AI Settings
Configure AI behavior and response parameters for your Support BV
AI Settings
Configure how your Support BV processes inquiries and generates responses. These settings control the AI's accuracy, creativity, and decision-making.
Advanced Feature
Start with recommended defaults and adjust based on your specific needs and customer feedback.
Quick Setup
Access AI Settings
Navigate to Support BV > AI Settings tab.
Choose AI Model
Select the model that best fits your requirements and budget.
Adjust Temperature
Set creativity level based on your use case.
AI Model Selection
Choose from multiple AI providers and models:
- Gemini 2.5 Flash - Fast, efficient (1 credit/message)
- Gemini 2.5 Pro - Advanced capabilities (2 credits/message)
OpenAI
- GPT-5 Mini - Cost-effective (1 credit/message)
- GPT-5 - Powerful performance (2 credits/message)
- GPT-5 Chat - Optimized for conversations (2 credits/message)
Anthropic
- Claude 4 Sonnet - Balanced performance (5 credits/message)
- Claude 4 Opus - Maximum capability (5 credits/message)
xAI
- Grok 3 Mini - Efficient (1 credit/message)
- Grok 4 - Advanced (5 credits/message)
Credits
Credits indicate resource usage. More powerful models consume more credits from your monthly quota.
Core Parameters
Temperature
Controls response creativity and variation.
Setting | Value | When to Use | Behavior |
---|---|---|---|
Precise | 0.0-0.3 | Factual info, policies | Consistent, focused |
Balanced | 0.4-0.6 | General support | Mix of consistency and variety |
Creative | 0.7-1.0 | Marketing, engagement | Varied, expressive |
Example - Low Temperature (0.2):
"Our return policy allows returns within 30 days. Items must be unused in original packaging. Refunds process in 5-7 days."
Example - High Temperature (0.8):
"Great question! We offer a hassle-free 30-day return window. As long as items are in original condition, we'll process your refund within about a week. Easy peasy!"
High Temperature Warning
Values above 0.7 may produce unexpected or less accurate answers. Test thoroughly before using.
Response Length
Option | Words | Best For |
---|---|---|
Short | 50-100 | Quick answers, mobile |
Medium | 100-200 | Most use cases (recommended) |
Long | 200-300 | Complex explanations |
Variable | Adaptive | Diverse question types |
Confidence Threshold
Minimum confidence required before answering (vs. escalating to human):
Setting | Threshold | Behavior |
---|---|---|
Low | 40-60% | Answer most questions, rare escalations |
Medium | 60-75% | Balance accuracy and coverage (recommended) |
High | 75-90% | High-confidence answers only |
Conversation Memory
How much conversation history the AI remembers:
Setting | Messages | Best For |
---|---|---|
Short | 2-3 | Simple, independent questions |
Medium | 5-7 | Multi-turn conversations (recommended) |
Long | 10-15 | Complex support sessions |
Full Session | All | Deep troubleshooting |
Answer Strategy
Best for: Experienced customers, quick facts
Example:
Customer: "What are your hours?"
Response: "Monday-Friday 9 AM to 6 PM EST, Saturday 10 AM to 4 PM EST."
Best for: General customer support (recommended)
Example:
Customer: "What are your hours?"
Response: "Happy to help! We're open Monday-Friday 9 AM to 6 PM EST, and Saturdays 10 AM to 4 PM EST."
Best for: Complex products, personalized recommendations
Example:
Customer: "Which plan should I get?"
Response: "I'd be happy to help you choose! How many team members will be using it?"
Best for: Complex topics, first-time users
Example:
Customer: "How does auto-renewal work?"
Response: "Auto-renewal ensures uninterrupted service. Your subscription automatically renews at the end of each billing period using your saved payment method. You'll get a reminder 7 days before renewal."
Escalation Rules
Define when conversations transfer to humans:
Triggers:
- Low confidence (below threshold)
- Customer requests human
- Sensitive topics (complaints, refunds)
- Negative sentiment detected
- After X messages without resolution
Recommended Settings by Use Case
E-commerce
- Temperature: 0.5-0.6
- Response Length: Short-Medium
- Confidence: 65%
- Focus: Speed
B2B/Enterprise
- Temperature: 0.3-0.4
- Response Length: Medium-Long
- Confidence: 75%
- Focus: Accuracy
Technical Support
- Temperature: 0.2-0.3
- Response Length: Long
- Confidence: 80%
- Strategy: Educational
General Customer Service
- Temperature: 0.5
- Response Length: Medium
- Confidence: 65%
- Strategy: Conversational
Best Practices
Start Conservative
- Begin with recommended defaults
- Monitor performance for 1-2 weeks
- Make one adjustment at a time
- Measure impact before next change
Monitor Escalation Rates
Ideal: 10-20%
- Too low (<5%): May risk inaccurate answers
- Too high (>30%): Underutilizing AI capabilities
Regular Updates
- Add new FAQ content monthly
- Update product information when changed
- Refine based on conversation patterns
- Remove outdated information