This piece first appeared in German. The English version is a rewrite rather than a line-by-line translation, and the answers below are translated: the substance is unchanged, the wording is not the speaker's own. The German original stays online in the archive: Interview: KI am Telefon.

Call a property management company after hours and the phone rings out. That missed call is the entire premise of Telefonagent, a Vienna startup whose product answers it instead. Jannek Drexl, its managing director, says around 80 percent of callers who reach the system go ahead and have the conversation. The interview below covers the technology underneath, where the data sits, and what he thinks the next three to five years do to first-level support.

What the product is

How would you describe Telefonagent in a few sentences?

Telefonagent is an AI telephonist. It is not meant to replace people; it takes calls automatically, on the basis of AI and software. You call it like any normal contact and have a conversation with the system. That makes most sense outside the opening hours of the many companies that have them. Our focus is the property sector, meaning building management firms, agents and so on. Where calls would otherwise be missed, the agent steps in and takes over.

How did the idea come about?

I met my co-founder Achilleas, who is the technical side of it. We are the classic duo: I do sales, he does product. I am good with people and he is good with software. We saw early that a market was developing here, secured the name and started building. I began selling before there was a finished product, when all we had was a minimum viable product. Achilleas comes out of chatbot development and I had worked in startups in service and sales.

How it works

How intelligent are today's voice models compared with a human operator, and where are the limits?

AI is only as intelligent as the data it is fed. In the property sector we want to cover specific requests, and the better the documents, the call guides and the datasets, the better the system performs. It can run anything from simple question-and-answer exchanges to more complex processes that open tickets in a system.

The limits sit in full automation. Many companies want entire workflows handled without staff, and the technology is not mature enough for that. Our customers use it mainly in first-level support or outside opening hours.

What does the product run on?

That depends on the use case. We use common large language model systems, frequently Mistral, a French provider that has established itself partly for data protection reasons, along with standard models from OpenAI and Anthropic. Those models form the base of our solution.

How do you make sure the system sounds natural and empathetic?

Through training, which in practice means prompting. You set how friendly or professional it should sound, rather like a dial for tone and style. We also define clear dos and don'ts. One example is instructing it never to address customers informally, where the client company wants it that way. The communication style can be steered quite precisely.

Where the data sits

How do you handle sensitive customer data?

Everything stays in Europe. We host in Germany and Austria, and the language models run locally, as with Mistral. For voices we partly use services such as Eleven Labs, which are hosted in Belgium, so that side is compliant too. Before every conversation the caller is told they are speaking with a digital assistant. We also concentrate on sectors like property, where no sensitive patient data arises.

How open are callers to speaking with an AI?

Many would rather speak to a person. When nobody is reachable, though, they prefer the AI to no answer at all. Once the solution is established in a company, acceptance is high. Around 80 percent have the conversation with the system and come away with a good experience. A small share hang up at the word AI, and after several unsuccessful attempts to reach someone they usually speak to the assistant anyway.

What happens next

How does AI-based telephony develop over the next three to five years?

First-level support will be covered entirely by AI, in my view. Large firms in the United States, UPS among them, already run such systems. At VivaTech in Paris it was obvious that the trend is arriving in Europe too.

Simple service calls, from delivery companies for instance, can be taken over as well. Those straightforward question-and-answer conversations will be fully automated within a few years.

Will AI eventually replace the customer adviser completely?

That depends on the work. Advisers handling simple standard requests will be replaced. Those with expertise or a genuinely advisory function stay important, because many people still want a person they trust. The specialized adviser is not going to be replaced for the foreseeable future.

Advice for people starting out

What should young professionals bring to an AI-driven economy?

I follow AI and artificial general intelligence closely myself, and studies forecast that a large share of jobs will be automated by 2031.

My advice to young people is either to go into the trades, where demand will be considerable, or, in an office job, to engage deliberately with AI tools. I look at new developments regularly, things like N8N, Eleven Labs, Zora 2 or Google Nano Banana. Staying current there is a clear advantage over anyone relying on a conventional business degree alone. AI will shape the future, without question.

The link has been copied!