Reviewing Policy Proposals

When the learning loop is enabled, oHallo analyses resolved conversations and may propose new policy rules in addition to knowledge base entries. Policy proposals capture behavioural patterns — situations where the AI should follow a specific rule that is not yet defined.

Where to find proposals

Open Policies in the sidebar and click the Pending review tab. A badge on the tab shows how many proposals are waiting for your decision.

Each proposal includes:

• Topic — The category the rule falls under (e.g. “returns”, “shipping”)
• Rule text — The specific instruction the AI would follow
• Source evidence — The conversation or pattern that prompted this proposal
• Confidence — How confident the AI is that this rule would be useful
• Rationale — An explanation of why the AI thinks this rule is needed

Approving a proposal

If the proposed rule is accurate and reflects how you want the AI to behave:

1. Review the topic and rule text.
2. Click Approve.
3. The policy becomes active immediately and the AI will follow it in future conversations.

Editing before approving

If the proposal captures the right idea but needs adjustments:

1. Click Edit on the proposal.
2. Modify the topic, rule text, or conditions as needed.
3. Click Approve to save and activate the policy.

It is common to tighten the wording, add specific thresholds, or narrow the scope of a proposed rule before approving it.

Rejecting a proposal

If the proposal is not useful, conflicts with existing rules, or does not reflect your business practices:

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

Rejection feedback improves the learning loop over time. When you explain why a proposal was rejected (e.g. “This conflicts with our updated return policy” or “We handle this case differently”), the AI adjusts its future suggestions accordingly.

Tips for reviewing policy proposals

Check for conflicts. Before approving, review your existing policies to ensure the new rule does not contradict an established one. Conflicting policies can confuse the AI and lead to inconsistent responses.

Be precise with thresholds. If a proposal says “escalate large orders”, edit it to include a specific amount before approving.

Review alongside KB proposals. Policy proposals and knowledge base proposals often come from the same conversations. Reviewing them together gives you a complete picture of what the learning loop identified.

Provide rejection reasons. Even a brief note like “Already covered by existing policy” helps the AI avoid proposing similar rules in the future.