70 per cent talk to the phone agent: AI in practice in Germany's mid-sized companies

Martin Jäger is an AI and automation consultant from the Böblingen area. He came to the field through AI-based spelling correction, because of second-degree dyslexia. Today, by his own account, he wins 70 per cent of his clients through TikTok.

What follows is not a study and not a statistic, but the direct assessment of a practitioner.

Two reactions, no third

Jäger's observation from the mid-sized sector is clear. He mostly sees either denial (“that won't happen here”) or fear for one's existence. A genuine engagement with the subject rarely takes place.

That is the actual finding, and it explains more than any adoption statistic. Both reactions lead to the same result: no decision. Anyone in denial does not decide, because there is no issue. Anyone who is afraid does not decide, because every decision looks too large.

For managers this yields a practical task: to set the frame so tightly that a decision becomes small. A bounded use case with a time limit and a budget is a decision somebody can take. “How do we deal with AI” is not.

Who is affected, and why intuition misleads

Jäger is hard on professions that are considered safe. He regards recruiting, sales and interpreting services as the most affected areas, because in his observation what happens there is frequently pattern recognition rather than genuine interpersonal work.

The sharpening is open to attack and it hits a point. A sales process that consists of working through a list is something other than a sales relationship. The first can be automated, the second cannot. Both run under the same job title.

At the same time he warns against regarding highly specialised experts such as tax advisers or lawyers as safe. His argument: self-contained expert knowledge is easy for a language model to open up, because it is clearly bounded and well documented. What remains hard to open up is experience with exceptions and the ability to deal with people in uncomfortable situations.

The most tangible finding: the telephone

The episode becomes most concrete on conversational AI. In Jäger's projects 70 per cent of callers willingly speak to a phone agent instead of leaving a message on an answering machine. The use cases range from onboarding at the tax adviser's office to taking orders on the building site in the trades.

The figure is more plausible than it first sounds, because the benchmark decides. Compared with a human being, an agent looks inferior. Compared with an answering machine, a hold queue or no availability at all, it is superior, because it answers immediately and takes down the request.

This is exactly where many rollout projects go wrong: they measure against the ideal case instead of against the actual state. In the trades the actual state is often that nobody picks up.

Jäger's thesis on this: by 2027 every company will have its own phone agent. Whether that comes true is open. The direction is well founded.

When agents negotiate with each other

One outlook concerns application processes. Instead of two DIN A4 pages, agents on both sides could negotiate with each other in future.

That raises a question the episode leaves open and that becomes practical for every HR department: if both sides deploy agents, the competition shifts from qualification to the quality of the agent. Anyone who does not want that has to keep the process deliberately human at one point.

Conclusion

This episode supplies no evidence, but observations from practice, and in places it is deliberately sharpened. Two things from it are nevertheless immediately useful.

First: measure new tools against the actual state, not against the ideal case. A phone agent loses against an experienced member of staff. Against an answering machine it wins clearly.

Second: make decisions small enough that somebody can take them. Denial and fear for one's existence are both forms of not deciding, and the way out is the same.

As an example of how to deal with employees, the episode cites the circular email from Fiverr chief executive Micha Kaufmann to his workforce: brutally honest instead of wrapped in cotton wool. That is uncomfortable and more respectful than any soothing wording nobody believes.