The Challenge
The potential for Generative AI to transform customer service is particularly significant in the financial services sector, where emerging FinTechs promise higher levels of customer service as a competitive advantage. However, it is also a sector where compliance with strict regulations remains critical. Any new solutions in the industry will need to deliver both.
Our client’s traditional chatbot with rigid dialog and keyword recognition delivered an unsatisfactory customer experience. Often, customers were unable to resolve their issues via chat or voicebot. In addition, the solution struggled to understand regional dialects. These issues resulted in a high number of customer requests requiring service agent intervention.
Despite its autonomy, the existing chatbot was limited by its inability to handle the complexity of human language without predefined dialog flows, requiring extensive manual maintenance. This inefficiency, combined with customer dissatisfaction with the bot’s user experience and design, underscored the need for a more sophisticated, AI-driven solution.
The approach
Given the speed at which GenAI is developing, there are already several different approaches being used by AI solution providers. At Eraneos, we decided to meet the customer’s requirements using a “Mixture-of-Experts” (MoE) technique. A machine learning architecture where a single neural network is organized into a stack of layers, and the overall problem is divided among smaller, specialized “experts”.
The challenge posed by the potential complexity of customer input could be solved by the divide-and-conquer model of the MoE approach. Rather than predefining the chatbot for the nearly infinite number of permutations that could arise during a conversation, our GenAI solution uses a gating function to ensure that customer input is sent to the expert with the highest probability of providing the right answer.
The goal was not simply to redefine dialog flows and layer GenAI on top to better understand customer intent. Instead, Eraneos wanted to completely rethink the current system and build an entirely new solution using pre-built AWS services.
In addition to reducing the number of agent-based interactions, another goal was to increase customer satisfaction. An advantage of using an MoE approach is that the solution is able to resolve a large number of customer intents by sending each query to the appropriate AI expert. Once implemented, the solution will significantly reduce latency and create more natural conversations.
If the GenAI voice and chatbot is unable to resolve an inquiry due to its complexity, Eraneos has established a methodology to connect the customer with a human agent. Whenever necessary, the customer is quickly transferred to the contact center, which includes various departments and a range of experts. Additionally, there are various escalation scenarios, with continuous emotion tracking used to gauge customer sentiment.
Our client is a major financial institution in Europe. They are committed to providing their customers with a wide range of innovative solutions in the areas of payments, investments, retirement planning and financing. With a history stretching back over a century, the client understands the pace and importance of embracing technological change. At the same time, it operates in a jurisdiction and industry with some of the most stringent regulatory requirements in the world. The company understands that while customers demand cutting-edge deployments, compromising on compliance is never an option.
The result
The company receives an average of 40,000 customer calls per week. Using GenAI, Eraneos’ voice and chatbot solution is expected to significantly reduce the number of agent-based interactions – ultimately to zero for many processes. In addition, in order to take advantage of the GenAI solution and realize its full potential, Eraneos assisted the customer in moving from their on-premises solution to the cloud.
To ensure the widest range of use cases, Eraneos’ GenAI chatbot can understand a variety of languages spoken by the client’s customers. To achieve this, Eraneos studied existing models and fine-tuned them to account for regional dialects. This is essential as existing speech-to-text tools do not work well with the common languages and dialects used by the client’s customers.
Our Mixture-of-Experts approach allows the solution to effectively handle 40,000 customer calls per week, ultimately reducing agent-based interactions to zero.
Knowing how important compliance is for this industry, we developed two different systems. The first is the bot itself. The second is a separate system dedicated to security. This ensures that the GenAI model works exactly the way the company wants it to. Because our approach is designed to allow a high degree of freedom within the conversational dialog flow, it needs rules to ensure that the dialog it contributes to is relevant and appropriate. For example, it is not allowed to advise the customer on banking products to ensure the security of sensitive financial information.
The final step in the process of delivering a GenAI solution that meets all of the customer’s objectives is to conduct extensive testing to ensure compliance. Following this, the client will not only be able to use GenAI to provide a higher level of customer service but will also be able to enjoy the benefits of the solution with peace of mind, knowing that this new technology will not create conflicts with regulators.
Let’s create sustainable change together.
Bodo Hechelmann
Head of Customer Service Transformation
Telco & Utilities | Financial Services
Generative AI | Organizational Excellence & Transformation | Data & AI
bodo.hechelmann@eraneos.com +41 58 123 93 89 LinkedIn