AI Transcription at Scale: A Proof of Concept for Gjensidige

Manual processes and the desire for efficiency

Gjensidige is one of Norway’s largest insurance companies, offering a wide range of services including health, property, and vehicle insurance. As part of their operations, they run a high-volume call center where support consultants handle countless customer inquiries daily.

A major challenge for these consultants is accurately capturing and documenting the key details from every call. Traditionally, this has been done manually during or after the conversation—an error-prone and time-consuming process. If consultants need to revisit a case, they often have to listen through the entire call recording, which can take up valuable time.

To streamline this process, Gjensidige wanted to explore how modern AI-powered transcription could help. While they were already using a transcription service, its quality was lacking. With the emergence of newer, more powerful AI models, Gjensidige approached us to create a proof of concept (POC) to evaluate whether upgrading would bring real value—both in quality and cost-efficiency.

Challenges between precision, speed, and technical integration

Gjensidige faced several challenges in optimizing their customer support operations. One major issue was accurately capturing and extracting important details from customer conversations. Speed was also critical—they needed transcriptions to be available almost immediately after a call ended to help consultants align their notes and streamline the onboarding process for new employees. Additionally, integrating external AI transcription services into their internal infrastructure posed a challenge, as all calls are routed through a centralized system that connects to various support agents.

Technical implementation: Precision through AI and intelligent mediation

To address these challenges, we implemented a solution that leveraged Cognigy as the intermediary between the phone system and the AI transcription service, effectively managing both call flow and data exchange. We focused on prompt engineering to guide the AI in transcribing conversations accurately and in identifying key information such as claim numbers, client requests, and required follow-up actions. Additionally, we tested and optimized several state-of-the-art AI models to ensure high transcription quality and minimal delay, delivering results almost immediately after each call.

More precise transcripts, faster processes, transparent scaling

The result was a robust and flexible transcription solution built on the latest AI models, with the capability to switch between them as needed. Compared to the previous system, the new solution consistently delivered more accurate transcripts, as demonstrated through comparative analysis. Transcriptions were available within one minute of calls lasting between 30 to 50 minutes, and the output included not only full conversations but also summarized notes that mirrored the manual notes typically taken by agents. In addition, we provided a cost projection for scaling the solution across Gjensidige’s operations and outlined any current limitations to support transparent planning and future improvements.