Legal advice in milliseconds

Project: The National Federation of House Owners (Huseiernes Landsforbund)

Morten Fæste, Lawyer at Huseiernes Landsforbund, has been managing the chatbot project with great success. (Photo: Huseiernes Landsforbund)

Text by Eli Rugaard.

Lucy, Norway’s first chatbot giving legal advice on real estate matters, is turning 1 year this month, and we decided to have a chat with her main colleague, Morten Fæste. Morten works as a lawyer at The National Federation of House Owners (Huseiernes Landsforbund, HL) and has also managed the chatbot project since the beginning.

HL is a nationwide consumer organization that works actively to safeguard homeowners’ rights in relation to the authorities. Members of the federation receive legal, economic and technical building advice as well as valuable discounts when buying goods and services connected to housing. Some of this interaction happens through Lucy.

Lucy was named after Norway’s first female law professor and the first female rector at the University of Oslo, Lucy Smith. Lucy the chatbot gives legal advice to both members and other visitors of HL’s website round the clock. Lucy went live in June 2018 and has since the start answered legal questions, as well as questions regarding membership of the federation.

In November 2018, after the chatbot had been live for five months, Lucy the chatbot underwent a bot analysis. A bot analysis entails evaluating chatbot performance, UX dialog flow, looking in detail at how the chatbot has been used in the period it’s been live and maybe most importantly: up-to-date knowledge transfer.

An in-depth analysis of UX dialog flow helps understand how end users interact with the chatbot. In our analysis for HL, we saw that the most frequently asked questions were regarding membership, so we focused on building dialog flows that gave the user the opportunity to learn more about membership and signing up, as well as informing about the many pros of becoming a member.

The bot analysis showed that users often used single words to communicate with Lucy. Such insights are very valuable in the further development of the chatbots content, which is used as training data for the deep learning model. The tweaks that were done following the bot analysis resulted in a reduction in fallback of 20%. This was due to a deeper analysis of how the chatbot had been used in the months it had been live, and adapting the chatbot content thereafter.

Morten Fæste (Photo: Huseiernes Landsforbund)

An analysis of incoming messages showed that more than one-third of inquiries to the chatbot came outside of office hours. Lucy is on call 24/7, so she is a valuable supplement to the customer support that opens at 8 am and closes at 4 pm. Contacting Lucy can also be very time saving, you’ll get a response within milliseconds!

Why has this project been so successful? Morten points to some factors that have been key to Lucy’s success. Huseiernes Landsforbund is an organization that provides many benefits to its users. Using a chatbot has been successful because the bot can give free legal advice and make law accessible to all users of the chatbot. Receiving an answer in chatbot format means that it will be short and to the point to fit the chat bubble window, quite the opposite to reading law text or receiving a document from a lawyer. Lucy makes law accessible, users can ask Lucy for advice, and be given sound advice that’s been quality assured by lawyers and law students.

“It’s fun to take the lead and setting a good example by using new technology.” — Morten Fæste

Law is a field that’s mostly known for long traditions and attention to detail, but there is no reason this can’t be combined with testing new technology.

To wrap things up, Morten gives three tips for a successful chatbot project:

  • Use UX writers to help build dialogue structure and content. Not only to build a solid foundation of training data for the AI and the chatbot, but also for knowledge transfer between the two parties.
  • Continue working on the chatbot content after the initial project phase. Establish solid and systematic routines to continue improving the chatbot’s scope and performance.
  • Each bot has unique users, try to adapt the bot to its users. After the bot goes live, you will see how it’s used. Use this data to improve the bot accordingly.