AI Performance and Safety

AI Performance and Safety is used to review the assistant's answers and identify topics that need improvement. From this section you can track knowledge gaps, safety evaluations and the related settings.

Sections

Overview: Presents the performance score, evaluation status and summary indicators.
Improvements: Lists knowledge gaps and answers that need review.
Safety Settings: Lets you manage review preferences and the safety restrictions the assistant should follow.

What Does the Performance Score Show?

The 0–100 performance score is calculated from that day's message count, detected knowledge gaps and safety evaluation results. HaveAI shows scores of 80 and above in the high-performance group. This classification belongs to the score HaveAI calculates; it does not replace reviewing answers one by one.

Do not interpret the score alone. On a day with no messages, the score may appear as 100. If safety review is off, the related evaluations are not produced either. Along with the score, check the message count, open issues and whether the review is active.

Knowledge Gap Detection

When the assistant cannot find enough business information for a question, the system can evaluate it as a possible knowledge gap. Not every unmatched message counts as a gap:

Messages like “Hello”, “OK” or bot commands are not treated as knowledge gaps.
General questions unrelated to your business are evaluated separately.
A question about your business being answered insufficiently because the needed information was not found can be an improvement topic.

When reviewing automatic detections, read the customer's question and the given answer together. If the relevant information already exists, check the clarity and accuracy of the existing record instead of adding a new copy.

Processing an Improvement Record

Open the Improvements section

Open the Improvements section and select the record.

Review the question and answer

Review the question, the given answer and the issue description.

Check the FAQ suggestion

For a knowledge gap you can get an AI-assisted FAQ suggestion. Check the suggested answer against your business's real information and edit it.

Save or dismiss

Save suitable content. You can close irrelevant records with the Dismiss option individually or with the offered bulk actions.

Re-test

Try the same question again on the AI Test screen.

A record being closed does not by itself show that the assistant now answers correctly. Also check the answer.

Safety Review

Safety review evaluates sent answers in the background. By examining risks such as price commitments, unverified information, requests for sensitive data or binding promises, it can produce a safety score and alerts.

This review runs after the answer is sent. Seeing a risk in the list does not mean the message was blocked before reaching the customer. Answers that need review appear in the Improvements section; when the notification conditions are met, an alert can be produced.

The availability of the review depends on your plan and account entitlements. It may be on by default on paid plans; you can check its status in Safety Settings. Since there may be differences on accounts shown as Free and with special account entitlements, base your judgment on your account's activity information.

Safety Settings

Industry: Used to determine rules suitable for your business.
Review sensitivity: Relaxed, Balanced or Strict options affect the evaluation thresholds. Choose the level suitable for your business's risks.
AI restrictions: Manage options for topics such as price commitments, medical or legal advice, competitor talk, requesting sensitive data and binding promises.
Custom keyword rules: Have the expressions you define tracked in evaluations. Do not interpret an alert rule as the word being removed from the answer text under all conditions.

Safety settings do not replace keeping business information accurate and up to date. After saving the settings, try the related sample questions on the AI Test screen.