How to fix AI Employee responses that do not match your business rules

    CategoryAI Employee
    Article typeDocumentation
    Estimated reading time 5 min
    Last reviewedRecently reviewed
    Applies toGEVADE CRM accounts where the feature is enabled
    Access requiredAccess may depend on your user permissions
    Plan/configuration noteFeature availability may depend on your selected plan and account configuration

    Summary

    Review your Business Rules and Custom Instructions for contradictory commands, ensure restrictions are explicit, and verify that your knowledge base doesn't conflict with your rules.

    Who this article is for

    This article is for administrators troubleshooting compliance and policy adherence issues with the AI Employee.

    Overview

    When the AI Employee ignores a rule (like discussing a restricted topic or offering a discount it shouldn't), it is usually due to conflicting instructions or ambiguous language in your setup.

    Before you begin

    • Identify the specific rule that was broken
    • Review the Business Rules tab
    • Check for conflicting information in the Knowledge Base

    Step 1: Step 1: Isolate the Incident

    Find the conversation log where the AI broke the rule. Note exactly what the AI said and what the customer asked to trigger it.

    Step 2: Step 2: Review Explicit Rules

    Navigate to Settings > AI Employee > Business Rules. Check how the broken rule is written. Is it vague? Change 'Try not to discuss pricing' to 'NEVER discuss pricing. Always use the fallback response.'

    Step 3: Step 3: Check for Contradictions

    Look at your Custom Instructions and Tone settings. Do you have a rule that says 'Always be helpful and answer every question'? This might override a rule that says 'Do not answer questions about competitors.'

    Step 4: Step 4: Audit Knowledge Sources

    Check your Knowledge Base. If you have a rule forbidding the AI from offering discounts, but you uploaded a PDF that lists a 10% discount code, the AI will likely offer it. Ensure your knowledge sources align with your rules.

    Step 5: Step 5: Prioritize Instructions

    If your system allows, move your most critical business rules and restrictions to the top of the Custom Instructions field to ensure they are prioritized.

    Step 6: Step 6: Sync and Test

    Click 'Sync' and rigorously test the scenario in the simulator to ensure the AI now adheres to the rule.

    Configuration note

    We do not guarantee AI accuracy, lead conversion, bookings, customer satisfaction, response quality, revenue, support resolution, or business outcomes. The AI Employee does not fully replace human staff. AI behaviour depends on knowledge sources, instructions, handoff rules, enabled features, connected channels, permissions, configuration, testing, and ongoing monitoring. AI-generated responses should be reviewed before relying on them for sensitive, urgent, legal, financial, medical, billing, or policy matters. The AI Employee may need human handoff for these situations. Knowledge sources must be reviewed, updated, and tested regularly. Usage charges, feature access, fair use rules, or add-on requirements may apply depending on your account setup. Always test before going live. Contact support if an option is not visible. The AI interprets instructions literally. It does not understand nuance or implied intent. Your rules must be absolute and explicit.

    Important clarification

    Writing rules that are too complex or conditional. Keep rules simple and direct. Instead of 'If the customer asks about X, and it's Tuesday, say Y,' use straightforward commands.

    Resolution guidance

    The AI is still breaking the rule after I updated the instructions.

    You may have too many instructions overall, causing the AI to lose focus. Try simplifying your prompt and removing unnecessary constraints.

    Was this article helpful?

    Need account-specific help?

    If your question requires account access, billing review, workflow inspection, message delivery diagnosis, AI response review, or private configuration review, submit a support request with examples and screenshots.