How to deploy chatbots successfully

Intelligent, self-learning chatbots are becoming an ever more popular means of handling customer questions. According to experts, chatbots are the future of customer service, with customer service staff really only being there for the more complex issues.

Nevertheless, companies are still wary. Our own research showed that two thirds of respondents consider trust that a chatbot gives the right answers the biggest obstacle to deploying such bots. It is essential to get off to a good start straight away. But how do you begin with chatbots in the right way, and how do you make sure you are actually successful with a chatbot?

1. Be patient

If you want to be successful in deploying chatbots, above all be very patient. In any case take plenty of time to make your chatbot smarter. By plenty of time I am not talking about a few weeks, but certainly two to three months! The very biggest pitfall with chatbots is that companies expect that as soon as they buy a chatbot, it will be complete in no time and know everything, and can therefore answer all questions adequately. That takes time. Use that time wisely, by filling your chatbot with information, by training it and clearing away mismatches.

2. Chat in a varied way

A chatbot learns by understanding several variations of the question. This way it learns to understand the intent of the question. So do not teach it keywords, but whole sentences and in different ways. Only then can it learn to understand the question. Make sure, therefore, that the different questions have the same intent.

Sometimes questions can differ, but the answer stays the same. In this case the idea is not that you merge the different questions – with different intents. Different question variations can perfectly well get the same answer. In this case you therefore add several questions with the same answer.

3. Go live!

Do not be afraid to put the chatbot live. Only when it is confronted with questions from real customers does the learning process start. This is the way the chatbot learns variations and gets smarter. Only for real do you notice which questions customers are going to ask, and the difficulty of the questions increases as well.

A chatbot only responds to a question when it understands it; if it does not, you can have it ask for the question to be asked again or have the question forwarded straight to a member of staff. The latter is called supervised learning. The chatbot learns from questions the customer service agent adds. That way there is also always oversight of what a chatbot does.

4. Manage expectations

If a customer calls on a chatbot but does not get an answer – or a clear one – there is a good chance he or she becomes frustrated. That is easy to head off by managing the customer's expectations in the chat, for example with a pre-set welcome message in which you indicate what the chatbot can help with. This is how we have done it ourselves:

Hi, I am Watermelon's chatbot. I am a beginner and still have a lot to learn; do I not have the answer to a question? Then probably tomorrow I will! What can I help you with? 😃

You can also give a list of the things you can help with, for example. That way the customer already knows that if he or she asks a question that is not in the list, the chatbot cannot give an answer.

Setting up an away message, for when the chatbot does not know the answer, is also advisable. Especially when the customer service staff are not continuously behind the dashboard, you can indicate that you will come back to it within a few minutes or ask for an email address to be left. It just depends on how the department is set up as to what works best for your company.

5. Replace FAQs that resemble each other with conversations

If you replace FAQs with conversations, you prevent the chatbot from giving an incorrect answer to questions that resemble each other. By setting up a conversation, you make a dialogue. It works like this: the customer asks a question, the chatbot recognises the question and so knows that it has to start a conversation and will ask return questions.

First of all the chatbot asks a return question. This can be what is known as a check question: have I understood correctly? Or a clarifying question: what exactly is it about? This way the chatbot guides the customer quickly and efficiently to the right answer.

An advantage of this is that you can steer the customer to the place on your website where you would most like to have them. If they have a question about a product, for example, and it has been answered by means of a number of return questions, you can ask the following return question: “Shall I take you to the product page?” Or, if it is a business product: “Shall I schedule a demo for you?”

Result

If you work through these steps properly, what does it ultimately deliver? Your chatbot gets steadily smarter. If you make sure you have managed customer expectations well and you have put the chatbot live, you will notice of your own accord that the learning goes faster and faster.

Ultimately your chatbot takes over frequently asked questions from the staff, who in turn can start occupying themselves with more complex questions. This means customers are helped better, faster and more enjoyably, both by the chatbot and by the customer service agent.

Nobody is made happy by answering the same simple questions over and over and the customer is not made happier by waiting a long time on the phone for a relatively simple answer.

If you set a chatbot up well and take the time to make it really successful, that ultimately offers more room for human interaction.

Would you like to know more about how to deploy chatbots successfully? Download the free whitepaper by clicking the button below!