Our support team is getting flooded with complaints. The new AI-assist feature in our platform (vendor name redacted for now) is aggressively suggesting step-by-step password reset guides on tickets where the customer is clearly reporting a compromised account or a suspicious login attempt.
This is not just unhelpful; it's a security risk. Telling a user who says "someone logged into my account from another country" to just "reset your password via the forgot password link" is negligent.
I have the deflection rate data, and it's bad. For tickets tagged "security" or "account compromise," the AI's suggested reply has a 92% inaccuracy rate based on manual sampling of 50 tickets last week. Agents are wasting time deleting its suggestions and escalating properly.
I need to know what configuration or training data parameter controls this. We're using a common support platform with a recently launched "Smart Reply" AI.
* Is this a ticket tagging issue? Our "Account Security" tag should theoretically exclude it from basic password reset logic.
* Is it a keyword trigger problem? The AI sees "password" and "can't login" and defaults to its most common solution.
* Has anyone successfully built an exclusion rule or intent filter for this specific scenario?
If you've solved this, I need to see the actual config approach. Generic "talk to your CSM" advice is not actionable.
Show me the query.