Enhancing Your AI Agent with Knowledge Sources

Overview

The CRM's Conversation AI feature has been upgraded with new Knowledge Sources and a smarter retrieval system. This allows your AI agents to access a wider variety of information, including structured tables, formatted documents, and custom text, enabling them to deliver more precise and context-aware answers to your contacts.

Key Benefits

These enhancements provide several major advantages for your automated conversations.

• Improved Accuracy: A new re-ranking process prioritizes the most relevant information, leading to more correct answers.
• Quick Updates: You can add or modify knowledge sources immediately without needing to retrain your AI bot from the beginning.
• Flexible Content: Support for different data types like spreadsheets, PDFs, and rich text means you can teach your AI from almost any document.
• Full Transparency: Every AI response includes clickable source links so you can see exactly which piece of information was used to formulate the answer.

Understanding the Smarter Retrieval Process

The system now uses a two-step process to find the best information. First, it performs a broad search through your knowledge base. Then, a re-ranking layer analyzes the results to score them based on how closely they match the user's specific question. Only the top-scoring information is sent to the AI to generate a reply, which reduces errors and keeps responses focused.

Supported Data Types

You can populate your Knowledge Base with several types of content.

• Tables: Upload CSV files containing structured data like price lists, product catalogs, or service details.
• Rich Text: Write or paste content directly into the editor for things like company policies, how-to guides, or frequently asked questions.
• File Upload: Supported file types include PDF, DOC, DOCX, and others. The system extracts and indexes the text from these documents.
• Other Sources: You can also use Web Crawler, FAQ, and Web Search sources to pull in information from your website or the internet.

Viewing Source Attribution

To maintain transparency, you can see which knowledge sources were used for any AI response. Inside a conversation, click the "Response Info" icon next to an AI-generated message. A side panel will open showing the specific text chunks that informed the answer, along with the source file name or URL. This panel also provides quick-edit options if you need to update the source material.

How to Set Up and Use Knowledge Sources

Step 1: Create or Edit a Knowledge Base

Navigate to the AI Agents section from the main menu. Click on the Knowledge Base tab. Here, you can click "+ Create Knowledge Base" to start a new one, or click on the name of an existing Knowledge Base (or the three-dot menu) to edit it.

Step 2: Add Your Knowledge Sources

Inside your Knowledge Base, click the "+ Add Source" button. Select the type of source you want to add (e.g., Tables, Rich Text, File Upload) and follow the prompts to upload your file or input your text.

Step 3: Connect the Knowledge Base to an AI Bot

Go to AI Agents and select Conversation AI. From the Agent List tab, either create a new bot or edit an existing one. Within the bot's settings, go to the Bot Training tab. In the Knowledge Base section, select the Knowledge Base you have created or updated.

Step 4: Test the Knowledge Retrieval

It is crucial to test the setup. Open your Knowledge Base from the AI Agents section and use the Knowledge Base Retrieval Tester. Enter sample customer questions to see which answers and sources are returned. If the results are not correct, update your Knowledge Base content and test again. Once retrieval is working properly, test the full bot experience.

Step 5: Inspect Sources in a Live Conversation

After your bot is live, go to the Conversations section and open a chat where the AI responded. Click the "AI Response Info" button next to the message and select "Knowledge Chunks" to view the specific information used for that reply.

Frequently Asked Questions

Do I need to retrain the bot after adding a new source?

No. Once the new source is saved and has finished processing, the Knowledge Base will automatically include it. It is recommended to use the Retrieval Tester to confirm the new information is being found correctly.

What file types are supported for upload?

The system supports PDF, DOC, DOCX, PPT, PPTX, TXT, and CSV files.

How many columns can a table have?

You can import CSV files with up to 500 columns, but you can select a maximum of 20 columns to be used by the AI.

Can I mix different source types in one Knowledge Base?

Yes. You can combine tables, rich text, files, and other sources within a single Knowledge Base. The retrieval system will treat all content equally and use re-ranking to find the best matches.