The model takes into consideration the previous chat/replies, the context of the text entered, and also the words and phrases used so that if the use of a particular word or phrase is causing the non-compliance of the response, the model will flag it. The model has a very short latency period which means that they flag a response as compliant or non-compliant as soon as the response is entered in the reply box.

No, the feedback process is governed by user permissions. Only the users with appropriate permission will be able to take the necessary feedback action. Users should have the ENABLE RESPONSE COMPLIANCE FEEDBACK permission granted.

Once a sufficient number of feedback messages are received, we will do an internal quality check on the feedback received before re-training the model. The model can be retrained once in a quarter, provided we have sufficient feedback to incorporate.

The publishing of a reply to the customer is not stopped if it’s non-compliant. The smart compliance check only generates a flag as a warning. In case, you want to stop publishing a non-compliant response or route it through an approval process, you can leverage the rule configuration.

You can create a new outbound rule and add non-compliant conditions under The Properties of the Outbound Message. They are Is Biased, Is Profane, Is Irrelevant, Is Improper Tone. Next, you can add actions as required. This could be either setting an approval path - where you set up an approval mechanism for the outbound replies or blocking such outbound replies.

Response Compliance ensures that each agent response adheres to – Biased Content, Profanity, Tonality, and Relevance and accordingly flags it as compliant or non-compliant, whereas Smart Responses are responses that are auto-generated by AI and suggested to agents.

Yes, it is possible to modify a non-compliant reply before publishing.