Overview
Your AI agent's effectiveness depends on how well it's configured to match your brand voice and handle edge cases gracefully. This guide covers the three core configuration areas:
Tone & Personality — How your AI agent communicates
Fallback Behavior — What happens when the AI can't help
Handoff Rules — When and how conversations transfer to humans
Tone & Personality
Setting Your Brand Voice
Your AI agent should sound like an extension of your team. In Intercom, navigate to Fin AI Agent → Customize → Tone of voice to configure these settings.
Available tone options:
Tone | Best For |
Professional | B2B, enterprise, financial services |
Friendly | Consumer brands, e-commerce, SaaS |
Casual | Startups, creative industries, younger audiences |
Empathetic | Healthcare, support-heavy products, sensitive topics |
Custom Instructions
Beyond the preset tones, you can add custom instructions to fine-tune responses:
Do's — "Always greet customers by name when available", "Use simple language, avoid jargon"
Don'ts — "Never discuss competitor products", "Don't make promises about timelines"
Brand-specific terms — Specify how to refer to your product, features, or company
Tip: Review your best human agent responses and extract patterns. What phrases do they use? How do they sign off? Use these as custom instructions.
Response Length
Configure how verbose your AI agent should be:
Concise — Short, direct answers (best for simple FAQs)
Detailed — Thorough explanations with context (best for technical products)
Adaptive — Matches response length to question complexity (recommended)
Fallback Behavior
When your AI agent doesn't know the answer or confidence is low, fallback behavior determines what happens next.
Confidence Threshold
Set the minimum confidence level required before the AI responds autonomously:
Threshold | Behavior |
High (80%+) | AI only answers when very certain; more handoffs to humans |
Medium (60-79%) | Balanced approach; AI handles most queries |
Low (40-59%) | AI attempts more answers; may reduce accuracy |
Recommendation: Start with a high threshold and lower it gradually as you review AI performance and expand your knowledge base.
Fallback Options
When confidence is below your threshold, choose what happens:
Graceful acknowledgment — AI admits it doesn't know and offers alternatives
"I'm not sure about that specific question. Would you like me to connect you with our team?"
Suggest related articles — AI offers knowledge base articles that might help
"I couldn't find an exact answer, but these articles might help: [Article 1], [Article 2]"
Immediate handoff — Conversation transfers directly to a human agent
Best for high-stakes support or complex products
Collect information first — AI gathers details before handoff
"Let me get some details so our team can help you faster. What's your order number?"
"I Don't Know" Responses
Customize how your AI handles uncertainty. Avoid generic responses like "I don't understand" — instead, make them helpful:
Good example:
"I don't have specific information about custom pricing. Our sales team can help with that — would you like me to connect you, or would you prefer to email [email protected]?"
Avoid:
"I'm sorry, I don't understand your question."
Handoff Rules
Handoff rules determine when conversations transfer from AI to human agents and how that transition happens.
Automatic Handoff Triggers
Configure scenarios that always trigger a handoff:
Keyword-based triggers:
Cancellation requests — "cancel", "refund", "close my account"
Complaints — "speak to manager", "this is unacceptable", "lawsuit"
Urgent issues — "emergency", "critical", "down", "not working"
Sentiment-based triggers:
Negative sentiment detected (frustration, anger)
Multiple messages without resolution
Customer explicitly requests human help
Topic-based triggers:
Billing disputes
Legal inquiries
Security concerns
Complex technical issues
Handoff Routing
Define where conversations go when handed off:
Scenario | Route To |
Billing questions | Billing team / queue |
Technical issues | Support team / queue |
Sales inquiries | Sales team / queue |
General / unclear | Default inbox |
VIP customers | Priority queue |
Handoff Experience
Configure what happens during the transition:
Before handoff:
AI summarizes the conversation for the human agent
AI collects any missing information (email, order number, etc.)
AI sets customer expectations ("A team member will respond within 2 hours")
During handoff:
Show a clear message that a human is taking over
Display estimated wait time if available
Offer callback option during high-volume periods
After handoff:
Human agent sees full conversation history
AI-generated summary appears at the top
Suggested responses based on the query (optional)
Tip: Enable "warm handoffs" so customers don't have to repeat themselves. The AI summary should include: the customer's question, what was already tried, and any relevant account details.
Testing Your Configuration
Before going live with changes, test your configuration:
Use Intercom's test mode — Simulate conversations without affecting real customers
Try edge cases — Test unclear questions, complaints, and handoff triggers
Review AI responses — Check tone consistency and accuracy
Monitor the first week — Watch for unexpected behavior and adjust
Key Metrics to Track
After configuration changes, monitor:
Resolution rate — Are more conversations resolved without handoff?
Customer satisfaction — CSAT scores for AI vs. human interactions
Handoff rate — Is the AI handing off too often or not enough?
Average handle time — Is the AI reducing time to resolution?
Common Configuration Mistakes
Setting confidence too low — Results in incorrect answers that frustrate customers. Start high.
Generic fallback messages — "I don't understand" helps no one. Always offer a next step.
No handoff for complaints — Frustrated customers need humans. Always route complaints.
Forgetting to update after product changes — New features or policy changes need AI retraining.
Over-customizing tone — Too many custom instructions can make responses inconsistent. Keep it simple.
Need Help?
If you need assistance configuring your AI agent:
Email: [email protected]
Phone: +41 79 741 86 13
Book a configuration session: Contact us
Clients on an active retainer can request configuration changes as part of their monthly support hours.