How does… a chatbot work?

We ask some of the industry's most knowledgable boffins to break down jargon to help you through those confusing meetings and indecipherable conferences. Here, George P. Johnson's Chris Hogben explains how chatbots work.

Chances are you’ve been hearing more and more of a buzz around chatbots recently. They’re gaining popularity rapidly – and with good reason. Chatbots are virtual agents that provide customers with a natural way to interact with a system.

They enable customers to receive instant responses to questions and interact with company services such as making bookings or giving product information.

For businesses large and small, chatbots can provide services to their customers 24/7 and can offer basic customer support so that customer service staff are able to focus on more complex customer issues. In short, they can help reduce costs whilst simultaneously improving customer service.

So, how does it work?

Chatbots are another product of the AI revolution. Depending on the type of chatbot, their architecture varies and different companies offer slightly different levels of intelligence – IBM, for example, uses sentiment analysis to infer the customer’s tone.

Chatbots start by using input decoders to understand spoken or written customer dialog and use Natural Language Processing to analyse the sentence to determine user intent. With most goal-oriented chatbots, intents are predefined by the architect and will be fed many sample user input phrases so that machine learning can be used to match a user input with the correct intent.

From there, the chatbot will be able to give a response to the user and provide fulfilment. Fulfilment is where the back-end of the chatbot interacts with your server, or other platforms APIs, in order to satisfy the user’s request. For example, if a user asks to be added to a mailing list, the fulfilment would be adding a user to an email database; or if a user wanted to know if it will rain today, it may use a weather API to find local weather information that it can respond with.

Again, not all chatbots are the same, and some more advanced chatbots will actually synthesise their own responses rather than requiring hardcoded answers.

Usually you can discover the limits of chatbots pretty quickly and if they have been trained by a single developer it can unwittingly acquire the biases of their creator.

This is why rigorous testing is required to create a more robust, user-friendly chatbot. By having testers try to break it, we can learn of different ways customers might phrase their questions and add in questions we hadn’t expected them to ask. Training is a critical step in a successful chatbot.

What are chatbots used for?

Chatbots are great for the dissemination of basic information, for example an FAQ chatbot or a bot that can be trained in basic company policies so that staff can easily refer to it when needed. They are incredibly versatile and they’re popping up all over the place; in law firms, ticketing websites to purchase event tickets, provide real-time quotes and assist with banking.

If you have a website, there is a strong chance that it can be enhanced by adding a chatbot. After all, we build websites to provide customers with information, using design to ensure that the information is presented in the most accessible, user friendly way.

UX design plays a large role in ensuring customers can quickly find what they need. With a chatbot, website visitors don’t need to work to find the information, learn your website’s interface or scroll through pages of information, they can simply ask a chatbot directly and receive instant feedback. That’s a huge time saver.

The case for chatbots is made stronger when you consider that, unlike an app or website, they can meet people where they are, on social media platforms. Some of the most popular apps, such as Facebook Messenger, Slack and Telegram, allow businesses to create chatbots that customers can converse with, making it incredibly easy to access a business via a chatbot. From here, it’s all up to the level of training and intents the chatbot has been provided with to ensure it can be a useful tool rather than an unnecessary distraction.

As a side note, it’s important to consider your brand’s tone of voice when designing a chatbot’s responses, as it is representing your brand.

Some inspiration

Chatbots shouldn’t be thought of as a little chat window in the corner of a website. It’s a chance to rethink how we interact with customers online and design chatbots to take on a larger role. One beautiful example of how this might work has been explored in this K2 Bank concept.

Chris Hogben, designer, GPJ Australia.


Get the latest media and marketing industry news (and views) direct to your inbox.

Sign up to the free Mumbrella newsletter now.



Sign up to our free daily update to get the latest in media and marketing.