Training AI
Last updated
Was this helpful?
Last updated
Was this helpful?
Was this helpful?
Zaapi's AI Agent is designed to be your frontline support, handling customer inquiries efficiently and accurately. To unlock its full potential, it's crucial to train it with the right information and instructions.
Training is built on three core components that work together:
Knowledge Source: The brain of your AI. This is where it gets its general information about your business, products, and policies.
Scenario Handling: The playbook for your AI. These are specific, step-by-step instructions for handling common situations.
Personality: The voice of your AI. This defines the tone, style, and specific rules for how the AI communicates.
Let's explore each component in detail.
The Knowledge Source is the foundation of your AI's understanding. It contains all the key information—like FAQs, policies, and details about your business, products, or services—that the AI will use to answer customer questions.
How to Add a Knowledge Source:
Navigate to AI Agent > Train from the side menu.
Select the Knowledge source tab and click + Add knowledge source.
You can add information in three ways:
Upload File: Upload a document containing your information. We support .txt, .csv, .docx, and .xlsx formats. You can download our templates to get started.
Add Website: Scrape information directly from a website page, like your help center or FAQ page.
Write it yourself: Manually type or paste the information directly into the editor.
Connecting Knowledge to Your Brands
Since you might manage multiple stores, it's vital to ensure the AI uses the correct information for each one. When creating a knowledge source, you must select which integrations (e.g., Facebook Page, Instagram account, LINE Official Account) it applies to.
This prevents your AI from sharing information about one brand with customers of another.
Example: Your "Lego Store Return Policy" knowledge source should only be applied to your Lego Store's Facebook and LINE accounts, not your Coca-Cola Store's accounts.
Scenario Handling allows you to train your AI on how to manage common, predictable customer situations with a clear set of steps. This ensures consistency and accuracy for frequent requests.
How to Add a Scenario:
Go to the Scenario handling tab and click + Add scenario.
You can start from pre-built templates for common cases like "Check order status" or "Return or refund," or you can create one from scratch.
When building a scenario, you will define:
When this scenario should trigger: Describe the customer's intent (e.g., "If a customer asks about delivery status"). The AI will use this to identify the right time to use the scenario.
Where should this scenario run: Select the specific integrations (your stores) where this scenario will be active.
How should AI respond: Choose one of two actions:
Follow instructions: Provide a step-by-step guide for the AI to follow.
Escalate to a human agent immediately: Pass the conversation directly to a member of your team.
Example: You can create a "Return Request" scenario. When a customer writes "I want a refund," the AI will follow your instructions, such as asking for the order number and checking the purchase date, before deciding whether to escalate the chat to a human agent.
The Personality defines the tone and character of your AI, ensuring it aligns with your brand's voice. You can create different personalities for each of your brands.
How to Add a Personality:
Go to the Personality tab and click + Add personality.
Assign the personality to the relevant integrations.
Configure the following settings:
Style your AI agent: Describe the overall personality and tone. For example, "You're a calm and witty tech expert who explains things like a helpful friend."
Custom guidelines for response generation: Set specific rules for how the AI replies. For instance: "Always respond in English," "Use emojis sparingly," or "Keep answers short and digestible."
Signature Settings: Add a signature at the end of every AI reply to let customers know they are talking to an AI, such as "Sent by AI Agent."
Example: For a toy store, you might create a "Fun & Friendly" personality that uses emojis. For a corporate B2B service, you could create a "Formal & Professional" personality that avoids slang and provides detailed, structured answers.
By effectively combining Knowledge Sources, Scenario Handling, and Personality, you can build a powerful AI Agent that not only resolves customer issues but also represents your brand perfectly.
PDF files cannot be uploaded.
Marketplace websites (Shopee, Lazada, TikTok Shop) cannot be scraped as a knowledge source.
AI does not use images in responses.
Regularly refining training content ensures the AI delivers accurate and effective customer support.