Skip to main content

Knowledge Assistant — External

Knowledge Assistant (External) is a customer-facing variant of a RAG system. It helps customers find answers in product documentation, FAQs, and the support knowledge base without contacting an agent.

Purpose​

The external Knowledge Assistant addresses the scaling of customer support:

  • reducing the load on agents
  • instant answers to routine questions
  • 24/7 availability
  • consistent answer quality

Use Cases​

Typical scenarios for a customer-facing Knowledge Assistant:

  • Support chatbot — answering customer questions about products and services
  • Documentation assistant — navigating product documentation
  • Self-service portal — independently resolving routine problems
  • Pre-sales consultations — answering questions about product capabilities

Data Sources​

The external Knowledge Assistant works with the following sources:

  • public product documentation
  • FAQs and knowledge bases
  • user manuals and guides
  • support request history (anonymized)
  • descriptions of pricing plans and terms

Key Challenges​

A customer-facing Knowledge Assistant imposes higher quality requirements:

  • Hallucination risk — customers must not be given inaccurate information; the stakes are higher than in the internal variant
  • Brand tone and voice — answers must match the organization's communication style
  • Escalation to an agent — the system must be able to determine when a conversation needs to be handed over to a human
  • Multilingual support — supporting multiple languages when working with international customers
  • Legal constraints — answers must not contain promises or commitments that the organization cannot fulfill
  • Personalization — taking the customer's context into account (product, plan, interaction history)

Escalation​

It is critically important to define escalation rules:

  • requests that fall outside the knowledge base
  • complaints and negative requests
  • questions that require access to customer data
  • situations in which the system is not confident in its answer

On escalation, the agent must receive the full context of the conversation.

Metrics​

The effectiveness of a customer-facing Knowledge Assistant is measured by the following metrics:

  • Deflection rate — percentage of requests resolved without an agent
  • CSAT (Customer Satisfaction) — customer satisfaction score
  • First response time — time to the first response to a request
  • Answer accuracy — percentage of correct and complete answers
  • Escalation rate — percentage of requests handed over to an agent
  • Resolution rate — percentage of requests fully closed by the system