Anyone who has ever had a technical issue affect their ability to access the Internet wants their support request to be received and addressed quickly. Service providers recognize this imperative, but clunky legacy systems can leave technicians with time-consuming manual work and customers waiting for extended periods.
That’s why R-KOM, an industry leader in telecommunications innovation, turned to Camunda to help implement a solution that empowered technicians to call upon intelligent assistants to work on specified cases, freeing up human resources for more complicated cases and driving better experiences for both technicians and customers.
R-KOM wanted to address a particular challenge in its support technician deployment – customer requests that arrive on weekends, outside of normal business hours. Before, a customer who called outside of hours and needed to wait for a technician to be manually assigned to their case and begin diagnosing the problem could wait an entire weekend without internet access.
Aside from the inconvenience a customer experiences with that level of service interruption and the potential to make the customer change providers, there are significant regulatory consequences that come with a delayed response. According to German regulations, internet service providers must offer compensation for internet outages lasting more than two days and must provide detailed documentation of service interruptions to consumer protection agencies. These providers can also face penalties for prolonged outages. One ill-timed outage could have serious financial consequences for R-KOM, who were operating with an average 8-hour response time.
The old process was completely manual; a customer phone call or email would be logged into the customer support system to create a ticket, which would then be manually selected by a human technician. The technician would then need to begin the diagnostic process to address and potentially reset the technical components causing the customer’s service disruption. Overall, the process was clunky and labor-intensive for technicians, with little visibility or tracking in the process for improvements.
That’s where Florian Hellmeier, Software Architect, and Thomas Kiefer, Automation and Process Manager, came in. The engineering team at R-KOM is small, so they are always looking for strategic opportunities to introduce automation that will help the team work as efficiently as possible. A highly manual and mission-critical process like customer support and communication felt like a great opportunity to make a meaningful impact, but there were some structural and technical hurdles to clear when building a solution.
The technician team managed their own diagnostic tools, so the engineering team had no insight or access into them. To build a solution that could automate any portion of this process would require buy-in from the technicians on the business side, working alongside IT to design and deploy a solution that supported everyone’s goals.
Additionally, the actual end-to-end business process that supports customer support ticketing and technician assignment was complex. The dynamic element of requiring customer interaction to move the process along introduced complexity that meant a fully deterministic-style approach to orchestration and automation wouldn’t really succeed. To make this transformation matter, R-KOM needed to find a way to introduce and orchestrate AI agents in the process while preserving the guardrails the company wanted.
“We thought that AI would be a perfect fit for us to solve this,” explained Hellmeier. He was reminded of a project a colleague, software development and AI specialist Peter Ruider, had been piloting. Peter had been using AI to generate and automate replies to customer communication requests. “It was interesting, but at the time we didn’t see many benefits,” Florian explained. But this pilot helped Florian and Thomas think about incorporating AI agents into the actual ticketing system and not just communications – and using a strong orchestration platform to incorporate those agents into the actual business process.
The project goal was two-fold:
R-KOM was already using Camunda for their deterministic process orchestration needs, and Camunda’s agentic orchestration capabilities meant that the team could address the dynamic element of customer communication and interaction without sacrificing the visibility and flexibility they needed in their platform. R-KOM had the same concerns many enterprises have when considering incorporating agents into customer-facing or other mission-critical processes; namely, concern over visibility into agents’ performance within the process, and the ability to direct their output. That’s why agentic orchestration was the right fit for R-KOM.
In particular, Camunda’s use of BPMN for processes and agents allows for transparency in both the overall process and in agent utilization within it. And since Camunda’s agentic orchestration engine allows agents to invoke tools and sub-processes as needed, R-KOM can reuse process assets across use cases to speed up the process and reduce engineering effort and duplication.
This reusability makes it easier and faster for the engineering team to deploy solutions, and it speeds up delivery to the customer. For example, technicians often need to provision a customer’s technical port during a service call to restore service; the provisioning process has already been modeled in Camunda, and now the agent can call that existing process and automatically reprovision after the diagnosis stage. With this functionality, the agent has the same capabilities as a human technician and acts as a new colleague that saves the complex work for the human technician.
Florian, Thomas, and Peter went to work to support the technicians with a Camunda-powered agentic ticketing system. R-KOM has implemented a core API to keep track of customer data, and Camunda sits on top to access the API to read and work with that data. Support services (diagnosis, communications, and other functionality) sit next to the API as well, where Camunda can call any service that is needed. R-KOM selected OpenAI as the LLM on which to base their service agents. Service technicians use Camunda Optimize to review the process, while Florian’s team manages the operational needs of maintaining and updating the process.
The whole process, including agentic support, is modeled in BPMN. One of the team’s favorite benefits of working with Camunda has been the ease of use of BPMN. Peter had never worked in the language before this project, but found it easy to understand. He was able to fully model the process and the agent prompts in under a week, greatly speeding up the entire project.
With the agentic process in production, service technicians can easily see and choose which customer requests should be diverted to AI agents or reserved for immediate human handling. This human-in-the-loop guardrail is included by design to help technicians better trust the AI agents, who were initially hesitant to hand over critical work to agents.
However, that reluctance has given way to excitement in the technicians and other business users, who are now seeing the effect the system has on their response times and efficiency. Florian credits the strong IT-business collaboration in the process, supported by BPMN, as a common language that gave stakeholders a seat at the table in designing a process that works for them.
The overall process works quickly, and the agent response time is very fast. This element has changed the nature of work for the technicians; they are spending less time and writing fewer responses to customers; now, they are concentrating on more complex tickets that require human intervention, and getting to those tickets more quickly in the queue.
Florian and his team had a modest scope for the outset of this project; namely, to build and deploy a simple agentic first step that helps support technicians and customers. Now that they have met this goal, the team is turning to add additional elements on their roadmap to scale the value of the agentic orchestration within this process.
First, the team is onboarding an entirely new customer ticketing system and integrating it into the process in Spring 2026. With this new system in place, there will be even less need for human technician guardrails at the onset of the process. AI agents will automatically try to solve any ticket categorized as “customer problem.” Aside from the short-term effect of automating more customer support, this approach will help Thomas’s team learn where the agent actually needs human intervention to solve a problem; with this information, his team can work to create automation solutions that further reduce friction and manual work from the technicians.
Additionally, the team is investigating a potential multi-agent opportunity within the process, using Camunda to connect another internal LLM-based agent that supports the customer hotline; the goal is to have that agent transcribe the customer message, detect if it is a support issue, and route it to the support system for that agent to begin working on solving the problem. This approach reduces the back-and-forth with customers, speeding up resolution time further, delighting customers, and keeping R-KOM safely within consumer protection regulations.
For Florian and his colleagues, the aspirational goal at R-KOM is to eliminate the need for manual support in the customer support process by further learning and automating work. With Camunda-powered agentic orchestration, R-KOM has taken a big step toward realizing that vision.
