Introduction
In today's fast-paced public sector, organisations face an increasing challenge: they must quickly and effectively handle numerous citizen complaints.
This is especially true for complaints related to fines, which require prompt resolution, clear communication, and scalable processes to meet increasing demand.
In this context, Möbius was tasked with designing a solution to automate and optimise the complaint-handling process for a public organisation that processes tens of thousands of complaints annually. By implementing a cutting-edge Generative AI (GenAI) system, Möbius aimed to alleviate the manual workload while maintaining high standards of accuracy, transparency, and citizen satisfaction.
Strategic challenge
Möbius faced the challenge of automating the processing of tens of thousands of complaints related to various fines that this public organisation receives yearly. Currently, a team of 10 full-time employees (FTEs) manually investigates these complaints and responds to citizens via email — a time-consuming and repetitive process.
The variety and complexity of complaints often mean that employees lack the specific expertise needed for each case, leading to frequent reliance on their team lead for guidance. To streamline this process and improve efficiency, the organisation identified the potential for AI to automate part of the workflow by leveraging its extensive historical data on complaint classification and response generation.
Approach
Möbius developed an AI-powered system using GenAI to automate key aspects of complaint handling, including classification, decision-making on fine cancellation, and generating appropriate responses to citizens.
The approach consists of three key steps:
1. Complaint classification 📂
The system uses a Large Language Model (LLM) to categorise complaints based on historical patterns. If the system's confidence in its classification is below a set threshold (e.g., 95%), the case is flagged for human review. This ensures uncertain classifications are validated or corrected by employees, with feedback continuously improving the system’s accuracy.
2. Complaint verification ✅
The system then performs an automated verification process, cross-checking external databases to determine if the complaint is valid and whether the fine should be reimbursed.
3. Response generation ⌨️
After verification, the system generates a response via the LLM, explaining the decision to the citizen. The response is reviewed by a human for accuracy before being sent, and the system’s multilingual capabilities allow for communication in the citizen's preferred language.
Human oversight is maintained at key points, ensuring control and quality while allowing the AI to handle the repetitive aspects of complaint processing. Möbius built this system within a secure, GDPR-compliant environment using Microsoft Azure, with Azure Functions managing the workflow and Azure OpenAI driving the LLM-based automation.
Expected results
The introduction of AI is expected to significantly reduce the time and manual effort involved in complaint handling, allowing the team to focus on more complex tasks and decreasing their reliance on the team lead. The combination of AI and human validation ensures that even low-confidence classifications are handled correctly, with continuous learning improving the system over time.
The automated verification process enhances accuracy and fairness, ensuring valid complaints are reimbursed. After testing and optimisation, the system will be fully integrated into the organisation’s platform, improving the speed and quality of citizen communication while streamlining overall operations.
