AI agents are supposed to answer inquiries, draft proposals, and confirm appointments. Regina Schnathmann, Head of Brand Management, Marketing & Communications at VDE, spoke with Carsten Ratzlaff about which of these functions are already working today and who decides how much an agent is allowed to do. The two know each other from the CXO Round Tables, which he moderated during his time at Oracle. On November 25, he’ll continue the series with Fraunhofer IAO.
Regina Schnathmann: AI agents are the big topic right now at conferences and in press releases. Anyone who works in communications knows the gap between announcements and everyday reality. What has actually taken hold in companies?
Carsten Ratzlaff: Less than the announcements suggest, and I think that’s a good sign. In their study “Understanding and Applying AI Agents,” Fraunhofer IAO and Fraunhofer IAIS developed a framework that measures, on a scale from 0 to 100 percent, how “agent-like” a system actually is. Their examples from sales and technical customer service score 36 and 42 percent, respectively, and in terms of autonomy, four out of five case studies are at the level where a human is in control. Fraunhofer even advises starting with internal applications that don’t require contact with customers or partners. That’s the right order.
Regina Schnathmann: For a brand, every customer interaction is a promise. When an AI agent speaks with customers, it makes that promise on behalf of the company. At what point is it allowed to do so?
Carsten Ratzlaff: As soon as the company allows it—and that happens step by step. The vendor provides capabilities; management decides how much of that an agent is allowed to use. At every step, I ask how easily a mistake can be rectified. Incorrect information can be corrected as long as the customer hasn’t acted on it yet, and a wrongly rescheduled appointment costs, at worst, a second contact. It’s different with goodwill gestures or a confirmed delivery date: In those cases, the agent speaks for the company, and the customer takes them at their word. It doesn’t matter to the customer whether a person or a software program made the promise. They associate the promise with the company’s name. Anyone with a technical background knows the principle: Approval is granted only after verification. For agents, this means determining in advance what needs to be documented, who signs off on it, and how to quickly scale back the decision.
“The manufacturer provides capabilities; management decides how much of that an agent is allowed to use.”
Regina Schnathmann: Many companies in the VDE sector sell technical products, and often the service determines whether a customer stays. What specific steps should these companies take now?
Carsten Ratzlaff: For starters, one page of paper is enough. Take every agent who deals with customers (or whom you’re currently planning to hire), and for each of their steps, write down whether they act independently, whether a person approves it, or whether a person oversees it. Then ask who determined that. If the answer is “the manufacturer” or “the project,” management hasn’t made a decision yet. This decision falls to service or sales management; IT implements it. We’ll discuss how other companies handle this on November 25 at the CXO Round Table with Dr. Maximilien Kintz and Jens Drawehn from Fraunhofer IAO—in a small group, off the record, and without product presentations.