Mishcon de Reya LLP is a law firm with offices in London and Singapore. Founded in 1937, it currently employs more than 900 people with over 500 lawyers. It is regarded as forming part of the “Silver Circle” of leading UK law firms.
Mishcon de Reya saw an opportunity to help internal users and visitors on their website to answer common questions quickly without the need for them to search the website. They approached ProLegalTech to build an automated tool that would help users to answer these queries by interacting with an artificially intelligent chatbot.
ProLegalTech built an intelligent chatbot solution for Mishcon de Reya that has three layers of intelligence. The first layer is the bot where we have created the logic that manages the user input.
The second layer is LUIS, which works out the intent and then communicates this with the bot to store the data as a conversation context which is then passed on to the next layer.
The third and final layer is the QnA maker, which attempts to match the input to a question in the knowledge base. There are two separate knowledge bases, one that handles default responses and the other that contains the main answers for the bots subject matter. If there is a conversation context, this is used to prioritise answers that match this context. If there is not a matching answer then a default is used for the current category.
The Mishcon de Reya chatbot also included advanced functionality to help the chatbot understand that questions which had preceded them would have influence over questions that came later in the conversation.
For example, when the user asks “what time?” in the below conversation, the chatbot is intelligent enough to understand that this refers to open days, based on the previous conversation.
The chatbot also includes follow-on question functionality which is shown as links under the latest answer. These can then be clicked which will automatically type in the question for the user and submit it which has a number of advantages:
- It can help the user understand how they can obtain further answers from the bot that relate to the answer just given.
- It can also be used to break up long answers by first providing a brief overview and then allowing the user to choose to find out more detailed information so that they aren’t overwhelmed with information.
- They are a good way to help a user find an answer when the bot has not been able to answer the question submitted.
Analytics and fine-tuning
Once the system was live, ongoing analytics were provided so that Mishcon de Reya could view any questions that didn’t match any answers so that they could be tweaked to ensure that any future questions asked, would be set up to provide the most relevant answer.
This enabled Mishcon to improve the chatbot by providing new answers or to add alternative phrased questions to ensure that the correct responses were provided.