Automation of incoming customer correspondence in the insurance industry

Machine-learning-based automation of an insurer's inbox with customer complaint recognition

Like many businesses, health insurance companies face the challenge of dealing with an increasing number of customer interactions. In particular, it is becoming increasingly difficult for service staff to respond to all inquiries in a timely, consistent manner and with appropriate care. This is essential, however, especially in the case of customer complaints, to obviate unnecessary irritation of customers  and potentially losing them to the competition. In a project with a large German health insurance company, we investigated and demonstrated how modern text-based systems can be used in combination with artificial intelligence methods to support service staff in (partly) automating the processes of analysing customer complaints, classifying them by topic, and filling in the corresponding form fields in the existing CRM system.

After intensive analysis of the in-house mail routing process and discussions with stakeholders within the organisation, a centralised solution for all inbound channels was identified as suitable. Customer communications received via mail, e-mail, online customer portal or fax were analysed with the aid of machine learning and the identified customer complaints were assigned directly to the responsible staff. Both structured and unstructured elements were considered. The prototype we implemented recognised more than 95% of the complaints, whereas less than 3% of the regular communication was incorrectly classified as a complaint.

Cooperation partner: German insurance company

Project period: May 2018 - September 2018