Agentic AI löst Kundenanliegen für die AVR Kommunal

Hundreds of conversations — and 7 AI agents ready to help 

AVR receives a high volume of inquiries about the same topics. At the same time, citizens want immediate answers, rather than waiting for office hours.

AVR manages over 470,000 waste containers throughout the district. The need for communication is correspondingly high: AVR received 204,707 calls last year. A large portion of these consistently revolve around the same points:

• Collection dates
• Opening hours
• Fees
• Bulky waste collection
• Waste separation

Furthermore, expectations have changed; citizens prefer to find information themselves rather than wait for a service response. It was precisely for these standard questions that AVR saw the greatest potential for AI.

It didn't start with the technology

Before speaking with a single provider, the AVR team clarified what the assistant should actually be able to do. AVR considers this preparatory work one of the most important success factors of the project.

During an extensive exploration phase, the focus was on four questions:

1. Which questions do our customers ask most frequently?
2. Which processes are suitable for automation?
3. Where do potential risks lie?
4. What information does the assistant need, and in what format?

Only then did AVR compare various solutions. They were looking for a solution that was technically impressive and a good long-term fit for a public waste management authority. The choice fell on moinAI.

The moinAI agents in action at AVR

For AVR, the combination of high automation and strong content control was paramount. These were the deciding factors:

• GDPR-compliant and based in Germany and specialized in digital customer communication
• Controllable generative AIthat only accesses verified sources
• Easy maintenance of the knowledge base and comprehensive dashboard analytics
• Integration with existing systems such as SAP
• Predictable costs: no implementation fees, no per-query costs, just a fixed monthly license

How AnTonn works

For citizens, interacting with AnTonn feels like a normal conversation. In the background, the knowledge base, specialized AI agents, and the SAP integration work together seamlessly:

1. A person asks their question in the chat.
2. The AI analyzes the request and identifies the topic.
3. The responsible AI agent takes over.
4. The answer is generated from approved sources and is clearly marked as AI-generated.

GRAPHIC: the four steps as a chain, similar to the flowchart in the TEAG case study.

The knowledge base as a source of information 

AnTonn only provides answers based on sources specified by AVR itself. These include:

• Content from the AVR website
• the waste management statutes
• PDF documents
• internal knowledge databases

The team can add new content at any time, often in just a few minutes. This allows the knowledge base to grow with every new topic. At the same time, control over the content is never relinquished. 

Seven AI agents with clear areas of responsibility

Each AI agent is an expert in its own specific field. This makes the answers more targeted and precise.

SAP integration for minimal maintenance

AnTonn accesses the AVR SAP system via an interface. The waste disposal guide is managed centrally there and is automatically available in the chat as well. This eliminates the need for duplicate data entry.

Testing from the citizen's perspective

Before the go-live, AVR conducted intensive internal testing of AnTonn. The most important takeaway: employees often ask significantly more complex questions than citizens do.

The team had to learn to stop thinking from the perspective of the specialist departments and start thinking from the perspective of the people who actually use AnTonn. The technology must be solid first; only then do you move on to the name, appearance, and tone of voice. 

The result is AnTonn, an AI chatbot with a clear connection to waste management that is intentionally designed not to give the impression that a human is responding.

Go-live without a major campaign

For the launch on October 22, 2025, AVR deliberately opted against a large-scale communication campaign. Even the best AI makes mistakes at the beginning. During the first two months, the team analyzed inquiries and refined content and individual AI agents. Since March 2026, spot quality checks, new content, and a monthly evaluation have been sufficient.

Download the compact case study now to learn how AnTonn has evolved since the go-live. Includes charts on maintenance effort and the most frequent topics.

Learn, among other things:

• How quickly the maintenance effort decreased after the go-live
• The five topics citizens asked about most frequently in May 2026
• The three key takeaways AVR shares with other municipal service providers

‍

• How Dreaming evaluates every conversation and identifies which topics are still missing
• How AnTonn uses AI Actions to move from providing a reference to completing a task