Skip to main content

How the AI Chatbot Works

CustomerFlows AI is no longer just a single WhatsApp chatbot. It is becoming a shared AI layer that supports:

  • website visitor chat
  • WhatsApp conversations
  • inbox assist for operators
  • lead qualification and routing
  • support guidance grounded in your knowledge base

The same AI foundation can support different channels, but it should not use the same rules in every context. Website visitors, WhatsApp leads, and internal operators may all need different knowledge and different safety boundaries.

The Core Flow​

When a conversation reaches CustomerFlows through website chat or WhatsApp:

  1. The conversation enters CustomerFlows through the channel transport layer
  2. Identity and context are resolved where possible using visitor, session, and contact data
  3. Structured logic runs first for qualification, routing, urgency, and handoff signals
  4. Knowledge retrieval runs next using only sources that are allowed for that context
  5. The AI drafts or sends a reply depending on the channel and workflow
  6. The conversation and AI activity are logged for review, monitoring, and improvement

This keeps deterministic business logic ahead of flexible AI behavior.

Website Chat vs WhatsApp Chat​

The experience differs by channel:

Website chat​

Website chat is best for:

  • product questions
  • pricing questions
  • demo requests
  • early qualification
  • support triage

It can use visitor identity, browsing context, and customer-safe knowledge to answer common questions and route the conversation properly.

WhatsApp​

WhatsApp remains a strong channel for:

  • lead capture
  • qualification
  • follow-up
  • service conversations
  • future inbox handling across human and AI workflows

WhatsApp also has its own messaging and billing rules. See WhatsApp Message Categories and Costs.

What The AI Can Use​

CustomerFlows AI can use:

  • structured conversation data
  • identified contact context
  • customer-safe knowledge sources
  • approved answer rules
  • routing and qualification logic

It should not use:

  • internal-only SOP knowledge in customer-facing replies
  • raw internal CRM data as customer-visible answer content
  • channel-specific behavior as a substitute for real knowledge policy

This is why the knowledge base and inbox should stay separate.

What The AI Captures​

Depending on the channel and the conversation, CustomerFlows AI may capture:

  • contact name
  • email
  • phone
  • service or product intent
  • urgency
  • qualification signals
  • demo or sales readiness
  • support or handoff needs
  • conversation transcript
  • related visitor/session context

That information can support routing and CRM workflows, but customer-facing answers should still come from safe, approved knowledge.

What The AI Can And Cannot Do Today​

It can:

  • answer common website and WhatsApp questions
  • reuse approved answer rules
  • retrieve from managed knowledge sources
  • qualify leads and route conversations
  • assist operators with summaries and draft replies
  • log runs, predictions, and feedback for review

It cannot fully do yet:

  • manage a complete self-improvement loop without operator review
  • expose production-ready bring-your-own-model or API-key controls
  • replace channel-specific provider rules
  • act as a free-form omniscient assistant across every internal and customer workflow

CustomerFlows AI works best when it is grounded, supervised, and given clear boundaries.

How Knowledge Safety Works​

Every knowledge source should be treated as either:

  • Customer-safe for visitor-facing replies
  • Internal-only for operator assist, SOPs, and team guidance

Recommended collections include:

  • general
  • sales
  • support
  • operator
  • team

This gives you a clean way to support both customer-facing and internal AI use cases without mixing them together.

For day-to-day control of knowledge, answer rules, monitoring, and feedback, see AI Console Overview.

Monitoring And Improvement​

CustomerFlows logs AI activity so your team can review:

  • recent runs
  • predictions
  • weak answers
  • missing knowledge
  • unsafe or incomplete behavior

The goal is not to let the AI silently rewrite itself. The goal is to let operators improve the system with a review trail.

FAQ​

Does the chatbot only work on WhatsApp?
No. CustomerFlows AI now also supports website chat flows, and the inbox architecture is being expanded for additional channels.

Is the knowledge base the same as the inbox?
No. The inbox is where conversations arrive. The knowledge base is what the AI is allowed to retrieve from.

Can the AI use internal CRM details in replies?
It can use contact context to help routing and support workflows, but internal-only information should not be exposed in customer-facing replies.

Can I control the AI's answers?
Yes. The recommended path is to use answer rules and managed knowledge sources rather than relying only on raw free-form behavior.

Can the AI improve itself automatically?
Not without supervision. CustomerFlows is moving toward an operator-reviewed improvement loop rather than uncontrolled self-modification.

Need help? Email [email protected].