Chatbots at Nestle: Improving Performance on Intent Detection

Nestlé built its first chatbot in 2016 with the rise of Artificial Intelligence and particularly Machine Learning and Natural Language Processing.

Brands are currently using chatbots to communicate with consumers on their preferred messaging apps, such as Facebook Messenger or WhatsApp. Instead of having an army of people, chatbots can hold personalized conversations with multiple consumers 24/7.

Since 2016, Nestlé has built and deployed up to 20 bots in production, with more than 15 projects in progress. Use cases for Nestlé can go from pre-purchase advice, recipe recommendation, and pregnancy advice, to machine troubleshooting and post-purchase support.

While the space of chatbots providers is rapidly changing and evolving, most of these platforms and technologies are focused on building conversations with a basic natural language processing, but don’t focus on the post-launch activities such as intent classification or combining several metrics sources to have an overview of the bot performance and user engagement.

As subject matter experts, we are focusing on not only selecting the best Natural Language Processing engine but also the conversation design that allows bots to chat as if we were chatting with friends. With these two components, we can identify if there is a good intent classification and enhance the bot over time.