AI

Partha
By Partha
2 articles

AI Brain & Knowledge Base

1) What is AI Brain in Thunaivi? AI Brain is your support knowledge layer. It uses your FAQs, policies, and process content to generate accurate replies across WhatsApp, Instagram, Facebook, Telegram, and Website Chat. 2) How to add knowledge content - Go to Client Portal > AI Brain - Click Add Knowledge - Add content using: - FAQ text - Policy docs - Website URLs - Internal process notes - Save and publish - Test with real customer-style questions 3) Recommended content format Use clear Q&A format: - Q: What is your refund policy? - A: Refund is allowed within 7 days for eligible plans... Best practice: - Keep each answer short and specific - Add exact limits, timelines, and exceptions - Avoid vague text like “usually” or “maybe” 4) Priority knowledge topics (must-have) - Pricing and plans - Trial and billing rules - Refund/cancellation policy - Setup/onboarding steps - Channel setup requirements - Escalation process to human team - Support contact details 5) How to improve AI accuracy - Add top 50 real customer questions first - Include synonyms and spelling variations - Add “do not answer directly” topics (force handoff) - Review wrong replies weekly and update content - Remove outdated policy text immediately 6) Human handoff safety rules Set AI to escalate for: - Payment disputes - Legal/compliance questions - Angry/frustrated users - Security/privacy requests - Any unknown/high-risk request 7) Versioning and change control - Keep one owner for knowledge approval - Update with change log (date + reason) - Test before publishing major edits - Roll back quickly if answer quality drops 8) Knowledge QA checklist - Is the answer factually correct? - Is policy up to date? - Is response short and clear? - Is escalation trigger defined? - Is link/contact path included if needed? 9) Common mistakes - Uploading marketing copy instead of support answers - Not updating knowledge after pricing changes - No escalation rules for sensitive cases - Too long answers that confuse users 10) Monthly maintenance workflow - Export top failed intents - Fix missing/weak knowledge entries - Add new FAQs from live conversations - Re-test top 20 intents - Track improvement in deflection and CSAT

Last updated on Mar 05, 2026

AI Replies & Human Handoff

1) How AI Replies Work in Thunaivi AI reads your AI Brain (FAQs, policies, support docs), detects intent, and sends a response when confidence is high. If confidence or risk is low/high, it escalates to human. 2) When AI should auto-reply - FAQs - Pricing basics - Trial steps - Setup instructions - Order/status type repetitive queries Keep these automated for speed and consistency. 3) When AI must handoff to human - Billing disputes/refunds with exceptions - Legal/compliance requests - Security/privacy incidents - Angry/frustrated conversations - Unknown/ambiguous requests Use strict escalation rules for these cases. 4) Confidence-based routing model Recommended: - High confidence: AI sends response - Medium confidence: AI drafts + agent approval - Low confidence: direct human handoff This balances speed and safety. 5) Human handoff workflow - AI marks conversation as Escalated - Adds concise handoff summary - Assigns to correct queue/agent - Notifies owner/team lead - Agent replies with full context available No context loss between AI and human. 6) Handoff summary template Include: - Customer issue in 1 line - Detected intent - AI responses already sent - Missing info needed - Recommended next action This reduces agent response time. 7) Prevent wrong AI replies - Keep knowledge base updated - Add “restricted topics” list - Use response guardrails - Limit AI on risky intents - Weekly audit failed conversations 8) AI tone and brand control Define: - Tone (professional/friendly) - Response length - Language preference - Forbidden claims - Escalation wording Consistency improves customer trust. 9) KPI tracking for AI + handoff Track weekly: - AI deflection rate - Escalation rate - Wrong-answer rate - First response time - CSAT for AI-only vs escalated chats Use these to tune automation safely. 10) Recommended rollout plan Week 1: AI only for top FAQs Week 2: Add routing + draft mode Week 3: Expand to more intents Week 4: Optimize handoff quality + SLA alerts

Last updated on Mar 05, 2026