Context and objectives

Our client, an entity managing public waste collection, faced a recurring problem: handling hundreds of citizen complaints each month. Many were irrelevant or duplicate requests, consuming valuable time and creating inefficiencies.

Overwhelmed by the manual handling of these requests, staff members spent nearly one-third of their working hours on complaint management. The emotional toll was quite concerning, with some complaints being aggressive or offensive, adding stress to the team.

Our goal was to test whether an AI-based solution could boost efficiency while integrating seamlessly with the client’s infrastructure. Automation needed to:

• reduce workload,

• speed up responses,

• and free employees for more critical tasks.

Approach

Our strategy centered around designing an AI assistant to automate most complaint interactions.

This complaint management project as a whole has been managed in two phases. Agilytic first worked on a proof-of-concept to validate the feasibility of replicating the complaints handling process with AI. After successfully completing it, we developed a Minimum Viable Product with a more robust and fully integrated solution.  

1. Data gathering and cross-referencing

We first needed to ensure that the AI assistant could make informed decisions about the validity of each complaint in order to propose a response for the citizen. The system integrated multiple data sources, cross-referencing them with the complaint to confirm accuracy:

• real-time garbage truck location,

• pre-planned collection routes,

• collection schedules.

Using data already available at our client’s, we replicated the existing complaint management logic into an algorithm that would power the assistant. Built on Azure Virtual Power Agents, the assistant automatically determines their validity and proposes a response that can be validated by the agent and then sent to the citizen.

2. Implementation and testing

A critical part of the process was enabling the AI assistant to handle unstructured data, especially emails. It needed to extract structured information from the free-text complaints in those emails. The chatbot’s success relied on its ability to process and convert these inputs into actionable data.

Everything was hosted within the client's pre-existing Azure environment to preserve a seamless integration with their existing IT infrastructure and service desk system.

During the testing phase, the team applied weekly patches, which refined the solution's ability to handle edge cases and unexpected complaints. The development team minimized the volume of complaints requiring manual intervention by improving the system iteratively.

Results

Over 1.5 months of testing, the AI assistant autonomously handled 83% of complaints. As a domino effect, the automation lightened the administrative workload, freeing staff to focus on more pressing tasks.

This success opens up new possibilities, like building an omni-channel complaint management system. This would unify customer interactions from email, social media, and phone, making it easier for citizens to get quick responses, regardless of how they reach out.

Still, this approach is not limited to waste management: industries like transport and logistics can also benefit from automating complaint management and customer service as a whole, with AI assistants adapted to various business needs. These systems both improve response times and enable teams to focus on more complex challenges.