Two admins building the same "skeptical CFO" agent used to end up with two different buyers, because the only control was a paragraph of prompt text, and no two people write that paragraph the same way twice.
Why this matters
Free-form prompting is fast, but it makes buyer difficulty inconsistent in ways that are hard to see and hard to fix. A "hard" agent used by a team in one region and a team in another needs to actually be hard the same way, or coaching benchmarks stop meaning anything across the org. And building distinct buyer types, a skeptical exec versus a curious evaluator versus a price-focused prospect, shouldn't require guessing at adjectives. It should be a setting.
How it works
Admins now tune roleplay agent behavior across four structured categories instead of prompt text alone.
Information strategy controls how proactively the buyer volunteers details like budget or timeline, how direct their answers are versus how often they deflect, and when in the conversation pain points surface and how deep they go. Engagement progression controls the buyer's starting stance, from open to guarded to cold, how quickly rapport and credibility build, and what regression triggers, like a pushy pitch or a wrong assumption, cause the buyer to step back.
Difficulty settings cover objection sophistication, skepticism, patience, and decision complexity, plus adaptive difficulty, which adjusts mid-call based on rep performance over a defined window of recent turns, with a separate adjustment-speed setting controlling how fast that change ramps in once it triggers. Conversation style governs verbosity, storytelling tendency, and how often the buyer asks questions back.
These settings apply to the buyer or decision-maker phase specifically. If the roleplay includes a gatekeeper, that phase keeps its own separate scorecard and persona, untouched by these behavior settings.
The background environment setting, ambient office or reception noise, follows the same split: it applies to the gatekeeper call when one is configured, and to the main buyer call when it isn't. Configurations can be saved as presets and applied to other agents, so a "skeptical exec, hard" profile stays identical across every team that uses it. Full setting definitions and examples are in Intro to roleplay persona customization.
What this unlocks
An enablement lead can build a skeptical exec profile with low proactive sharing, high deflection, and high credibility establishment, then build a curious evaluator with the opposite settings, without touching the underlying prompt at all. A team standardizing training across regions can save one difficulty preset and apply it everywhere, instead of hoping five different prompt-writers landed on comparable toughness.
And a self-serve practice pool can run adaptive difficulty, so the same agent gets harder for a rep who is breezing through and eases up for one who is struggling. An account management team prepping for renewal negotiations can build a price-focused prospect profile instead, high information withholding on budget, a regression trigger that turns the buyer guarded the moment a rep offers a discount too early, so reps rehearse holding on value before conceding on price.
The bigger picture
The prompt still tells the buyer's story. These settings decide whether every rep who steps into that story gets the same version of it.
See it in Action
Tell us what you're working on, whether that's sales calls, support tickets, or leadership coaching, and we'll show you how it fits.
Talk to Sales