Business Requirements

Problem Statement

Organizations need to provide intelligent customer support and information access through public-facing websites without requiring users to log in. The Website module enables public AI chatbots powered by knowledge bases, allowing customers to find answers instantly while providing public pages for marketing and information.

Use Cases

1. Public Customer Support Chatbot

Actor: Website visitor (anonymous)

Flow:

  1. Visitor lands on website with embedded chatbot
  2. Visitor asks a question (e.g., “How do I reset my password?”)
  3. System searches knowledge base for relevant answers
  4. LLM generates response based on knowledge base context
  5. Response displayed to visitor in real-time
  6. Visitor can continue conversation or get human support

Acceptance Criteria:

  • No login required
  • RAG retrieves relevant knowledge base articles
  • Response quality matches knowledge base content
  • Conversation history stored locally on client
  • Fallback to human support link if needed
  • Response time < 5 seconds

2. Public FAQ/Helpdesk

Actor: Support team, customers

Flow:

  1. Support team configures AI chatbot with FAQ content
  2. Customers use chatbot to find answers
  3. Chatbot searches knowledge base with customer queries
  4. Answers automatically generated from relevant docs
  5. Analytics show which questions are most common
  6. Support team uses data to improve knowledge base

Acceptance Criteria:

  • Knowledge base integration working
  • Relevant document retrieval
  • Answer generation from documents
  • Query analytics tracking
  • No hallucinations (only from knowledge base)

3. Marketing Landing Pages

Actor: Marketing team, website visitors

Flow:

  1. Marketing team creates landing pages with CMS
  2. Pages published to public website
  3. Visitors view marketing content
  4. Chatbot available for product questions
  5. Chat leads captured for sales follow-up

Acceptance Criteria:

  • Page customization available
  • SEO meta tags configurable
  • Responsive design on mobile
  • Chatbot integration seamless
  • Lead capture working

4. Tenant-Specific Customization

Actor: Tenant administrator

Flow:

  1. Admin configures AI chatbot for their tenant
  2. Selects LLM model (GPT-4, GPT-3.5, etc.)
  3. Configures system prompt (tone, behavior)
  4. Sets temperature and token limits
  5. Provides knowledge base URL/API
  6. Deploys chatbot on public website

Acceptance Criteria:

  • Easy configuration interface
  • Model selection with cost/performance tradeoffs
  • Prompt customization for brand voice
  • Knowledge base integration verified
  • Configuration tested before deployment

Business Rules

Chatbot Rules

  1. Knowledge Base Constraints:
    • Chatbot only answers from configured knowledge base
    • No general knowledge (avoid hallucinations)
    • Confidence score tracking
    • Fallback to human support if uncertain
  2. Response Generation:
    • Must cite source document when possible
    • Limit response length (configurable)
    • Include links to full articles
    • Suggest related topics
  3. Conversation Management:
    • Chat threads stored per session
    • Local storage for client-side persistence
    • Optional user identification
    • Configurable chat history limits

Configuration Rules

  1. Model Selection:
    • Per-tenant model choice
    • Cost tracking per query
    • Rate limiting per tenant
    • API key management
  2. Customization:
    • System prompts editable per tenant
    • Temperature and token limits configurable
    • Language support (multi-language)
    • Timezone awareness
  3. Knowledge Base:
    • Dify integration supported
    • Custom provider support
    • Document indexing
    • Relevance scoring

Public Access Rules

  1. No Authentication Required:
    • Visitors don’t need to log in
    • Anonymous chat capability
    • Optional email capture
    • GDPR-compliant tracking
  2. Rate Limiting:
    • Per-IP rate limits
    • Prevent abuse/spam
    • Graceful degradation when limits hit
    • Whitelist for trusted sources
  3. Content Moderation:
    • Input validation and sanitization
    • Offensive content filtering
    • Spam detection
    • GDPR compliance

Key Metrics & KPIs

Engagement Metrics

  • Chatbot sessions per day
  • Average session duration
  • Questions per session
  • User satisfaction (CSAT)
  • Resolution rate (answered vs. escalated)

Content Metrics

  • Knowledge base coverage
  • Document hit rate
  • Missing answer queries (gaps)
  • Answer relevance score
  • Hallucination rate

Business Metrics

  • Support ticket reduction
  • Cost per interaction
  • Customer satisfaction improvement
  • Lead generation (email captures)
  • Time to answer

Technical Metrics

  • Response latency
  • System uptime
  • API error rates
  • Knowledge base search performance
  • LLM API usage and costs

Success Criteria

  1. Usability:
    • Chat interface loads in < 1 second
    • Response generation < 5 seconds
    • Mobile-friendly interface
    • Intuitive navigation
  2. Quality:
    • Answer relevance > 85%
    • Hallucination rate < 5%
    • User satisfaction > 4.0/5.0
    • Support ticket reduction > 20%
  3. Reliability:
    • 99.9% uptime
    • Knowledge base searchable
    • No message loss
    • Graceful error handling
  4. Security:
    • GDPR compliance
    • Data privacy maintained
    • Secure credential storage
    • No data leakage
    • Input sanitization
  5. Performance:
    • Scales to 1000s concurrent users
    • < 500ms document search
    • < 2s response generation
    • Efficient caching

Constraints

  1. Knowledge Base:
    • Dependent on knowledge base quality
    • Limited to configured documents
    • Search accuracy limited by indexing
    • Maintenance overhead
  2. LLM Integration:
    • Dependent on external APIs
    • API rate limits apply
    • Token usage costs
    • Model limitations and hallucinations
  3. Public Access:
    • Cannot restrict to authenticated users
    • Subject to abuse/spam
    • Privacy compliance required
    • Rate limiting necessary
  4. Content:
    • Limited to text-based content
    • No image/video support initially
    • Document format limitations
    • Language limitations per model

Integration with Multi-tenant Platform

Website module:

  • Per-tenant Configuration: Each tenant customizes chatbot
  • Shared Database: Uses same PostgreSQL as main app
  • Shared LLM Budget: Optional cost allocation per tenant
  • Public Access: No tenant boundaries for visitors
  • Separate from Authenticated Areas: Public pages separate from internal apps

Future Enhancements

  • Multi-language support per knowledge base
  • File upload for knowledge base documents
  • Analytics dashboard for admins
  • A/B testing for prompts and responses
  • Voice/audio chat support
  • Integration with support ticket systems
  • Conversation branching and suggestions
  • Custom branding and theming
  • Video and image content support
  • Sentiment analysis and feedback
  • Integration with CRM for lead capture

Back to top

IOTA SDK - Multi-tenant Business Management Platform