Inside a Chatbot

Posted on: January 31, 2018 by Guillaume Lefebvre

It is now more than 10 years since at Worldline we first began working on solutions that would later become known as chatbots. Back then, we were approached by a number of brands to help them provide faster responses to their customers’ queries on the Internet or by email. They knew that 48 hours was not good enough for a response time, so we worked on a new technological solution to serve their customers better. At that stage it was not even called a chatbot.


Fast forward a decade, and the technology has changed beyond all recognition, incorporating artificial intelligence (AI) and machine learning.

However, the underlying intention is the same. Brands want to understand their customers better and develop better, increasingly real-time relationships with their end-users. A chatbot can be the ideal mechanism for achieving that. It is an instant relationship builder.

In the last couple of years, we have seen a considerable increase in the number of brands who are using chatbots to interact with their end users, often using open application programming interface (API) on social messaging platforms or adding new skills to voicebots such as Amazon Echo Alexa and Google Home. There have been huge advances in voice recognition. Customers can now order food just by speaking to the latest generation of smart speakers such as Amazon Echo Alexa and Google Home. More and more applications are becoming available on these devices, taking chatbots and voicebots into the mainstream. The technologies are more mature and the customers are ready.

Supervised learning

Whereas the previous generation of chatbots was based on rules, today’s chatbots use machine learning and artificial intelligence to establish simple, natural conversations with end users.

However, while the application of AI may produce a great chatbot, at this stage in its development, it is not sufficient.

For example, it is not enough for a chatbot to understand a user’s questions. It must also engage in a conversation with users. This is where contextual analysis and other domains come in. To improve the conversation, and guide end users to their final destinations, chatbots need external information. They need to understand where end users are coming from. They need to appreciate the context for their questions and adapt accordingly.

That is the job of machine learning. But today this still needs to be supervised by humans. When chatbots log questions from users that they do not understand, we still need a human being to validate them and decide whether they need to be integrated. The machine learning process also needs supervision to ensure that any understanding and conversations are genuine before a chatbot learns from them.

The risks of non-supervised machine learning are all too clearly illustrated by the problems of Tay in 2016.

Context and content are king

A chatbot’s main advantage over websites is its ability to act on contextual information. On websites it can be hard to find opening times for hospitals or work out when a social security office is open. Asking a chatbot a question about times for a specific date can result in much more specific information.

A chatbot’s ability to make provide contextual solutions to end users can help organizations deepen their relationship with their customers. For example, at Worldline we are working with a transportation company on projects that would enable travelers to plan their entire journeys using only chatbot technology on Facebook Messenger and Amazon Echo Alexa.

A chatbot cannot answer everything that a customer can ask it, so it is better to make it specific and useful for end-users, avoiding the risk of a bad user experience.

For all organizations who are interested in developing and launching chatbots, I would offer one key piece of advice: Keep It Simple!


Chatbots and AI: 2018 theme of our international student competition “Atos IT Challenge”

Artificial Intelligence is a game changing technology but some are also saying it could be a huge threat – with calls for regulation to protect humanity against AI running out of control.

We believe it is an important area for students to explore as one thing is certain: if you master the chatbot, you master the world!

More about the Atos IT Challenge competition

Share this blog article

  • Share on Linked In

About Guillaume Lefebvre
R&D Manager, Worldline
Guillaume Lefebvre is currently Head of R&D User eXperience Department in Worldline. He has 17+ years of experience in innovation and Research & Development. An expert in user interface technologies, he investigates all kinds of new customers experiences to develop new online services for Worldline customers. Guillaume has worked on innovative concepts around new web technologies, Social Networks, chat bots & conversational agents, connected TV, Cross Channel & mobile uses, and new digital journeys thanks to mobile, tablet, camera, wearable device, digital signage, physical web, and Internet of Things technologies.

Follow or contact Guillaume