Skip to main content

AI Console Overview

The AI Console is the control panel for CustomerFlows AI. It gives workspace admins one place to manage the knowledge your AI can use, define approved answers for common questions, review AI behavior, and improve results under operator supervision.

The AI Console is separate from your inbox. Inbox channels like website chat, WhatsApp, SMS, and email bring conversations into CustomerFlows. The AI Console controls the knowledge, guardrails, and review workflows those channels depend on.

What You Manage In The AI Console​

1. Answer rules​

Answer rules are short, approved responses for specific questions your team wants answered consistently.

Examples:

  • "What do you do?"
  • "How does pricing work?"
  • "Can I book a demo?"

These rules are useful when you want a predictable answer instead of relying only on general knowledge retrieval.

2. Knowledge sources​

Knowledge sources are the documents and text content CustomerFlows AI can search before drafting or sending a reply.

Common examples:

  • support FAQs
  • service details
  • onboarding guides
  • pricing notes
  • internal SOPs

When you upload or edit a source, prepare it before relying on it in live answers. Preparing makes the content searchable for the AI and helps the console show whether the source is ready, still processing, or needs attention.

3. Monitoring​

The AI Console shows recent runs, predictions, and review activity so your team can see what the AI has been doing and where it needs improvement.

4. Supervised feedback​

Operators can flag weak, unsafe, or incomplete AI behavior and record the preferred outcome. This creates a review trail that helps improve the system without silently changing customer-facing behavior.

Improving a Bad Answer​

If the AI gives a weak or incomplete answer:

  1. Open the conversation or run in the AI Console
  2. Mark the answer for review
  3. Write the preferred answer or explain what was missing
  4. Add or update the supporting knowledge source
  5. Prepare the knowledge source again
  6. Add an answer rule if the question should always receive a predictable response
  7. Test the question before relying on the change in live conversations

Use answer rules for exact, repeated questions. Use knowledge sources when the AI needs broader context across documents.

Knowledge Safety Model​

CustomerFlows separates knowledge from channels.

  • Channels are where conversations happen: website chat, WhatsApp, SMS, email, and future inbox sources.
  • Knowledge is what the AI is allowed to use when helping visitors or operators.

Every knowledge source should be treated as one of two types:

  • Customer-safe: allowed to support visitor-facing replies
  • Internal-only: reserved for operator assist, SOPs, and internal workflows

Customer-safe and internal-only sources should not be mixed casually. If a document contains team-only notes, escalation rules, or internal process steps, mark it internal-only.

As a simple rule: if a customer could read the document without exposing private process, pricing exceptions, or staff instructions, it may be customer-safe. If it is written for your team, keep it internal-only.

Suggested Knowledge Collections​

We recommend organizing knowledge into separate operational collections:

  • General for broad company and product answers
  • Sales for pricing, demos, and pre-sales qualification
  • Support for troubleshooting and service guidance
  • Operator for agent assist and internal workflows
  • Team for SOPs, internal help, and process documentation

This helps your team reason about which knowledge belongs in visitor-facing flows and which knowledge should stay behind the scenes.

Inbox vs Knowledge Base​

These are different parts of the product:

  • Inbox: where messages from channels are received and worked
  • Knowledge base: what the AI can retrieve and use

A WhatsApp message might arrive in the inbox, but the answer should come from the right customer-safe knowledge. An operator assist suggestion might use the same conversation plus internal-only SOP knowledge. The channel stays the same; the retrieval policy changes.

Plan Access​

AI access is plan-aware.

Depending on your workspace tier, CustomerFlows may limit:

  • how many AI responses are included each month
  • how many knowledge sources you can manage
  • whether supervised feedback tools are available
  • whether advanced preparation, monitoring, or feedback actions are enabled

If your workspace plan does not include an AI feature, the console may show the section but block the action until the workspace is upgraded.

Current Limitations​

The AI Console is still evolving. Areas still being expanded include:

  • deeper retrieval diagnostics
  • stronger performance dashboards
  • provider and model configuration
  • automated application of approved feedback
  • richer multi-language behavior

Need help? Email support@customerflows.com.