Eneco's AI agent handles 70% more customer conversations without human hand-off

Overview

Eneco, a leading sustainable energy provider serving over 1.5 million customers in Belgium, replaced an underperforming chatbot with a new AI agent built on Microsoft Copilot Studio, combined with the Conversational Language Understanding model in Azure AI, developed in just three months -- half the time needed for the original chatbot. The new AI agent achieves an intent recognition score of over 95% and handles 67% of customer conversations without requiring hand-off to a live agent, compared to only 40% with the previous chatbot -- a 70% relative improvement. The agent now handles 24,000 chats per month, up from an average of 10,000 with the previous chatbot, with Azure AI Translator enabling cross-language handling between customers and agents.

The challenge

Eneco's existing chatbot was a black box that was difficult to train and refine, so it did not perform as accurately as hoped, and a large number of simple customer enquiries continued to be routed to increasingly strained call center channels.

The solution

Eneco replaced its chatbot with a new AI agent built on Microsoft Copilot Studio, combining its standard Natural Language Understanding model with the Conversational Language Understanding (CLU) model in Azure AI, connected to its live chat platform via technology partner Seamly, which also uses Azure AI Translator to translate conversations across languages.

Reported business value

The new AI agent was developed in just 3 months, half the time needed for the original chatbot, and achieves an intent recognition score of over 95%; it handles 67% of customer conversations without hand-off to a live agent (versus 40% previously), a 70% improvement, while volume rose to 24,000 chats per month from an average of 10,000 with the previous chatbot.