successful chatbot in five steps

How to create a successful chatbot in five steps

More and more companies are starting to use chatbots to automate part of the customer service or to automate lead generation. Chatbots have a bright future ahead of them, but you still often hear about people talking to a chatbot that is not capable of offering the right help. This leads to frustration for the customer that could’ve easily been prevented, provided you know what you are doing.

In this blog we explain the steps that are necessary to turn your chatbot into a super smart customer service employee!

Add knowledge

The way to a successful chatbot starts with the amount of knowledge it possesses. This is the step where you fill the brain of your new digital employee with questions he needs to understand and the answers he has to give to those questions. At Watermelon this works as follows:

add knowledge

You start by formulating the questions that your customers will be asking. You do not do this in one, but at least three different ways. This is necessary because people will ask the same question in different ways and the chatbot must understand those variations of course. This does not mean that the chatbot only understands exactly these three variations. Watermelon’s chatbot has AI (artificial intelligence) and can therefore understand many more variations on the basis of the three variations given. The chatbot looks for the intention of the sentence and does not pay attention to the order of words or keywords. The answer given by the chatbot is always the answer that you have inserted in the answer field.


After you have added all questions and answers, it’s time to train the chatbot. This too is very simple and can be done with a single push of a button. When you start training, the chatbot trains his brain with all the newly added knowledge. In a few minutes, the chatbot analyzes all the variations in order to discover the intentions. This process is fully automatic and runs in the background so you can do other things in the meantime.

Training Chatbot


After the chatbot has finished training, he possesses all the knowledge that you have added in the first step. You can put that to the test in this step! There is room in the Watermelon dashboard to chat in a closed-off environment with your own chatbot to test his knowledge. You can check here whether he answers the questions that have added and to what extent he already understands the different variations to the question. If you are satisfied with the result you can go to the next step. If the chatbot does not know enough, you can always go back to the first step to add more knowledge.

Go live

If the test has been satisfactory, it is time for the most exciting part. Putting your chatbot live! Releasing your new chatbot to the world can be daunting, but it is an essential part of improving the chatbot. That is why you should not wait too long to put it live!


Now that the chatbot is live, it’s time for the fun to begin. The chatbot can only really get smart when he talks to your customers. You will find out that your chatbot will struggle a bit in the beginning, but thats ok! Because what he does is remember what questions he has not understood. Again artificial intelligence comes into play here, because the chatbot will show you these questions to you as so-called ‘mismatches’. Then it is up to you to choose whether the chatbot should learn these questions or not. This way, the input of your customers play an important role in optimizing the chatbot.


To set up an optimal chatbot, it is crucial that you continuously repeat these five steps. You add knowledge, you train this knowledge, test the knowledge, put the chatbot live and let the chatbot save mismatches. You add these mismatches to the knowledge of the chatbot. You are never completely finished improving your chatbot, but as you progress in the process, fewer and fewer mismatches come forward, making it less and less work!

Curious about what Watermelon can do for you? Book a free demo here!

Related articles