Deloitte chatbot

Case study

Chatbot as a search tool for an online technical library

Deloitte’s online technical library is so extensive that you could easily get lost in it. That is why employees will soon see the introduction of a smart chatbot that will guide them through the jungle of laws, regulations, auditing and accounting standards, and specialist literature.

Question-and-answer game

“Although the technical library has a search engine, you have to enter very specific search criteria to use it,” explains Twan van Gool, Director Innovation & Analytics within the Audit department at Deloitte. “People use the search function in the same way as they use Google. They type in ‘materiality’ when searching for how to determine the materiality of a public interest entity. But the library contains a vast array of specialist information, and the contents are closely related. You won’t find the answer by using a general search criterion; it generates far too many search results.”

In order to help employees to search more efficiently, the analytics and innovation team is working on a smart chatbot. “The chatbot takes people to the search result more quickly by playing a question-and-answer game,” says Van Gool. “It acts like an intermediary between the user and the library.” For example, the chatbot asks whether the user is looking for an accounting or an audit issue, and then poses follow-up questions.

“The smart algorithm uses those questions to guide you to the appropriate information,” continues Van Gool. Users chat with the bot in a separate pop-up window, “just as if you were typing a message in Skype. Because it feels as if you are chatting with a real person, you use natural sentences, and you find it easier to give details about exactly what you are looking for.”

AI email alert

Get 16 cases in your inbox 

Sign up here

Japanese model

This is not the first time the team has attempted to make the technical library more user-friendly. “We spent a long time tagging content and enhancing the search function, but that achieved very little,” admits Van Gool. “That is why we are now devoting our energy to improving the input side of it. The chatbot is relatively easy to create.”

The chatbot is currently being built for one collection in the technical library, and the plan is to roll out the chatbot for all forty collections in due course. The first MVP (Minimal Viable Product) is now in use to gain experience. The chatbot will also be integrated in Deloitte's Global Technical Library. Van Gool’s team is using the chatbot built for Deloitte Japan’s technical library as the model for development, and the results of the Japanese pilot are encouraging. “The technical library has been consulted far more frequently since the introduction of the chatbot,” notes van Gool. “Users have also started performing slightly more detailed searches, and entering far more advanced search criteria.”

Valuable data

The chatbot not only simplifies the search process, but also produces valuable data. “While the present search engine shows us what is being searched for and what is being found, it doesn’t tell us whether the user considered the answer satisfactory,” says Van Gool.

The chatbot shows what exactly is being searched for, and the route by which the user arrives at the result. “We can use that information to feed back into the algorithm in order to make the chatbot faster and more efficient.” The data are also of value outside of the technical library. “If we identify trends, developments and current issues, we are able to anticipate them immediately,” adds Van Gool.

Vond u dit nuttig?