{"id":17967,"date":"2026-10-02T17:28:11","date_gmt":"2026-10-02T15:28:11","guid":{"rendered":"https:\/\/agcommtech.de\/2026\/10\/02\/process-clarity-for-ai-an-interview-with-timo-radzik-of-siemens\/"},"modified":"2026-10-02T17:28:32","modified_gmt":"2026-10-02T15:28:32","slug":"process-clarity-for-ai-an-interview-with-timo-radzik-of-siemens","status":"publish","type":"post","link":"https:\/\/agcommtech.de\/en\/2026\/10\/02\/process-clarity-for-ai-an-interview-with-timo-radzik-of-siemens\/","title":{"rendered":"Process Clarity for AI: An Interview with Timo Radzik of Siemens"},"content":{"rendered":"\n

\u201cDon\u2019t Start with the Solution\u201d: Why AI Needs Process Clarity<\/strong><\/h2>\n\n

AI promises productivity. But automating processes that haven\u2019t been clearly defined only accelerates chaos. At the regular meeting of the Organization & Processes<\/a> Cluster of the AG CommTech, Timo Radzik<\/a> from the Operational Excellence division at Siemens Corporate Communications explained why process clarity is a prerequisite for scalable AI\u2014and why, before introducing the next bot, you should first put your own work to the test.<\/p>\n\n

<\/div>\n
\n
\"\"
Timo Radzik, Siemens<\/figcaption><\/figure>\n<\/div>\n
<\/div>\n\n
\n\n
<\/div>\n\n

AG CommTech: Timo <\/strong>, bots, agents, and AI workflows are popping up everywhere. Why are you focusing on processes, of all things?<\/p>\n\n

Timo Radzik:<\/strong> Because at some point, the question arises: What does this actually do for us? At Siemens, we have a wide range of options for building AI solutions. That encourages experimentation. But every solution also involves effort\u2014through token consumption, maintenance, and testing. We need to know where AI creates economic value. If we automate an unclear process, we\u2019re often just automating inefficiencies. That doesn\u2019t result in productivity gains\u2014it just leads to chaos sooner.<\/p>\n\n

\n

When we automate an unclear process, we often end up automating nothing but inefficiencies.<\/p>\n Timo Radzik\n<\/blockquote>\n\n

AG CommTech: So, process clarity first, then AI?<\/strong><\/p>\n\n

Timo Radzik:<\/strong> Exactly. One of our challenges is maintaining domain-specific knowledge in AI solutions. When multiple bots require the same knowledge, every change requires updating each knowledge base and then testing it. This effort increases with every additional bot.<\/p>\n\n

AG CommTech: What helped you identify the problem?<\/strong><\/p>\n\n

Timo Radzik:<\/strong> We mapped out the process of updating knowledge and testing our bots step by step. When is knowledge updated, how is it tested, and when is the process complete? Only then did the cost drivers become apparent. For us, these were primarily manual updates and testing.<\/p>\n\n

AG CommTech: How did you get the many knowledge bases and the testing under control?<\/strong><\/p>\n\n

Timo Radzik:<\/strong> We now treat subject matter knowledge much like our performance data. The knowledge is stored centrally in a structured, reusable format that can be utilized by multiple AI solutions. And during testing, a workflow automatically prompts the bots and compares their responses with expected results. A language model handles the evaluation of the responses. A human only intervenes if an evaluation appears unusual.<\/p>\n\n

AG CommTech: You set up <\/strong>rely heavily <\/strong>heavily on the Markdown<\/strong>file format<\/strong>. Why?<\/strong><\/p>\n\n

Timo Radzik:<\/strong> Markdown is stripped down to pure content and easy to version; at the same time, machines can process it very well without consuming many tokens. Word or PDF files, on the other hand, add extra overhead for AI. This is very well-suited for knowledge objects and process descriptions, and I can store them on our company\u2019s GitLab and distribute them through it. Similar to our performance data, this creates the desired \u201csource of truth.\u201d<\/p>\n\n

AG CommTech: What role do clear responsibilities play?<\/strong><\/p>\n\n

Timo Radzik:<\/strong> In our examples, a clear separation between subject-matter responsibility and process responsibility has proven effective. Someone must ensure that the relevant knowledge is accurate. The process owner is responsible for the workflow. It must be clear what triggers the process, who approves the result, and when it is completed.<\/p>\n\n

AG CommTech: Does this process have to be perfectly documented before it can be automated?<\/strong><\/p>\n\n

Timo Radzik:<\/strong> No, our approach here is also iterative. Process clarity doesn\u2019t mean drawing the perfect diagram once and for all. You document the current process as best you can, automate meaningful steps, and update the process documentation. This creates a cycle of documentation, testing, and improvement.<\/p>\n\n

Real-World Example: Automating YouTube Metadata<\/h3>\n\n

AG CommTech: Another example is your work on improving YouTube metadata. What did you learn from that?<\/strong><\/p>\n\n

Timo Radzik:<\/strong> There was a process in place involving transcripts, subtitles, translations, tags, and our communication standards. Bots already existed for individual steps, but people had to operate them one after another. After we documented the current process, we were able to consolidate several steps into a single workflow. Today, a transcript is uploaded to a form, and the required information is returned within a few minutes. For YouTube metadata, we save about 30 minutes per run by partially automating the process.<\/p>\n\n

AG CommTech: So process documentation also makes the business case more robust?<\/strong><\/p>\n\n

Timo Radzik:<\/strong> Absolutely. Once I know the steps, the time required, and the resources, I can do the math: What work will be eliminated through automation? What new effort will AI require? What are the costs of operation and token consumption? Then I can decide whether a solution is worth it. When it comes to prioritization, the classic logic of business value and effort helps: develop solutions with high value and low effort first; avoid those with low value and high effort.<\/p>\n\n

Governance and the New Role of Communicators<\/strong><\/h3>\n\n

AG CommTech: Many communicators see process documentation as a chore. How do you motivate them?<\/strong><\/p>\n\n

Timo Radzik:<\/strong> Not by telling them, \u201cGo ahead and document all your processes now.\u201d That doesn\u2019t work in the long run. A better starting point is a necessary decision, a specific problem, or the desire to implement AI. Often the question is: What is a time-consuming or costly task? And what does the process surrounding that task look like? Suddenly, documentation has a purpose. I recommend using the Standard Operating Procedure (SOP) format here.<\/p>\n\n

AG CommTech: How much <\/strong>governance<\/strong> is needed?<\/strong><\/p>\n\n

Timo Radzik:<\/strong> Currently, when developing AI solutions, we rely on a balanced mix of enablement and governance. Teams should be able to gain experience. At the same time, we try to create transparency regarding solutions, quality, and knowledge structures so that we can demonstrate productivity gains, especially with more complex developments.<\/p>\n\n

AG CommTech: How does this change the role of communicators?<\/strong><\/p>\n\n

Timo Radzik:<\/strong> People are increasingly becoming organizers of knowledge, standards, and quality. In the YouTube example, the social media manager doesn\u2019t have to personally share their knowledge with every inquiry. They maintain it centrally, and processes access it. Not every communicator has to calculate business cases or build automations. But everyone should be able to describe what they do, which steps are repetitive, and where patterns lie.<\/p>\n\n

AG CommTech: What\u2019s your advice for the next AI project?<\/strong><\/p>\n\n

Timo Radzik:<\/strong> Don\u2019t start with the solution. Identify a specific pain point, map out the actual process, assign responsibilities, and estimate the effort and benefits. Then automate only the steps where it\u2019s worth doing so. Afterward, document the new process and analyze it again. To me, being \u201cAI-ready\u201d means, above all, understanding my own work well enough to use AI in a truly productive way.<\/p>\n\n