iSales AI

Data Preparation

Mastering AI Memory: The Ultimate Guide to Preparing Content for Your Business Bot

Your AI Business Bot is only as smart as the data you feed it. To provide accurate, context-aware answers, the bot uses a process called RAG (Retrieval-Augmented Generation). Before “remembering” your files, the bot cuts them into smaller pieces called “chunks.”

Our system uses Smart Chunking technology that automatically detects the type of content you upload—whether it’s an FAQ, a sales script, or a technical manual—and splits it in the most logical way. However, you can significantly improve the bot’s performance by formatting your documents correctly.

This guide covers how to prepare your files for the four main strategies our bot uses: FAQ, Dialogue, Structured Documents, and General Text.

1. Preparing FAQs (Question & Answer Format)

Best for: Knowledge bases, list of common questions, support cheat sheets.

The bot has a specialized FAQ Chunker that keeps questions and their specific answers tied together forever. If you split them up, the bot might find the question but lose the answer.

✅ How to Format

  • Use Clear Prefixes: Start questions with Q:, Question:, or Вопрос:. Start answers with A:, Answer:, or Ответ:.
  • Keep Them Close: The answer should immediately follow the question.
  • One Pair at a Time: Don’t group multiple questions together before giving answers.

📝 Example (Good)

Q: What are your working hours?
A: We are open Monday to Friday, from 9 AM to 6 PM EST.

Q: Do you offer refunds?
A: Yes, we offer a full refund within 30 days of purchase if you are not satisfied.

Why this works: The bot detects the “Q/A” pattern and treats each pair as a single “memory unit.”

2. Preparing Sales Scripts & Dialogues

Best for: Training the bot on how to sell, handle objections, or follow specific conversation flows.

The Dialogue Chunker is designed to understand the flow of conversation between two parties (e.g., a Manager and a Client). It preserves the context of “Who said what.”

✅ How to Format

  • Use Role Names: Explicitly state who is speaking using colons.
    • For the User side: Client:, User:, Guest:, Customer:.
    • For the Bot side: Manager:, Bot:, Assistant:, Support:.
  • Use Numbering (Optional): If you have a sequence of steps, use numbering like 1.1, 1.2 to show the flow.

📝 Example (Good)

# Handling Pricing Objections

Client: It's too expensive.
Manager: I understand price is a concern. However, considering the ROI of 300% within the first month, the investment pays for itself.

Client: Do you have a discount?
Manager: We can offer a 10% discount for annual commitments.

3. Preparing Structured Documents (Manuals & Guides)

Best for: Employee handbooks, technical documentation, complex product descriptions.

For large documents, context is king. If the bot reads a paragraph about “Settings,” it needs to know which settings (e.g., “Profile Settings” vs. “Notification Settings”). Our Structure-Aware Chunker uses headers to attach this context to every piece of text.

✅ How to Format

  • Hierarchy Matters: Don’t skip levels (e.g., don’t go from H1 directly to H3).

📝 Example (Good)

# Employee Handbook

## 1. Onboarding
### 1.1 First Day
On your first day, please report to HR at 9:00 AM...

## 2. Benefits
### 2.1 Health Insurance
We provide full coverage...

Why this works: Even if the bot retrieves just the text “We provide full coverage…”, it will attach the context: [Employee Handbook > Benefits > Health Insurance].

4. General Tips for All Files

Regardless of the content type, following these “Golden Rules” ensures the fastest processing and highest accuracy.

  1. Preferred Formats: .txt, .md (Markdown), and clean .docx are best. .pdf is supported but can sometimes contain hidden layout issues.


  2. Avoid “Walls of Text”: Break long texts into smaller paragraphs (3-5 sentences).

  3. Check Your Encoding: Ensure files are saved in UTF-8 format. This supports all languages and special characters correctly.

  4. Remove Visual Noise: Remove page numbers, headers/footers repeated on every page, and decorative images before converting to text.

Summary: Help the bot help you. Structure your data logically—use Q&A for FAQs, Role names for scripts, and Headers for manuals—and your AI assistant will become significantly more intelligent and responsive.