contains two sections:
| Section | Description |
|---|---|
| Suggested Articles | AI-drafted articles pending review. Shown only when drafts exist. |
| Published Articles | All active: true articles, as before. Manually added articles continue to work exactly as before. |
The existing Add Article form remains available for manually authoring articles at any time.
Self-Improvement Loop
Every approved article feeds back into the AI support agent's retrieval context. This means:
- A novel issue is resolved by the AI.
- The resolution is validated by a high customer CSAT.
- A KB article is drafted and approved.
- The next customer with the same issue gets a faster, more confident resolution — because the agent now has an explicit article to cite and build on.
- Repeat.
Over time, the knowledge base grows proportionally to support volume rather than requiring proportional manual effort.
Limitations (v1.0.39)
- Drafts are only generated from AI-resolved tickets. Human-escalated tickets do not trigger auto-generation.
- The CSAT threshold is fixed at 4 or above on a 5-point scale.
- Duplicate detection across similar ticket types is not yet implemented — reviewers should check for overlapping articles before approving.
- Article generation requires the CSAT follow-up to complete; tickets closed without a CSAT response will not produce a draft.