AI Performance & Safety
The AI Performance & Safety tab continuously monitors the answers your assistant gives customers: questions it could not answer, risky replies and improvement opportunities β all on one screen.
Performance Score
The 0-100 score on the Overview sub-tab summarizes today's answer quality. It is calculated from today's message volume, detected knowledge gaps and safety evaluation results. Above 80 is considered healthy.
Knowledge Gap Detection
When the assistant cannot find a match in your knowledge base, the system inspects the exchange automatically. Not every unmatched message counts as a gap β messages are classified first:
- Greetings / bot commands / short replies: messages like "Hello", "/start", "OK" require no action and never enter the list.
- General information requests: questions unrelated to your business (weather etc.) do not count.
- Real knowledge gaps: a business-specific question was asked, your knowledge base had no answer and the assistant fell short β only these are listed.
In the Improvements sub-tab you can get an AI-drafted FAQ suggestion for each gap with one click, edit it and add it to your knowledge base. Irrelevant entries can be dismissed individually or in bulk.
Safety Review
On paid plans, every assistant reply is automatically reviewed in the background after it is sent: risky patterns such as unauthorized commitments, unverified price confirmations or requests for sensitive data are detected, and the reply receives a 0-100 safety score. Low-scoring replies appear in the Improvements list; critical cases trigger a notification.
Which plans?
Safety Settings
- Industry: a rule set matching your industry is applied automatically (e.g. no medical advice in healthcare).
- Sensitivity: Relaxed / Balanced / Strict β "Strict" is recommended for regulated industries.
- AI restrictions: toggle individual guards such as price commitments, medical/legal advice, competitor talk, sensitive-data collection and binding promises.
- Custom keyword rules: get flagged whenever words you define (e.g. competitor brand names) appear in replies.
Related Topics
Was this page helpful?