Knowledge Builder
Knowledge Builder in the sidebar of the admin console at app.flexivity.ai finds the questions your knowledge base does not answer, and drafts articles for them.
A run groups your resolved tickets into topics, then compares each topic with your published articles. A topic that no article covers becomes a candidate. For each candidate you can generate a draft article, edit it, and push it to your help desk’s knowledge base.
An Admin or a Knowledge Manager can use every part of the Knowledge Builder; a Read Only user can see it but not change it. See Users.
Brands
The Knowledge Builder works on one brand at a time. If you have more than one, choose it from the Brand selector at the top of the page.
A run reads every help desk connected to that brand. There is no account-wide run, because topics from different brands would be mixed together.
The Knowledge Builder needs a brand’s ticket history. Run a Historical Sync on the Help Desk page first. Until the sync finishes, Run Analysis shows Waiting for ticket sync.
Runs
The page shows:
- Total Candidates, Articles Published and Coverage for the selected run
- Run History: every run for the brand, with its status (Pending, Running, Completed). Select a run to see its candidates.
Run Analysis starts a new run. A run takes a few minutes; the page updates when it finishes. Delete Run removes a run and its candidates.
Running an analysis every month or so works well for most teams, so new patterns are caught as your ticket mix changes.
Candidates and drafts
Each candidate is a topic: a group of similar resolved tickets that no published article covers. Candidates with more tickets are the topics that generate the most support work, so start with those.
For each candidate:
- Generate Draft writes an article from the tickets in the topic, using the brand’s Analysis Settings.
- Push to Zendesk sends the draft to Help Center as an unpublished draft, in the section you choose each time you push. Review and publish it in Zendesk.
- Push to osTicket creates a new FAQ article in the category you choose. Review the draft before you push it.
- Dismiss removes a candidate that does not need an article, such as a one-off issue or an internal process.
A draft is only as detailed as the tickets it was written from. Check it for accuracy before publishing.
Analysis Period
The Analysis Period card sets how far back a run looks for resolved tickets. The periods available depend on your plan.
A run can only read tickets that have been synced. If the period is longer than your Historical Sync, the page warns you; extend the sync on the Help Desk page to cover it.
Analysis Settings
Configure on the Analysis Period card opens Analysis Settings: how a run finds topics and how drafts are written, for the selected brand. ← Knowledge Builder takes you back.
Each setting shows where its value comes from: Built-in default, Inherited from the account-wide value, or Set for this brand. A new run uses the settings in force when it starts. Drafts already pushed are not rewritten; generate a draft again to pick up new settings.
| Setting | What it does | Default |
|---|---|---|
| Minimum topic size | How many similar resolved tickets make a topic worth an article, from 3 to 20. Higher values focus on topics that recur often; lower values catch niche issues. | 5 |
| Topic granularity | Finer (more topics) finds more, narrower topics. Broader (fewer topics) groups them into fewer. | Finer |
| Noise sensitivity | Sensitive, Balanced or Conservative. Higher settings leave more one-off tickets out of every topic. | Sensitive |
| Coverage threshold | How close an existing article must be before a topic counts as already covered, from 0.50 to 0.90. Raise it to find more gaps; lower it if topics you already cover keep appearing. | 0.65 |
| Article format | Knowledge-base topic (a longer article) or FAQ (a short question and answer). Pick the one that matches how your knowledge base is organised. | Knowledge-base topic |
| Draft temperature | How much a draft’s wording may vary, from 0.1 to 0.5. Lower keeps drafts closer to the tickets’ own wording. | 0.3 |
| Draft instructions | What the AI is told when it writes a draft. Until you change them, the article format’s built-in instructions are used. | The format’s own |
Draft instructions
Use built-in default loads the article format’s built-in instructions into the editor. Saving them removes the brand’s own instructions, so the brand follows the built-in ones from then on, including any later improvements to them.
History lists the brand’s earlier instructions, newest first, with who saved each one. Restore loads a version into the editor so you can review it; nothing changes until you save. A version marked went back to the built-in default restores the built-in instructions.
Troubleshooting
The run shows no candidates. Your articles may already cover the main topics, or the coverage threshold is too low. Try raising the Coverage threshold.
A topic already has an article. Sync your knowledge base so the run compares against your current articles, and archive out-of-date ones so they do not count as coverage.
Topics are too broad or too narrow. Adjust Topic granularity and Minimum topic size, then run the analysis again.
For help-desk-specific steps, see Knowledge Builder for Zendesk and for osTicket.