The Three Buckets

At botBrains, we’ve developed a mental model that helps categorize all customer queries into three buckets.

1. Simple Questions

These are the most basic types of customer inquiries. They ask for information and do not require any customer-specific data.

Examples:

• “How long does shipping take?”
• “What is your refund policy?”
• “What are the opening times?”

Answers to these questions should be direct, consistent, and instantly accessible from your knowledge base.

2. Personalized Questions

These require read access to third-party systems to retrieve user-specific data.

Example:

• “Where is my order?”

Unlike simple knowledge questions, these require:

• Authentication of the user
• Permission checks to ensure the user has access to the requested data
• A system connection to fetch the correct data

This means responses are personalized, and care must be taken to ensure security and data integrity.

3. Tasks

These involve taking action on behalf of the customer, requiring write access to third-party systems.

Examples:

• “Please update my invoice address and resend the invoice.”
• “Cancel my subscription.”

Tasks often involve:

• Permission Checks
• Hard Business Rules to prevent injection

This is the most advanced type of interaction and generally carries the highest risk and operational complexity.

Bucket 2 and 3 usually require collaboration with botBrains. Bucket One is where you’re in charge to monitor conversations and improve answers. We offer two main levers for this:

1. Guidance
2. Knowledge + Insights

Knowledge Gaps

If we look at our three-bucket model, one thing becomes clear: Knowledge is foundational. But in many organizations, knowledge is fragmented or undocumented. Here’s how to systematically improve it:

Step 1: Ingest What You Know

Start by uploading knowledge that you are confident is correct. It’s better to exclude anything that is likely outdated or possibly incorrect, as this may confuse the AI and result in misleading responses. It is better for the system to say “I don’t know” than to answer with outdated or wrong information.

Step 2: Use Data to Identify Gaps

Use botBrains’ analytics tools to:

• Use Topic AI to identify common intents
• Review conversations that escalated to human support.
• Review conversations that are marked as unresolved.
• Monitor your resolution metrics

This will reveal what your bot cannot currently answer and help prioritize what new content to add.

Step 3: Capture Knowledge from Human Conversations

Customer calls, video chats, and support sessions are goldmines for knowledge. With permission, record these sessions and:

• Transcribe them