Knowledge base for an AI agent
The knowledge base of an AI agent is a centralized file where frequently asked questions, model answers, product data, policies, and all the information that the artificial intelligence consults to suggest responses are stored. In the WhatsApp Business ecosystem, this base is what gives the bot context to understand what to respond, but be careful: it is not a rigid script, but rather a reference material that the AI uses to craft more accurate responses.
Why it matters to your SME
If you serve customers on WhatsApp, it has surely happened to you: they ask the same thing a thousand times (hours, prices, shipping, payment methods). Without a knowledge base, your AI agent responds haphazardly or has nowhere to get the info. With a well-built base, AI suggestions are more useful and the team just has to approve with a click. Plus, you save time and the customer does not get upset waiting for an answer that never comes.
Concrete example
Imagine a clothing store that sells via Instagram and WhatsApp. Its knowledge base includes: return policy, size chart, shipping cost by area, and template responses for 'do you have this jean in another color?'. When a message arrives, the AI searches that base and suggests a response that the seller approves or adjusts. If the customer asks something not in the base, the AI says so and the human takes over. That simple.
Common mistakes when building it
- Copying and pasting endless texts: the AI does not need a 50-page manual, but rather short, direct responses.
- Not updating it: if the price or hours change, the base becomes outdated and suggestions lose meaning.
- Using only frequently asked questions: it is also useful for answering product questions, complaints, or shipping inquiries.
- Thinking it replaces the human: the base is support, not a robot that handles everything alone without supervision.
How to start without dying in the attempt
- 1Review the last 50 WhatsApp chats and note down the most frequently asked questions.
- 2Write 2-3 sentence answers, clear and in your brand's tone.
- 3Group them by topic: sales, shipping, complaints, hours, etc.
- 4Upload them to your management tool (or a Google Sheets if you're just starting out).
- 5Update it whenever something important changes and test with real messages.
In Wando, the AI learns from the answers you approve, so the base adjusts itself over time. It's not magic: it's hot training.
How it's used in practice
Imagine a real estate agency in Cordoba that receives 30 inquiries per day on WhatsApp. Without a knowledge base, the AI agent doesn't know what to say when asked about the price of a rental or the requirements to rent. With a built base, the agent suggests answers with exact data: 'The rent for the apartment in Nueva Cordoba is $450,000 per month, includes HOA fees. The requirements are a pay stub and a property guarantee.'
The team approves with one click and that's it. If the customer asks something not in the base, the AI says it doesn't know and a human takes over the chat. This way, the business serves faster without losing control.
The key: the base updates itself. Wando learns from the answers you approve, so over time the suggestions become more and more accurate.
What it's commonly confused with
Many people think the knowledge base is the same as a flow chatbot. It's not. A flow chatbot follows a fixed script: 'if the customer says X, reply Y'. It's rigid and has to be programmed for each case.
The knowledge base, on the other hand, is an information file that the AI consults to build responses. It's flexible: if the customer asks something that wasn't anticipated, the AI can combine data from the base and suggest a useful answer.
Another common confusion is thinking that the base replaces the human. No: it's a support. The AI suggests, the human approves. That's what's called human-in-the-loop, and it's the difference between a reliable system and one that makes up answers.
In short: the flow chatbot is a script, the knowledge base is a library. Wando uses the latter.
What happens if you ignore it
If you don't have a knowledge base, your AI agent has no source of information. The result: generic answers, errors, or awkward silences.
The customer who asks about a price and gets 'I don't have that information' will go to the competition. That translates into lost sales and a reputation that suffers.
Furthermore, without a base, the team wastes time correcting responses or writing from scratch. The savings that AI promised vanish.
The concrete consequence is simple: an agent without a knowledge base is an empty promise. It won't save you time or improve service. That's why, if you're going to use AI on WhatsApp, the base is the first step, not an extra.
How it relates to Meta's official API
The knowledge base does not replace the WhatsApp API: it uses it. Meta's official API (WhatsApp Cloud API) is the channel through which messages travel. The knowledge base is the brain that decides what to respond.
When a customer writes to you, the message enters through the API, the AI consults the base, puts together a response, and sends it back through the API. Without the API, there is no channel. Without the base, there is no intelligence.
One detail: approval of message templates (HSM) depends on Meta and can take up to 48 hours. The knowledge base does not influence that, but it does help you draft templates that comply with Meta's policies.
At Wando, integration with the official API is automatic. You handle the base; we handle the rest.
When it does NOT apply or is not advisable
Not every business needs a knowledge base. If you handle 5 messages a day and respond to them yourself on the spot, it might not be worth setting one up.
It is also not advisable if your business is so specialized that every inquiry is unique and does not repeat. In that case, the base will not save you time, because there are no frequently asked questions to automate.
Another situation: if you are not willing to keep it updated. An outdated base is worse than having nothing, because the AI will suggest prices or schedules that are no longer valid.
The rule is simple: the knowledge base pays off when there is volume and repetition. If there isn't, it's a waste of effort that you can avoid.
Common mistakes of those who are just starting to understand it
The biggest mistake is thinking that the knowledge base is a 50-page manual. No: they are short, direct answers, 2 or 3 sentences long. The AI doesn't need a book, it needs clear data.
Another mistake is not updating it. If the price or schedule changes, the base becomes outdated and the suggestions lose meaning. What was valid yesterday may be a disaster today.
There's also the one who believes that the base replaces the human. That's not the case: the AI suggests, the human approves. Without supervision, the AI can invent answers, and that is a risk.
Finally, many copy and paste texts from the web. That doesn't work: the base must have your tone and your real info. Otherwise, the customer notices that it is a robot.
Frequently asked questions
What's the difference between a knowledge base and a flow chatbot?+
A flow chatbot follows a fixed script (if this happens, respond that), while a knowledge base is a file of information that the AI consults to generate responses. The base allows more flexibility and adapts better to unexpected questions.
Do I need to be a programmer to create a knowledge base?+
No. Most tools, like Wando, let you load frequently asked questions and answers from a simple panel, without writing code. The important thing is to keep the information clear and up to date.
What happens if the AI doesn't find the answer in the base?+
In a human-supervised setup, the AI can say it doesn't know or route the query to a person. It should never make up information. That's why having a good response approval process is key.
How long does it take to build a knowledge base?+
It depends on the size of the business. For an SMB, spending a few hours reviewing chats and writing responses can give you a functional base in a day. After that, it gets refined with use.
Does the knowledge base update itself?+
In Wando, yes. The AI learns from the responses you approve, so the base adjusts over time. But it's always a good idea to review and update it when prices or hours change.
Can I use a knowledge base without knowing how to code?+
Yes. Most tools, like Wando, let you load frequently asked questions and answers from a simple panel. No coding required.
How much does the knowledge base cost?+
It depends on the tool you use. In Wando, plans start at US$9.99 per month and you can try it free for 14 days without a card. Check current plans at wando.online/pricing.