AI roleplay shortens SDR ramp when it adds repetitions on the situations your reps actually meet. Scenarios built from your own call recordings do that. Generic personas do not. We have not seen an independent, audited benchmark showing that roleplay halves ramp time, so the honest approach is to measure your own: conversation-to-meeting rate at day 30 and day 60.

Why ramp is a repetition problem

A new SDR does not lack knowledge as much as reflexes. Knowing the answer to "send me an email" is easy. Giving it calmly, at second eight, with a stranger who is about to hang up, takes repetition.

Real repetitions are expensive. In the 1,000,000 French B2B cold calls we analysed with Quentin Despas, the average connect rate is 11.3%. Every connect a new rep wastes is an account you cannot easily call again. Parallel dialing gives reps 5x more conversations, which helps, but a beginner who has more conversations also burns territory faster.

The usual alternatives each fail in a specific way:

• Manager mock calls. The manager has a handful of objections, plays them politely and has thirty minutes a week.
• Listening libraries. Listening teaches what good sounds like. Only speaking builds the reflex.
• Live calls as practice. It works, and it costs you prospects.

Roleplay fills the gap: many speaking repetitions, no prospect burned. The condition is that the practice resembles the calls.

The own-calls loop in five steps

1. Tag. Take two weeks of recordings and label how the first 30 seconds ended: hang-up, objection, engaged. You are looking for where calls die.
2. Extract. Keep the five most frequent objections in the prospects' exact words, not your paraphrase. In France that is often "envoyez-moi un mail" or "on a déjà un prestataire".
3. Build. Turn each into a 60-second scenario: a persona, a trigger, and the objection at second ten. Keep it short. The opener, one discovery question and one objection are what an SDR needs.
4. Drill. Ten minutes a day, by voice, same scenario until the behaviours score well, then the next one.
5. Check. After a week, compare the same behaviours in live calls. Pull the new objections, replace the stale scenarios, repeat.

A scenario can be this small: an HR director, a trigger (the company just posted three sales roles), and an interruption at second ten, "on a déjà un prestataire". The rep has sixty seconds to open on the trigger, ask one "how" question and ask "What do you mean by already have a provider?" before answering. Repeat it twenty times and the reflex is there.

The loop matters more than the tool. It keeps the training current, because the buyer in the simulator changes when your market changes.

What to score: behaviours, not confidence

A score for "confidence" or "rapport" is an opinion. Score behaviours that moved results in our benchmark:

• Trigger or pitch. 65% of reps open by explaining what they sell. Trigger-based openers reach a 7-10% meeting rate, against 2-3% for pitch-first ones.
• "How" questions. Discovery questions starting with "how" lift conversion by 190%.
• The objection reflex. Asking "What do you mean?" before answering an objection lifts conversion by 197%.

These are binary and easy to check, in the simulator and on live calls. Our wider guide to choosing an AI roleplay tool covers the buying criteria; this article is about how to run it.

Measure ramp, not minutes practised

What to track

Practice minutes show effort, not output. Use them as a leading indicator and judge the programme on results:

• Behaviours on live calls. Share of trigger openers, share of "how" questions.
• Time to first meeting booked. In our experience, a week-one meeting does more for a rep's confidence than any workshop.
• Conversation-to-meeting rate at day 30 and day 60, against the team average. A rep who matches it by day 60 is ramped.

Keep the data for coaching. If reps think the tool is surveillance, they will game it, and the scores will mean nothing.

What a ramp week costs

Take an assumption and do the arithmetic. For example, a French SDR with a loaded cost of 45,000 euros a year costs about 3,750 euros a month. Cutting two weeks from ramp saves about 1,700 euros in salary, plus the meetings those two weeks would have produced.

That is an illustration, not a benchmark. Replace it with your own loaded cost, then compare ramp across cohorts. Be careful with the comparison: a cohort that got better lists or a calmer quarter will look better regardless of training. Compare reps who worked the same type of list.

Four mistakes that waste the programme

• Generic personas. A "skeptical VP" teaches generic answers.
• Long scenarios. Ten-minute discovery calls train the part of the call that matters least for an SDR.
• Monthly sessions. An hour a month is a workshop. Ten minutes a day is practice.
• No loop back. If practice scores never meet live-call scores, nobody knows whether it works.

How Sonik approaches AI roleplay

Sonik's roleplay is built from your own call recordings, so the buyer a rep practises against is the buyer your team actually loses to. It sits inside the dialer, which means the practice score and the live-call score can be read side by side.

We pair it with AI call prep and coaching on the behaviours above. We would rather you measure your own ramp than trust a headline, ours included.

Conclusion

Roleplay is useful when it is specific, short and tied to real calls. This week, pull twenty recent recordings, write down the three objections that end calls fastest, and turn each into a 60-second scenario. Have your newest rep run one of them for ten minutes a day, and compare their opener on live calls on Friday.