What works and what does not
- Works: objection drills, difficult-customer practice, product knowledge checks, generating alternative phrasings.
- Does not work: training without reviewing real recordings. In the drill a rep says one thing; with a real customer, another.
role-play prompt
You are playing a customer. Stay in role until I write "stop". Who you are: [customer description - industry, role, situation] What you want: [the customer's goal] Your main objection: [too expensive / need to think / we already have one / no time] Your manner: not rude, but sceptical. You do not dodge direct questions, and you do not help the seller either. Rules of the game: - Reply briefly, the way people write in a messenger: one or two sentences. - Do not agree on the first attempt. Give in only if the seller has done three things: found out what you actually need, quoted a price directly, and proposed a specific next step with a date. - If the seller starts pushing or arguing, cool off and shorten your replies. - If the seller asks about features instead of asking about your situation, answer flatly and do not open up. - Never hint at what the seller should do next. When I write "stop", drop the role and give me a debrief: 1. At which moment could you have walked away, and why. Quote it. 2. Which question were you waiting for and never heard. 3. What exactly would have made you agree sooner.
Building it into onboarding
- Take five real objections from your own matrix, not invented ones.
- Write a customer description from your own segment for each.
- A new rep runs five conversations in their first week.
- Every conversation is saved and read by their manager: what does this person do under pressure.
- In week two, the same five with a harder customer.
The difference from a live review matters. The drill shows what a rep is capable of. Recordings of real conversations show what they actually do. The second is what counts, which is why the drill supplements reviews rather than replacing them.
Other things that work
- product knowledge: ask the model to fire twenty customer questions and check the rep's answers against your documentation
- alternative phrasings: three versions of the same clarifying question, and the rep picks the one they would actually say
- your own conversation seen from the other side: give the model a transcript and ask it to answer as the customer explaining why they left
What not to do
- do not let the model write your scripts from scratch: you will get the market average instead of yours
- do not let the drill replace the weekly review of real recordings
- do not judge a rep on role-play performance: there is no cost of failure in it