Training Meta Business Agent: How to build a robust knowledge base
The Meta Business Agent is intelligent by design. But just how intelligent it appears to be in practice depends almost entirely on what you teach it. The Knowledge Base is the difference between an agent that sounds generic and one that sounds like your best employee. This article shows you how to set it up properly.
This article is part of our comprehensive Meta Business Agent Guide – there you’ll find all the topics relating to the Meta Business Agent at a glance.
What exactly is the Knowledge Base?
The Knowledge Base is the agent’s source of knowledge. Everything the agent knows about your company, your products, your processes and your communications comes from there. Without a well-maintained Knowledge Base, the agent falls back on generic model knowledge, which is rarely a good fit for your company.
Meta allows you to input various types of content: text, documents, website URLs, product catalogues and past chat histories. Each of these sources has its strengths and limitations.
An overview of the six sources of knowledge
Past chats
This is the most valuable source of all – and the one most frequently underestimated. Your past WhatsApp chats contain genuine customer queries in the language customers actually use, your previous responses, and typical conversation flows. An agent trained on this basis will sound authentic right from the start, because they have learnt how your company actually communicates.
Be sure to enable this source during setup, provided your chats contain relevant business communications.
Website-URLs
You can provide the agent with your website URL and it will automatically crawl the content. This is useful for basic company information, product descriptions and general FAQs. The downside is that websites are often written for people, not for AI agents. Long texts, marketing jargon and unstructured content are processed, but not always utilised to their full potential.
Product catalogue
If you have a Meta Commerce catalogue, link it directly. The agent will then be able to access up-to-date product names, prices, descriptions and images, and can recommend and display products directly in the chat. This forms the basis for genuine sales conversations in the chat.
Documents and PDFs
Price lists, product manuals, returns policies, delivery terms, internal FAQs: you can upload all of these as PDF or text documents. These sources are particularly valuable because they are structured and precise. The agent can use them to provide specific, reliable answers.
Facebook-Page
The agent automatically retrieves the content from your Facebook page: company description, posts, events and opening hours. This is useful as a supplementary source, but should not be your sole source of information.
User Manual
This is the most important – and at the same time the most neglected – source of knowledge. Instructions are direct guidelines for the agent: how it should behave, how it should sound, what it should do and what it should not do. More on this in the next section.
Instructions: The decisive factor
You can explicitly tell the agent how to behave. These instructions aren’t just a nice-to-have; they’re the most important tool for turning a generic AI agent into a genuine brand ambassador.
Key points you should include in the instructions:
- Tone and style: Does the agent use informal or formal address? Is their style formal or relaxed? Do they use emojis or not? How long are their replies typically?
- What they should do: Book appointments, recommend products, qualify leads, quote prices.
- What they shouldn’t do: Not offer discounts below X per cent, not make statements about competitors, always involve a human when certain topics arise.
- How they handle unfamiliar questions: Do they admit when they don’t know something? Do they refer the enquiry elsewhere? Do they give a standard reply?
Writing instructions that actually work – rather than just sounding good on paper – is more complex than it sounds. In this video course, we’ll show you the method we’ve developed through hundreds of projects.
The correct order for assembly
Not all content is equally important. If you try to enter everything at once, you’ll quickly lose track of things and end up with an agent who knows a lot but gives unstructured answers.
Our recommended order:
Firstly: Write the instructions. Decide how the chatbot should sound and what it’s allowed to do before you provide it with information.
Next: Link the product catalogue and price list. This is the core of all sales conversations.
Then: Upload frequently asked questions as a structured document. Use the 20 to 30 questions you’re asked most often.
Next: Share past chat histories and add the website URL.
Finally: Documents for specific related topics: returns policy, delivery terms, technical manuals.
A knowledge base is not a one-off project. Prices change, products come and go, and processes are adapted. An agent providing out-of-date information causes lasting damage to trust.
Schedule a regular monthly review. Check whether prices and product information are still correct, whether any new frequently asked questions have emerged that should be added to the knowledge base, and whether the instructions still reflect the company’s current situation.
In our video course, we’ll show you how to set up this process so that it takes no more than 30 minutes a month.
What distinguishes a strong knowledge base from a weak one
A weak knowledge base contains too little information, is unstructured or out of date. The agent will then either give no answer, an incorrect answer, or an answer so generic that the customer ends the conversation.
A strong knowledge base is precise, up to date, well-structured and supplemented by clear instructions. The agent will then provide specific, helpful answers that assist the customer and build trust.
The difference between the two lies not in the quantity of content entered, but in its quality and structure. More is not always better. An agent who knows 500 poorly structured documents often gives worse answers than one who knows 10 precisely structured FAQs.
Anyone wishing to build a knowledge base that really works right from the start will find what they need in the Memacon® Meta Business Agent Package – currently under development. The package will contain the complete methodology: from the correct structure of instructions, through systematic development, to ongoing optimisation. Video courses, guides and best practices – all in one place.
Until then: sign up now for the Memacon® Meta Business Agent newsletter using the button or the QR code below, and don’t miss any updates.
WhatsApp-Newsletter
Regular updates on the
Meta Business Agent
