EnBW: Redesigning customer service processes with GenAI

EnBW is one of the Big Four energy providers in Germany. Their new customer care agents undergo an onboarding process that can take up to two months. During this time, they are trained to handle a wide range of customer issues, from explaining billing details and exploring new contract options to clarifying tariffs, assisting with electricity meter readings, and helping with address changes.

Customer care agents aim to resolve client calls within a specified amount of time, handling tasks such as identifying the customer, understanding the issue, finding a solution, taking action, and summarizing and documenting the call within this tight time frame.

The primary pain point for agents under this strict time constraint is finding the correct information fast and efficiently when on the job.

Studies suggested that the greatest business potential for GenAI in the energy industry lies in the areas of sales and customer support. After initial testing and prototyping, we arrived at the same conclusion. Our joint idea was to build an AI-driven chatbot to pull information from the Knowledge Management CMS into a format where agents could quickly receive direct answers to their customers' questions.

The information in the Knowledge Management system had typical issues of an intranet, some outdated content, contradicting articles, and structure that was at times challenging to navigate. This made obtaining clear answers to the questions at hand challenging.

We started by pulling the existing data and creating a Retrieval-Augmented Generation (RAG) chatbot. This combined the sources in a vector database that fed GPT-3.5 to provide customer care agents with answers based on EnBW's knowledge.

We used the Ragas evaluation framework to systematically improve answer quality, experimenting with various optimization approaches in chunking, hybrid search, summarization, document metadata, and search relevance optimization.

After evaluating the RAG system, we identified the ‘retrieval’ component (i.e. the underlying search) as the biggest bottleneck in response quality. We decided to focus our efforts on 'better data' and 'better search’ as the first step towards developing a high-quality RAG application. This meant creating an AI-enabled search tool that used GenAI to provide snippets and summaries for each search result before going into a full-blown chatbot.

With this approach, EnBW customer care agents quickly expressed excitement about the new system. Trainers saw significant potential in saving time during training and boosting agents' confidence in their work.

EnBW is meticulous in its data expertise, relying on numbers and statistics when evaluating PoCs. Together, we elevated the company's GenAI capabilities to new levels. The project emphasized the learning aspect of bringing AI to a large organization. We continuously formed hypotheses, tested them, gathered insights, and then adapted. It was a highly collaborative journey.

Several use cases were explored, tested, and validated within 7 months with large data sets and different users. Most cases focused on increasing efficiency and improving customer experience (CX). The knowledge about GenAI was systematically built up within the organization, and broad awareness of the topic was created through numerous tests and demos.

Through a collaborative approach, we helped EnBW transition from understanding the potential of GenAI to having a clear picture of the highest GenAI opportunities across their customer service processes. We provided a plan for turning ideas into action and developed greater internal capabilities to make it happen. The project boosted the company's GenAI maturity and delivered real-life improvements in efficiency, customer satisfaction, and cost savings.