Situation
One of the Netherlands’ largest health insurance providers faced a fundamental customer service challenge. While the company’s website already contained clear and accurate answers to the majority of customer queries, users struggled to find the information they needed. This resulted in a high volume of avoidable inbound calls, placing unnecessary pressure on the contact centre and diverting agents from situations that genuinely required human support.
Approach
Valcon built a production-grade AI assistant, the Conversational Engine, directly into the insurer’s public website, fully powered by Databricks. A phased rollout approach was adopted, initially covering five high-frequency customer questions before expanding to hundreds of topics.
The solution uses Retrieval-Augmented Generation (RAG), with website content governed in Unity Catalog, semantic search powered by Mosaic AI Vector Search, and responses served through Databricks Model Serving. Scheduled Databricks Jobs automate content refreshes, while MLflow tracing captures user feedback to support continuous improvement and optimisation.
Results
The AI assistant now handles approximately 25,000 customer interactions per month, scaling to 40,000 during December, the peak renewal period in the Dutch insurance market, without any degradation in performance.
In a structured evaluation, 92% of responses were rated acceptable, and 97% were found to be substantively correct. Delivered in approximately six months through part-time development effort, the project also demonstrated the speed and efficiency benefits of Databricks as an integrated AI and data platform.
By enabling customers to self-serve on general queries, the solution significantly reduces avoidable inbound calls and allows contact centre agents to focus on interactions that genuinely require human judgement, empathy and support.
