Showcase – Voicebot

Challenge: The Road Traffic Office of the Canton of Aargau receives over 330,000 telephone enquiries per year, which it is unable to handle satisfactorily for capacity reasons. At peak times, less than 50% of the requests can be treated in a timely fashion. Outside office hours, an answer machine provides limited information. Because of the poor accessibility, customer satisfaction is low, which results in pressure on office staff.

Solution: Combine speech recognition and bot technology to implement a voicebot that enables automated answering of simple requests.

Hypothesis: We believe that it is possible to build – with relatively little effort ­– a voicebot that can answer the most frequent questions independently. We also believe that this will increase customer satisfaction and relieve employees so that they can perform their actual work more effectively.

Approach: To find out whether our hypothesis is correct, we ran a proof of concept (PoC) based on a cloud-based voicebot infrastructure put together from different commercial and open-source products. We selected five frequently asked questions and built language samples in Swiss-German to train a natural language understanding model (NLU). In iterative tests with employees and customers, we continuously optimized the solution to improve both understanding of questions and customer experience along the call flow. Finally, we switched on the solution selectively outside of office hours.

Result: Our PoC solution successfully handled simple customer requests - from welcoming customers to answering their questions. Despite the small amount of data and the short training time, the recognition rate was surprisingly high (over 70%). Contrary to our expectations, acceptance of the solution was also high: the majority of users were open to and even enthusiastic about using this new possibility to get information.