Skip to navigation

AI Replies and Bot Configuration

6MM AI Customer Support · Tenant guide

View as Markdown

Give AI reliable business context

Use System Settings → Intelligent Services for service switches and Bot Configuration for reply behavior. Menu availability and enabled capabilities depend on your tenant configuration and account permissions.

The reply pipeline first handles special intents, then evaluates enabled matching stages. FAQ or vector matches can answer before the large-model stage. The model is not invoked for every message. Use the flow test to see which path answered.

ConfigurationPurpose
Basic settingsConfigure the bot’s general behavior using the available controls
FAQ repliesMaintain eligible question sources and similar phrasings
AI vector repliesMatch semantically related questions; review thresholds and index readiness
AI large-model repliesUse retrieved business knowledge and conversation context to compose an answer
Identity and capability repliesMaintain approved answers to questions such as who the assistant is and what it can do
Reply buttonsChoose eligible content sources for suggested follow-up buttons

Improve answer quality

Write articles with a clear question, applicable audience, prerequisites, steps, exceptions and escalation path. Keep fees, limits and timestamps explicit. Remove conflicting outdated rules. Provide related facts together, especially deposit networks, transfer conditions and account verification requirements.

Ask a documented question, a follow-up and an ambiguous question in the visitor chat. Check whether the answer preserves your actual rules and asks for clarification when information is insufficient. Do not place passwords, signing secrets or internal operational instructions in customer-facing knowledge.

Diagnose a weak reply

A matched FAQ can return its configured text without using the large model. Update that content when it is wrong. If the model is not used, check the actual matched step and service switches. For vector failures, inspect retrieval status and service availability before rebuilding. Index updates and rebuilds may consume intelligent-service balance under the configured billing rules.