Reviewing KB Proposals

When the learning loop is enabled, oHallo analyses resolved conversations to identify gaps in your knowledge base. If the AI notices questions it could not answer well — or where a human agent provided a better answer — it proposes new KB entries to fill those gaps.

Where to find proposals

Open Knowledge Base in the sidebar and click the Pending review tab. The tab shows a badge with the number of proposals waiting for your decision.

Each proposal in the list shows:

• Question — The proposed question the entry would answer
• Answer — The proposed answer text
• Source conversation — A link to the conversation that triggered this proposal
• Confidence score — How confident the AI is that this entry would be useful
• Motivation — A brief explanation of why the AI thinks this entry is needed

Approving a proposal

If the proposed question and answer are accurate and useful:

1. Review the question and answer text.
2. Click Approve.
3. The entry status changes to Approved and the AI can immediately use it when answering future customer questions.

Editing before approving

If the proposal is on the right track but needs adjustments:

1. Click Edit on the proposal.
2. Modify the question, answer, or topics as needed.
3. Click Approve to save your changes and activate the entry.

This is common — the AI often captures the right idea but may need refinement in wording, completeness, or tone to match your brand.

Rejecting a proposal

If the proposal is not useful, inaccurate, or duplicates an existing entry:

1. Click Reject.
2. Optionally provide a reason for the rejection.

Rejection feedback helps the learning loop improve over time. If you explain why a proposal was rejected (e.g. “We already cover this in the shipping FAQ” or “This policy has changed”), the AI is less likely to propose similar entries in the future.

Tips for reviewing proposals

Review regularly. Proposals accumulate over time. A weekly review keeps your knowledge base growing steadily without a backlog piling up.

Check against existing entries. Before approving, scan your knowledge base for entries that already cover the same topic. Approve only if the proposal fills a genuine gap.

Edit liberally. Treat proposals as drafts. The AI identifies what is missing, but you know your business best — adjust the wording to match your standards.

Use rejection reasons. A short note on why you rejected a proposal goes a long way toward improving future suggestions.