Chatbot is a term that has popped up more and more often over the last few years. Many companies have on their agenda to implement chatbots in an array of different cases. Customer service, HR, IT support, E commerce -- are all great uses for a chatbot. But what does it really mean when we say chatbot? What is the technology behind it actually doing? And why have chatbots had a bad reputation for such a long time? These are the questions we'll be trying to answer in this article.

History of chatbots

Chatbots have actually been around for quite a while. The earliest computer games where a form of chatbot. Choose your own adventure style games where you typed your action into the computer and then got a set of responses based on what you typed. This is the essence of a chatbot, and provided the limited technology at the time, resulted in a suprisingly immersive gaming experience.

In the 90's the bot Watson became famous for winning a jeopardy game against two humans. Watson is still available today as a chatbot for businesses.

As an application for businesses to help customers and employees in various ways, chatbots have been around for at least 10 years. The first chatbots that started popping up in the bottom right corner on website where very rudimentary and more often than not, was not able to help the visitor. This lead to an increasing frustration in general over chatbot as a channel for support.

But as technology matured, chatbots have become better and better. And in since 2020 many businesses have been able to implement them with great success.

Chatbot technology

The first chatbots used by businesses for support purposes were triggered by certain keywords typed by the users. If someone typed a sentence with the word "invoice" for instance, the chatbot was programmed to give a specific reply. As you can imagine, this resulted in many misunderstandings and lead to a lot of frustration. That is because a sentence can have millions of different meanings and still contain the word "invoice". "Where is my invoice?" requires a completely different reply than "My invoice is way too high!". You get the picture.

What made the so called chatbot revolution start was a Machine learning technology known as NLP (Natural Language Processing). This technology made it possible for a chatbot to interpret complete sentences instead of only specific keywords. The alogrithm collects similar sentences and uses the sentences as training data to understand all the remaining ways that sentence can possibly be expressed in a specific languages. This leads to an extremely high level of understanding. Many chatbots today can understand up to 90% av what a customer is saying.

A competent chatbot platform can also easily collect new data from the incoming conversations. If a previously unfamiliar sentence is stumbled upon, the chatbot can be trained to recognise the sentence going forward. This is what is known as training the chatbot.

On top of the NLP technology that makes chatbots understand more than ever before, some platforms lets businesses build contextual understanding on top of this. This means that chatbots can understand follow up questions as well as handle multiple topics without getting confused.

When a chatbot needs to be able to handle large amounts of data, for instance a whole product library, other techniques can be use to make it easier to separate all the different parameters involved. At Ebbot, we use entities to map certain properties with a specific product. This way a chatbot can answer questions about all properties on a specific product. This can also be done by integrating to a PIM system, more on that later.

Building a chatbot

The most user friendly way to build a chatbot is to structure the questions and answers in a tree format. That way it's easier to map follow up questions to let the bot relate to a whole topic.

When you train a question with an NLP algorithm you start out by giving the bot a bunch of training phrases. A training phrase is a variation of a sentence that the chatbot should base their language comprehension for that specific question on. For instance you write 5-15 variations of the sentence "where is my invoice". The bot will then fill in a multitude of different ways that question can be expressed in (usually over 95%). The more so called training phrases that has been trained, the higher the understanding.

When the questions are set up, the next step is to setup a welcome message. This is the message that comes up when a visitor opens upp the chat widget for the first time. The welcome message is crucial for creating a first impression as well as making sure the visitor interacts with the chatbot. A welcome message should be short and portray the brands tonality as well as proactively leading the visitor to the next step. This can be achieved by letting them know what the chatbot can do; sometimes this is even shown with additional buttons that will activate certain answers.

The last step is to set up a so called Catchall. This is the message that comes up when a situation arises where the chatbot does not understand. What is important here is that there is always a next step for the visitor. Either suggest answers to questions with buttons, ask a follow up question, or offer to pass the chat on to a human agent. This keeps the visitors from getting frustrated.

When the bot is set up it is ready to launch. From here on the work starts with maintaining the chatbot and making sure it gets better over time. It is important here that the chat platform offers a training center, with a user friendly interface to make the training over time easy.

When something is written that the chatbot does not understand, this phrase will be sent to the training center so that it can be matched with the proper phrase that is already trained in the chatbot. Alternatively, a new set of questions and answers can be created if does not already exist.

Integrations

To get the most out of a chatbot, the chatbot should be integrated to other systems that can support the actions the chatbot is trying to complete for the customers. By integrating to a Ticketing system for instance, a chatbot can create tickets as well as retrieve information from open tickets. If the chatbot is integrated to a payment platform. the chatbot can suddenly complete orders without the customer having to leave the chat.

With open APIs, integrations are easier than ever to set up, and businesses should not be afraid to embark upon a chatbot project if integrations are required to make it work. Rather, they should embrace the possibilites and the increased results that can be achieved then the integrations are in place.

Other integrations include:

CRM: Update company, and contact records as well as retrieving data from the CRM. Collect leads and update customer records.

  • Logistics system: Retrieve status of shipping orders, and log cancellations and returns.
  • Identification system: Identify customers and log them into my pages.
  • Other messengers: Integrate a chatbot to an otherwise regular livechat tool.
  • HR systems: Call in sick, check you shift or plan your holiday without leaving the chat.
  • PIM (Product information management) systems: Answer questions about all products properties.

What is the best experience?

Technology is simply not enough to create a great customer experience with a chatbot. It is important to keep in mind what the actual customer journey looks like when setting up all the questions and answers the chatbot should be able to handle.

Sometimes a great experience is not equal to a long text answer, no matter how accurately interpreted. Sometimes, a simple button could give the best experience. At Ebbot, we work a lot with interactive chat-cards that can create a more interactive and visual expreience with the chatbot.

Another thing to keep in mind is when to pass on to a human agent. As the enquiries gets more and more complex, it's sometimes way better to let a human handle the case. Basically, a chatbot does not mean that human agents become obsolete. Quite the opposite, a chatbot is rather making the human agent more important (and effective than ever) -- handling more customers, quicker.

In the end, it's all about streamlining the support experience. A human agent will at most time be able to deliver a better more personal customer experience. But at what cost? More waiting time? Longer handling time? By letting a human and chatbot work together businesses can make sure that the customer always get help instantly, and that no step in the customer journey is delayed.

Benefits of using a chatbot

When a chatbot is implemented correctly, it can revolutionise how a business handle customers. The instant availability of a chatbot is changing expectations for wait times in support. The quick handling times can resolve more cases quicker. And the fluidity of language can make one bot help customers in many different countries. It also tends to be cheaper in the long run to have a chatbot handle the majority of customer enquiries.

There are a few things to keep in mind, like knowing when to pass on to a human or what integrations to use. The key is to have a great partner when setting it up, and to start with small goals, and work your way up.

At Ebbot, we have a lot of experience setting up chatbots for some of the largest E commerce and Energy companies in Sweden. Feel free to reach out for a free demo or consultation here.