AI sales roleplay for the conversations that matter.
Voice roleplay against AI prospects with a mood, a backstory and an agenda of their own. Infinitely patient. Identically hard for everyone. Free to fail in front of.
Five levels, named for the prospect.
Named for the prospect's disposition, not the rep's skill, because that is how a salesperson sizes up a call. A rep can say “that was a Level 3” and the whole floor knows what they mean.
| Level | Prospect | Behaviour |
|---|---|---|
| 1 | Receptive | Curious, one soft objection, folds if handled at all. Never interrupts. |
| 2 | Busy | Neutral, time-pressured. Two objections. Needs a reason to keep listening. |
| 3 | Skeptical | Wants proof. Three or more objections. Remembers what you said and holds you to it. Leaves silences. |
| 4 | Resistant | Hostile or hurried open. Interrupts. Assumes you are like the last rep. A weak first ten seconds ends it. Softens noticeably to calm, specific competence. |
| 5 | Expert Buyer | Informed, pleasant, precise. Has competing proposals. Actively baits the compliance traps as if genuinely wanting a yes. Dangerous because likeable. |
The persona is instructed to be winnable at every level. A prospect who never says yes teaches learned helplessness.
Not “watch a video, then do a roleplay.”
Reps move through all three, per beat, per level. Drills are the surface used most, so the mobile experience is built for that screen first.
Recite it, out loud.
Read the script with a why-this-line note on every beat. Chunk it on a teleprompter that shrinks. Then the prompt goes blank and the rep says the beat aloud. Speech-to-text scores intent match, not word match. Gate: 85% on every beat.
One objection, fired cold.
“We already work with someone.” The rep has 45 seconds. High volume, low stakes. This is the batting cage, and where muscle memory forms. Ten of these before a shift.
The whole call.
Start to finish against a persona with a mood, a history and an agenda of their own. Scored against your rubric with transcript evidence. This is what gets certified.
Every score comes with the 20 seconds that earned it.
Weights are per track and live in your rubric. Compliance is not weighted; it is a gate. Open any row to see the transcript span behind the number. The flag card is what a manager receives within a minute of the call ending.
Ask one model to grade and police, and it softens the violation.
A single call that scores the rubric and checks compliance at once under-flags, because it has already decided the call went well. So they are two separate passes. The scoring pass has the rubric in context. The compliance pass has your rule text, verbatim, and asks one question per rule: did this occur?
The detector is tuned to over-flag. False negatives are the expensive direction. False positives cost a manager one click and make the detector better.
Rubric in context
Per-criterion score with the transcript span that justified it, and a two-line rationale a rep can act on.
Rule text, verbatim
“We can be flexible on the fee” and “don't worry about the number” share no keywords and both violate the pricing rule. This is not keyword matching.
Gate, then mastery
Any critical flag fails the attempt regardless of score. Then the trailing window is re-evaluated and any unlock, or re-lock, is recorded.
What a rep actually does, minute to minute.
- 1Rep
Opens a drill or a full roleplay. For phone tracks, the platform rings the rep's actual handset.
- 2Persona
Decides, turn by turn, which objection fires next: escalate, soften, bait, or leave a silence. Speaks in a voice chosen for the persona and the locale.
- 3Loop
Continues until the objectives are met, the time cap hits, or the persona ends it. A weak first ten seconds against a Level 4 prospect ends it early, on purpose.
- 4Scoring pass
The full audio and diarised transcript are stored. The rubric pass scores each criterion with its transcript span.
- 5Compliance pass
A separate pass reads the transcript against your rules. Any critical flag fails the attempt.
- 6Scorecard
The rep sees scores with evidence, the one thing to work on, and their level progress within a minute. A manager is messaged only if a human must act.
Tracks are yours.
A track is one selling motion with its own script, rubric weights and mastery thresholds. All tracks share the objection bank, the persona library and the compliance ruleset.
| Org type | Typical tracks | Channel |
|---|---|---|
| Inside sales | Outbound cold call · Inbound qualification · Renewal call | Phone |
| Field sales | Site visit · Consultation · Renewal visit | Browser mic |
| Retail and showroom | Floor approach · Consultation · Financing conversation | Browser mic |
| Call centre | Inbound · Retention · Upsell | Phone |
| Written channels | Chat · Email follow-up | Text |
Nothing punitive, by design.
Heatmap by criterion
Which criterion is weakest across the team, and for whom. One coaching conversation per rep, not a ranking.
Certification funnel
Who is at which level on which track, what the next requirement is, and median time to first certification. The metric to hold us to.
Compliance queue
Every flag with a listen-at-timestamp link and a false-positive button. A rep cannot dismiss their own flag.