Skip to main content

How to Train Periskope AI

1

Go to the AI Knowledge Base

To get started with training your Periskope AI, head over to AI > Knowledge Base
AI Knowledge Base interface
2

FAQs vs Documents

You have two ways to train the AI: by setting up FAQs or uploading a Training Document
FAQ and Documents training options
  1. FAQs: Better for handling specific questions and instructions to respond. (e.g. have AI ask follow up questions and debug an issue)
  2. Documents: Better for giving AI large contexts (e.g. for finding information about company services and answering a relevant question)

Structuring Your Knowledge Base

How to Structure FAQs

FAQs are essentially prompts for the AI Agent and should be structured as such. A question in the FAQ is used by the AI to search for an similar questions and the Answer tells the LLM what to do in response. Creating and tuning your FAQs to get the results that you want, may take multiple tries to acheive. If you have little experience with prompt engineering, or would like a deep dive and learn the best practices, you can read up on it here: Prompt Engineering Guide In general, when creating FAQs, follow these best practices for optimal AI responses:
  1. Include Multiple Question Patterns
    • Add various ways users might ask the same question
    • Example for a pricing question:
      Q: How much does your service cost?
      Q: What are your pricing plans?
      Q: Can you tell me about your subscription fees?
      A: Our service has three tiers:
      - Basic: $10/month
      - Pro: $25/month
      - Enterprise: Custom pricing
      
    AI Knowledge Structuring 1

    Adding multiple questions to an FAQ for better identification

  2. Add Instructions for controlled AI Responses
The new FAQ structure includes a separate Instructions field where you can guide the AI’s response behavior without using special tags.
New FAQ Structure

New FAQ Structure with Separate Instructions Field

FAQ Structure Components:
  • Question: The customer question or query variations
  • Answer: The response the AI should provide to customers
  • Instructions: Special behavior instructions for the AI (separate field)
Types of Instructions:
  • Basic Instructions
    Ask for the user's account email before providing specific account details
    
  • Conditional Instructions
    If user mentions error code, ask for screenshot before troubleshooting
    If user asks about pricing, confirm their region first
    
  • Follow-up Instructions
    Add follow-up instructions for multi-turn conversations if you wish the AI to follow up with different responses or escalate the issue by raising a ticket. Yes, the AI Agent can raise a ticket automatically to escalate an issue to a human! 😄
    ONLY If they are facing issues with adding a new user, like they are unable to add any more users, it is likely their plan supports limited number of users. To add more, they'll need to upgrade. Find out if that is the case and if so you can ask them to upgrade from Settings > Manage Plans or notify the sales team to step in.
    

How to Structure Documents

When uploading training documents, organize information for better AI comprehension. The following formats are recommended, listed in order of preference:
  1. Q: Ans: Format (Most Favored)
    • Structure your content as question-answer pairs
    • This format is highly effective for training as it directly mirrors how the AI processes queries
    • Example:
      Q: How do I reset my password?
      Ans: To reset your password, go to the login page and click on "Forgot Password". Enter your email address and you'll receive a reset link within 5 minutes. Click the link and follow the prompts to create a new password.
      
      Q: What payment methods do you accept?
      Ans: We accept all major credit cards (Visa, Mastercard, American Express), PayPal, and bank transfers for enterprise accounts. Payment is processed securely through our encrypted payment gateway.
      
      Q: How can I contact customer support?
      Ans: You can reach our customer support team through the in-app chat feature, email at [email protected], or by creating a ticket from your dashboard. Our team typically responds within 2-4 hours during business hours.
      
  2. Paragraph Structure
    • Group related information into single, focused paragraphs
    • Keep each paragraph focused on one main topic
    • Example:
      [Product Features]
      Our platform includes real-time analytics, custom reporting, and team collaboration tools. All features are accessible through our dashboard and come with detailed documentation. The analytics module provides instant insights into customer behavior and engagement metrics.
      
      [Security Measures]
      We implement industry-standard encryption for all data transmissions. Our servers are hosted in ISO-certified data centers with 24/7 monitoring. We conduct regular security audits and maintain SOC 2 compliance.
      
  3. List Structure
    • Organize information in bullet points or numbered lists
    • Useful for step-by-step instructions or feature lists
    • Example:
      [Getting Started Guide]
      - Sign up for an account using your email address
      - Verify your email through the confirmation link
      - Complete your profile setup
      - Connect your first integration
      - Invite team members to collaborate
      
      [Available Integrations]
      - HubSpot CRM
      - Google Sheets
      - Slack
      - Zapier
      - Salesforce
      

Knowledge Base Optimization Tips

FAQ Best Practices

Question Variations
  • Include different ways customers might phrase the same question
  • Consider regional language differences and terminology
  • Add common misspellings or abbreviations
Answer Quality
  • Keep answers concise but comprehensive
  • Use bullet points for multiple items
  • Include relevant links or references
  • Test answers with real customer scenarios
Instruction Effectiveness
  • Be specific about when and how AI should behave
  • Use conditional logic for complex scenarios
  • Include escalation criteria for human handoff
  • Test instructions with edge cases

Training Optimization

Regular Updates
  • Review and update content based on customer feedback
  • Add new FAQs for recurring questions
  • Remove outdated information
  • Monitor AI response quality and adjust accordingly
Performance Monitoring
  • Track which FAQs are most frequently used
  • Identify gaps in knowledge base coverage
  • Monitor customer satisfaction with AI responses
  • Adjust training based on conversation outcomes
Start with a small set of high-quality FAQs and gradually expand based on actual customer interactions. Quality is more important than quantity when training your AI Agent.

AI Agent Behavior Configuration

The Prompts section allows you to define how your AI Agent should behave, communicate, and handle different scenarios. Access this through AI > Prompts.

🤖 AI Nickname

Provide a nickname for the AI. This will be used to identify the AI in chat conversations with customers.

🎭 AI Personality

Select a predefined personality that defines the AI’s behavior, tone, and response style. Each personality comes with pre-configured settings for communication style, response length, and how to handle unknown information.
AI Agent Prompts Interface

AI Agent Personalities

Available Personalities:
  • Strict Grounded: Fact-based—only verified info, no guessing or assumptions.
  • Spartan: Ultra-brief, authoritative, zero fluff.
  • Friendly: Warm, conversational, customer-focused and approachable. Conversational and polite with moderate detail. Ensures customers feel heard without fabricating info or over-apologizing. Ideal for general support and onboarding.
  • Sales Forward: Confident, value-driven, conversion-focused.
Each personality has a specific:
  • Tone Style: Defines the communication tone (e.g., warm and polite)
  • Response Length: Sets how detailed responses should be (moderate, concise, detailed)
  • Unknown Info: Guidelines for handling situations when information is unavailable
  • Restrictions: Specific behaviors to avoid (e.g., no fabricating information, no excessive apologizing)

🏢 Company Context & AI Role

This information will be used by the AI to understand your company and what the AI’s role is within your organization. Define what the AI should do in your company and specific roles it should take when responding to customers. This field combines company context, AI role definition, team structure, and operational guidelines into a single comprehensive configuration.
AI Agent Prompts Interface

AI Agent Prompts Configuration

Example Configuration:
Role & Identity

You are an AI-powered Sales and Customer Support Assistant for [Company Name].
Your primary responsibility is to handle customer conversations across sales, support, onboarding, and operations. Customers may ask about features, raise doubts, report issues, request demos, or ask for meetings.

You work closely with the internal team at [Company Name] to ensure customers receive accurate information and timely support.

Communication Guidelines

You are warm, polite, patient, and solution-oriented in all responses.

You provide information strictly based on the relevant context and knowledge base provided to you.

If the context is irrelevant or insufficient, you actively search the knowledge base for the correct information.

If you do not have the required information even after searching:

Inform the customer once that you do not have that information.

After that, stay silent on that topic unless the customer asks a different question.

You never guess, hallucinate, or fabricate information.

Internal Coordination & Delegation

Think of yourself as the team coordinator.

When there are actionable items, issues, or follow-ups in a conversation, you delegate them to the appropriate internal team member.

You use:

Private Notes for low-priority communication, internal clarifications, or FYIs.

Tickets for high-priority issues, bugs, outages, or customer-blocking problems.

When informing the customer that a ticket or note has been created:

Do not disclose internal email addresses or personal identifiers.

Keep the response professional and reassuring.

About [Company Name]

[Insert a clear, product-focused description here. Example structure below:]

[Company Name] is a platform designed to help businesses [primary value proposition]. It enables teams to:

Core Capability 1 – Short description

Core Capability 2 – Short description

Core Capability 3 – Short description

Automation & Workflows – What can be automated and how

Analytics & Reporting – What insights users can get

Integrations – Which tools or systems it connects with

Exports & Data Access – What data can be exported and in what formats

In essence, [Company Name] helps businesses [one-line summary of outcome and value].

Team Structure

You have access to the following internal teams. Route issues and requests accordingly.

Engineering Team

[Engineer Name / Role] – Owns: [Feature areas, systems, or modules]

[Engineer Name / Role] – Owns: [Feature areas, systems, or modules]

[Engineer Name / Role] – Owns: [Feature areas, systems, or modules]

Sales Team

[Sales Rep Name / Role] – Handles: pricing, plans, offers, demos, contracts, and commercial queries

[Sales Rep Name / Role] – Handles: pricing, plans, offers, demos, contracts, and commercial queries

You should route:

Technical bugs, integrations, product issues → Engineering

Pricing, discounts, plans, demos, negotiations → Sales

Operational Rules

Do not expose internal processes, tooling, or personal details to customers.

Do not mention internal role names unless necessary.

Do not promise timelines unless explicitly provided in the context.

Always prefer clarity, correctness, and minimalism in responses.

If a task is delegated internally, confirm to the customer that it is being handled.
Text fields in the Control Panel are System Instructions for the AI Agent Model. You will achieve best results if you apply good prompt engineering practices. To learn more about prompt engineering, go here: Prompt Engineering Guide

📝 Custom Instructions (Optional)

Custom Instructions Define custom instructions for the AI to follow when responding to customers. These instructions help fine-tune the AI’s behavior for your specific use case and operational requirements. Common Use Cases:
  • Language requirements for internal communication (e.g., tickets and private notes)
  • Notification preferences when creating tickets or notes
  • Response length and style preferences
  • Feature request handling
  • Specific workflow requirements
Example Configuration:
Ticket subjects and private notes should ALWAYS be in English.

If you are raising tickets or notifying team mates via private notes, ALWAYS tell the customer about it.

If you don't have enough relevant context to answer, put a holding message to the customer and notify the team about their query.

No need to raise a ticket or a private note if someone just says 'Hi - I need some help on Periskope.' without prior context.

Your responses should be short and crisp unless details are required to get the point across.

We collect feature responses via this link: https://periskope.featurebase.app/roadmap, feel free to send this across with your requests if someone is talking about a feature that we don't have.

🚫 Custom Restrictions (Optional)

Define more restrictions for the AI to follow when responding to customers. Use this field to specify actions the AI should avoid or limitations on its capabilities. Common Restrictions:
  • Actions the AI cannot perform (e.g., scheduling calls, making refunds)
  • Information the AI should not share
  • Processes the AI should not initiate
  • Limitations on decision-making authority
Example Configuration:
You cannot schedule calls yourself, always inform a sales agent.

Response Quality Factors

The AI Agent’s response quality depends on several key factors: Knowledge Base Quality
  • Well-structured FAQs with multiple question variations
  • Comprehensive documents covering your business domain
  • Regular updates based on new customer queries
Prompt Configuration
  • Clear, specific behavior instructions
  • Appropriate tone and communication style settings
  • Well-defined escalation criteria
Team Role Definitions
  • Accurate team member responsibilities
  • Clear escalation paths for different issue types
  • Updated contact information and availability
Activation Criteria Tuning
  • Balanced activation rules (not too broad or narrow)
  • Custom rules that reflect your business needs
  • Regular refinement based on performance monitoring

Advanced Configuration Tips

Monitor and Adjust
  • Regularly review AI Agent conversations to ensure quality responses
  • Adjust activation rules if the AI is responding too frequently or not enough
  • Fine-tune response delays based on your team’s availability
  • Update knowledge base content based on common customer queries
Best Practices
  • Start with conservative activation rules and gradually expand
  • Ensure internal contacts are properly marked to avoid AI interference
  • Regularly update team member roles and responsibilities
  • Test changes in the playground environment before full deployment
Troubleshooting Common Issues
  • AI not responding: Check activation criteria and ensure AI Agent is enabled
  • Too many responses: Tighten activation rules or increase response delay
  • Wrong team escalations: Update human agent roles and responsibilities
  • Poor response quality: Review and update knowledge base content
Start with default settings and monitor performance. Make incremental adjustments based on actual conversation outcomes rather than making large changes all at once.

Next Steps

After setting up your knowledge base: