At a Glance
| Outdoo AI | DIY build in ElevenLabs | |
|---|---|---|
| Persona creation | Built from a prompt, transcript, call recording, or LinkedIn profile | Hand-write one system prompt per persona |
| Stakeholders per call | Up to three AI stakeholders in the same call | One agent at a time; transfer hands off, it does not add a second voice to the room |
| Scoring | Scorecards graded against your own playbook, consistently across every rep | Custom pass/fail/unknown checks you write yourself, per agent |
| Manager visibility & assignment | Manager dashboards, courses, batches, and call-blitz assignment | Admin/Member roles only; no "my reports" view, no due dates |
| CRM / LMS integration | Built-in CRM, LMS, and Slack integrations | Webhook tools you wire up and maintain |
Building the Agent, Step by Step
Open the ElevenLabs dashboard, create a new agent, and start from the Blank template. That gives you an empty shell with five tabs, and that's the whole surface area you're working with.
The Agent tab is where the persona actually lives. You set a First message, the line the agent opens with, and a System prompt, one text field holding the personality, the objections it's supposed to raise, and any guardrails.
Write it like you're briefing an actor: skeptical VP of Ops, mid-market budget, has been burned by an overselling vendor before.
Knowledge Base is next. Drop in a pricing sheet, a competitor comparison doc, whatever the persona should be able to reference.
ElevenLabs retrieves from these documents so the agent cites specifics instead of improvising, and there's a source-attribution toggle if you want it naming which document it pulled from.
On the Voice tab, pick a stock voice from the library or clone one.
Instant Voice Clone needs one to two minutes of clean audio and a consent check.
Professional Voice Clone wants up to three hours of source audio and gets you a much tighter match if the goal is a specific real voice.
Tools covers webhook tools, which let the agent hit an external API mid-call, useful for logging the conversation to a CRM.
There's also a system tool called agent transfer, which hands the conversation off to a different agent when a condition fires, something like "user asks about pricing."
Hold onto that detail: it's a handoff between two agents, one at a time, not three personas sitting in the same call.
On the Analysis tab, you define Success Evaluation criteria: did the rep handle the pricing objection correctly, scored success, failure, or unknown, with a rationale attached.
Hit Test AI agent and talk to it live, or run the Simulate Conversation API for text-based tests at volume before anyone touches it for real.
To put it in front of people, paste a widget script tag onto a page (`<elevenlabs-convai agent-id="...">`), or route it through Twilio or another SIP provider so reps can call in from a phone. Either way, in an afternoon, you have a working AI buyer that talks back.
It Genuinely Works, Which Is the Whole Point
ElevenLabs has its own published case study on this: a team coaching reps with AI roleplay agents they built entirely in-house.
They wrote persona reveal logic by hand, so the buyer only gives up budget and timeline after real discovery questions, and a three-criteria, 1-to-3 scoring rubric, per scenario, with no engineering support.
Completion rates cleared 90 percent, well above the 40-60 percent typical of peer-to-peer roleplay.
That's a real result. It's also a fair preview of the labor involved: every persona, every rubric, every scenario is a separate build.
Where It Breaks Once It's a Team, Not a Demo
Go looking for the gaps and they show up fast, and they show up specifically at the team level, not the single-agent level.
The Analysis tab reports calls, cost, latency, and pass rate per criterion, for one agent at a time.
There is no rollup showing how an SDR cohort trended over the quarter, and no line showing whether one rep is actually getting better week over week.
Workspace roles are Admin and Member, with Full and Basic seats and user groups for bulk permissions.
None of that gives you a Manager role that only sees their own reports, and there is no way to say "everyone in this cohort completes Scenario 4 by Friday."
You can grant access. You cannot assign work.
Multi-stakeholder roleplay, the kind where a buyer, a technical evaluator, and procurement all push back in the same conversation, is not something you configure.
Agent transfer moves a call to a different agent; it does not seat two more personas at the same table.
Security is a real strength here, not a gap. ElevenLabs carries SOC 2 Type II, ISO 27001, and PCI DSS Level 1, plus HIPAA and GDPR attestations, zero-retention mode, and PII redaction.
That's platform-grade compliance, and it's worth crediting. It answers a different question than sales-training governance, though: who's allowed to see whose scores, and who assigned what to whom.
Then cost. Billing runs $0.08 to $0.10 per minute depending on tier, on top of separate LLM and telephony charges, and that's per agent.
A buyer persona, a technical evaluator persona, a procurement persona: each one is its own prompt, its own knowledge base, its own upkeep, and its own running bill.
The "no engineering support" story from ElevenLabs' own case study gets a lot harder to repeat once you're maintaining a dozen scenarios for fifty reps instead of one scenario for a pilot group.
Where Outdoo AI Picks Up
None of this is a mark against ElevenLabs as a voice platform. It's a strong way to build one agent, and the case study proves it. The gap opens specifically where "one agent" has to become "a program everyone runs on."
Outdoo AI starts from what reps already have: roleplays built from a team's own call transcripts, playbooks, battlecards, or a rep's LinkedIn profile, so the practice matches the actual job.
Calls can run with up to three AI stakeholders at once, so committee-based B2B deals get rehearsed the way they actually happen.
Scoring runs against your own playbook, and the same scorecard grades roleplay practice and live calls, so a manager can point to a coaching change and see it move a real number, not just a completion count.
Admin, Manager, and User roles come with actual assignment: courses, batches, and call-blitz campaigns a manager can push to a cohort with a due date, plus dashboards scoped to just their own reports.
CRM, LMS, and Slack connections are already built, so this plugs straight into the enablement stack that's already running. Pricing is usage-based, so you pay for training that's actually happening.
If you're building one agent to see whether this idea works at all, ElevenLabs is a fine place to spend an afternoon. If you're trying to run it across a sales team with any consistency, that's the problem Outdoo AI is built for.
See how Outdoo AI handles multi-persona calls and manager dashboards out of the box: book a demo.
Frequently Asked Questions
.svg)
Yes. The Blank template in the ElevenLabs dashboard gives you a persona, knowledge base, voice, and scoring rubric through five tabs, with no engineering help required.
Billing runs $0.08 to $0.10 per minute depending on tier, plus separate LLM and telephony charges, and that cost applies per agent.
No. The agent transfer tool hands a conversation off to a different agent when a condition fires, but that's a one-at-a-time handoff, not a second persona joining the same call.
ElevenLabs carries SOC 2 Type II, ISO 27001, and PCI DSS Level 1, plus HIPAA and GDPR attestations, along with zero-retention mode and PII redaction.
Up to three AI stakeholders can be in the same call at once, for rehearsing committee-based B2B deals where a buyer, technical evaluator, and procurement all weigh in.
Table of Contents
Talk to Sales
Have questions about training and enablement for your sales, CS, support, or leadership team? Let's talk.
Talk to Sales
