/ What is a smart chatbot? Artificial intelligence!
Alexander Wijninga
|
5 min
Keeping a customer satisfied generally works best by giving your customers the right time and attention and being reachable when the customer wants it. Very valuable, but it does take a lot of time. That is why companies use various ways of being easily reachable and serving customers as fast as possible. One of those ways is deploying a smart chatbot. An effective and easy-to-manage way of being reachable 24 hours a day. Even so, there are still companies that have some doubts about the effectiveness of chatbots. Understandable, but really not necessary. Because chatbots are more versatile and smarter than you think!
In this article we explain how chatbots work and what effect they have. In doing so we take Watermelon's software and chatbots as an example. With Watermelon's software a customer can make their own chatbot, without needing any technical knowledge for it. Watermelon makes use of artificial intelligence, also known as AI, to make chatbots ever smarter and let them work more efficiently.
For the record: a chatbot is of course not a real robot sitting behind a computer. It is a fully automated and digital conversation partner, which can be used to lighten the workload of customer service staff, for example. On the basis of information from the FAQ, the chatbot handles the most common and obvious customer questions, so that the service employee can focus on the more complicated matters.
In the Watermelon Dashboard a chatbot can be added to various channels such as live chat (the chat on your company website itself), but also apps such as Telegram and Facebook Messenger. Via this system the chatbot can quickly answer your customer's questions without you having to worry about it. On top of that it simplifies your communication management, because you now have all communication channels clearly and completely in one tool.
So you can deploy a chatbot very well as a source of answers within one of your service channels. But how does a chatbot know that a customer is asking a question and how is it possible that the chatbot can answer it? That works with artificial intelligence, also known by the English term ‘artificial intelligence’, abbreviated to AI. This is software that responds to data or impulses from their digital environment and on that basis makes decisions independently without human intervention. With this kind of software it is therefore not about computing power (processing many impulses at once), but about the ability to learn independently and make decisions, based on information learned in advance. For Watermelon these kinds of machines are the chatbots, which can be made by the customer with the help of Watermelon's software.
Natural Language Processing (NLP) is deployed to convert human language into something a computer can understand and use. You can add conversations or frequently asked questions to the chatbot so that it has a basis of questions and answers. As it holds more conversations and is trained it learns more and more, becomes more accurate and answers the questions better. If the chatbot does not understand something it registers this as a mismatch. It is then up to you to determine whether the mismatch should be added to the existing knowledge or whether you select another answer as the correct response.
Conversations are triggered by means of a question-and-answer model. All the questions a customer could possibly ask can be added to the chatbot's knowledge. This lets the chatbot learn to recognise the meaning of of different questions, even when the same answer applies to them. You add the answers as well.
An example: a customer asks which Dutch branches your company has, but also which branches of your company are located in the Netherlands. At least three variations on questions always have to be filled in. Watermelon recommends six, so that the smart chatbot can give an answer as accurately as possible.
When the smart chatbot gets a question, a classification of the question takes place first. For this, machine learning is used, in which with the help of statistics and algorithms the questions asked by the customer are converted – with the help of classification – into the language the chatbot can ‘understand’. The more questions are linked to an answer, the clearer the intent behind the question becomes for the chatbot.
Through the use of the question-and-answer model, Watermelon's software is set up in such a way that you, as a company and customer of Watermelon, never have to program when making the end user's (your customer's) questions and conversations with the chatbot.
By filtering out unnecessary words with the help of NLP the sentence is shortened to its essence. Artificially intelligent techniques have come a long way but are not yet perfect. So occasionally it happens that a question is not understood. We then speak of a “model of reality” with which the computer makes its predictions. This model is based on the answers the smart chatbot gives. For this you have two ways: retrieval-based and generative-based.
In the case of retrieval the chatbot can make use of a predefined set of answers it has learned. So the chatbot ‘retrieves’ the answers from the database and then sends them to the customer. With generative the chatbot can think up answers itself to the questions it gets. At Watermelon we enter the questions and answers and so create a database, so we use the retrieval-based model.
With Watermelon you always have the possibility of working with a human back-up when the chatbot has not yet learned a question or conversation. On top of that, Watermelon's software always has the option of letting a human employee take the conversation over seamlessly. Watermelon calls this the hybrid model. If the human employee is suddenly not present, this can be conveyed to the end user by means of an away message and it is then followed up later by the human employee.
A smart chatbot from Watermelon has the option of asking the customer for feedback after the conversation. This can be set up with the feedback module. The feedback is given by an emoticon, a thumbs-up or as text in a text field. By processing the feedback into the question-and-answer model, you also make the chatbot ever smarter at the same time.
Many companies are wary of deploying chatbots because they think a chatbot's possibilities are limited. But that is certainly not always the case! Watermelon makes chatbots ‘smarter’ with the help of artificial intelligence and handy software, combined with NLP. That way chatbots are trained to respond effectively to questions. And should the chatbot not work it out after all, there is always the possibility of bringing in a human employee.
Would you like to know more about how to deploy chatbots successfully? Download the free whitepaper by clicking the button below!
Share this article
