Faster correspondance, 24/7 availability and stress reduction for the customer care center: Various departments of R+V insurance have been dreaming of a chatbot implementation for a long time. Therefore, a decision was made in 2018: The topic “Chatbot” was placed on the official agenda and started as a trainee-project.
About R+V Insurance
The R+V insurance is one of the biggest German insurance companies for private and corporate customers and belongs to the cooperative banking group Volksbanken Raiffeisenbanken. Since the founding of Insurlab, R+V is an active member and shaper of the community.
The startup Kauz GmbH from Düsseldorf develops high-quality chatbots with natural language understanding (NLU). In addition to the complete NLU-chatbot, the company also offers Chat & Search, a chatbot-based search module which can be used for internal employee-service.
Each year, R+V Insurance provides recent university graduates with the possibility to start their professional career as part of a trainee-program. In this program, they have to complete an interdisciplinary project for a given topic.
Within the following six months, the trainees had to complete three tasks: Creating a classification of chatbots, collecting data about the expectations within the different departments and – the final test – creating a prototype.
Step one, writing down the following classification, was completed quickly: The main expectation of users is to receive a correct answer to their question. Of course, many users like to engage in small talk as well, but users do not want to engage in an actual conversation with a chatbot.
Step two: Collecting data about the different companywide expectations. The trainees conducted dozens of interviews and created a longlist with ideas for the use of a chatbot. The scope of potential applications reached from customer care via marketing scenarios to the internal services for employees. The longlist was then reduced to a shortlist and in the end, the internal technical support of R+V won the round. This was the chosen application for the prototype.
One big reason for choosing this usecase was the fact that the already existing content for self-help was not met with much enthusiasm.
Last task: Quickly developing a functional chatbot, as the given time of the project was ending.
At that moment, Kauz was chosen. A “No-Code-Solution” that could be implemented directly by the interdisciplinary trainee team without much help from the IT-department. Based on the realization that users want their answers fast and correct, the team focussed on the chatbot-component “Chat&Search”
Already six days later, the textbased chatbot came to life and was named ”Lara“. The first testrun presented sensational results, even without any adjustments. The rate of corrects answers scored 77 per cent compared to the search function of the intranet which only scored 8 per cent.
Mission accomplished – what now?
The internal technical support department took on the prototype as a “digital apprentice” and even assigned the chatbot to a trainer. Gradually, Lara was provided with all the available information. Technical questions could be supported with self-help texts and other topics were registered and sent to the correct contact.
The trainees already planned to extend Laras skills. In the future, she (the chatbot) should be able to support the R+V-employees in many more tasks. Therefore, Lara also received other information next to the technical content. She even knows company-specific abbreviations.
Already in 2020, the original technical support content made up only half of the chatbot’s knowledge. Lara made much progress towards becoming an assistant for all employees.
Only 8.4 per cent of the overall questions are answered incorrectly.
The expansion continues in 2021. Some new topics were already integrated in the beginning of the year and more are waiting in the pipeline.
Conclusion: Kauz is a “No-Code”-solution for the fast and easy use throughout departmens. It is simple in maintenance and can be continuously refined. As it is a curated chatbot, the developers do not need to worry about language understanding or small-talk abilities.