Smart Role vs. Eigenbau mit einem LLM

Prompting a general-purpose assistant to "act as a difficult customer" works, and it is a genuinely good way to find out whether role-play helps your team. Several of our customers started exactly there.

It stops working when you try to make it a programme.

What you get for free

A persona, in any language, immediately. If one person wants to rehearse before a call this afternoon, this is the right tool and there is no reason to buy anything.

What you have to build

The gap is not the conversation, it is everything around it:

• Consistency. The same prompt produces a different difficulty each time, so two people are never tested equally.
• Voice. Real-time speech with interruption handling and noise suppression is an engineering project, not a prompt.
• Scoring. Weighted criteria, evidence quotes, passing scores and the plumbing to keep them stable across model versions.
• Everything organisational. Accounts, roles, assignment, progress tracking, retention policy, SSO, audit.
• Data governance. A consumer assistant is the wrong place for a recording of an employee's performance, and your DPO will say so.

The honest recommendation

Prototype with a prompt this week. If people run it twice and ask for more, you have your answer — and the question becomes whether you want to maintain the other ninety percent yourself.

Our API exists precisely for teams that want the engine without our interface.