We're the Evolve team
We've been building AI into corporate learning since 2023. We've created courses, knowledge bases, and AI simulations for companies like Carlsberg, Schneider Electric, and Freedom Pay Having built hundreds of courses and simulations across different markets and industries, we've learned that AI simulations are the most effective way to boost your team's performance and get them genuinely engaged in training. We also know exactly how to make course creation several times faster. That's what we're sharing in this playbook.
Enjoy the read!
AI Simulations for Practice & Skill Building
You want your training to actually engage people — not just make them click through slides. You want interactive scenarios, real-world practice, something that makes employees think instead of just passing a test.
Companies like Bank of America, Walmart, Accenture, Schneider Electric, Carlsberg, etс. already use AI simulations for exactly this — and it's the most effective way to train the specific skills each person is missing, without pulling managers into the process every time.
How it works
1. You identify each employee's weak spots based on assessments, completed courses, or call analysis (you can build this with AI agents)
2. Then an AI agent plays the role of a client, mentor, or negotiation counterpart.
3. Sales, customer success and other team members go through at least one role-play or simulator with AI every day.
4. Every session generates a unique scenario focused on that person's specific gaps.
5. The interface can range from a text-based simulator to a real-time voice conversation with an AI avatar.
6. There can be two types of scenarios: a client scenario, where the employee practices with a simulated customer or counterpart and a mentor scenario, where AI acts as a senior colleague walking them through a situation step by step.
7. After each simulation, AI delivers a detailed breakdown: mistakes, strengths, specific recommendations, and an assessment visible to HR and managers.
It is fully customizable to the specific employee and goals: client persona, script, agent behavior, difficulty level.
How to built a simple version
Step 1. Define your avatars, scenarios, and system prompts
For each role and scenario you configure:
Who the agent is (persona, background, personality, role). For sales or support call simulations, you describe your ICP or a typical client. For example:
You're Martin, 45, chief engineer at a manufacturing plant. Your facility runs 24/7, any downtime costs the company thousands per hour. You're evaluating equipment upgrades but need to justify every purchase to management. You're technical, direct, and expect the rep to know the product inside out.or
You're a senior project manager with 12 years at the company. A new hire just joined your team and is about to run their first client kickoff meeting. Walk them through the process step by step — ask what they would do first, how they'd prepare the agenda, what questions they'd ask the client. If they miss something important, don't give the answer right away — ask a guiding question to help them figure it outHow they talk (tone, style, conversation guidelines), for example:
Maintain a calm, supportive tone. Speak in a conversational, encouraging manner. Ask questions more often than make statements. Pause after each question, give time to think.Customize the prompt for each scenario and role From here, you tailor the prompt depending on what exactly you need from each simulation. You define an objective and tasks for the agent and specific instructions for different situations. Pro tip for your prompt:
If the customer's response is unclear, ask clarifying questions. If you encounter any issues, inform the customer politely and ask to repeat.How to score the rep after the session Define evaluation criteria with weights, for example:
Opening (greeting, introduction, confirming availability) Discovery (open-ended questions, identifying pain points, understanding context) Value pitch (connecting the offer to needs, specific benefits, relevant examples) Objection handling (active listening, reframing, confident responses) Closing and next steps (asking for commitment, setting specific next steps)
The assessment agent should be a separate agent that reviews the full conversation after the session and scores the rep's performance based on your criteria.
Resources on prompt engineering to help you write better system prompts:
Step 2. Assemble the stack
If you want a voice-based simulator, you can use Vapi, a platform for building voice AI agents. It works like a constructor where you connect all the components together:
LLM (GPT-4o works best for this task) generates the agent's responses based on your prompt and scenario.
ElevenLabs converts those responses into natural-sounding speech in real time.
Deepgram transcribes the employee's spoken responses into text for the LLM to process.
For a text-based simulator, you can build one on N8N, for example, where you hook up GPT working off your prompt to any interface you need (web chat, Slack, etc.).
Some learning platforms also offer built-in simulation features.
For example, Evolve built functionality where simulators are auto-generated from course content and the employee's existing assessment results. The platform identifies the employee's weak spots based on how they performed in the course and generates a simulation that specifically targets those gaps. Each session is fully unique, regenerated every time. Results feed into the employee's profile, where the platform combines simulator performance with course completion, open-ended answers, and 360 reviews, giving the manager a full picture of each employee's strengths and gaps.
Why it matters
Role-plays with a manager or trainer are the most effective method for sales training, but they eat up the sales manager's time. On top of coaching, managers deal with processes, cross-team coordination, and a dozen other things. Running daily role-plays with every rep on the team is simply not realistic.
An AI simulator takes this entire routine off the manager's plate and runs it for the whole team, every single day. Previously, between discovering a problem (say, a rep handles objections poorly) and fixing it, a week or two would pass and during all that time the rep was wasting leads.
An AI simulator collapses the feedback loop to a single day.
AI-Enhanced Content Creation
You have SOPs, product documentation, onboarding guides, compliance materials sitting in PDFs and Word files, and turning all of that into actual courses normally takes weeks of instructional designer's time, because someone has to read through everything, figure out the structure, rewrite the content in a way that's actually learnable, and then proofread the whole thing. With AI agents, you can compress this entire pipeline into hours, because each stage of course creation is handled by a separate agent with its own role, and they work sequentially, one after another, passing the result to the next.
How it works
You upload your source materials — PDFs, Word docs, presentations, whatever you have — and set the parameters that will be the part of the prompt for the first and the second agents:
Company description: what the company does, what products or services it offers, which regions it operates in, key business specifics. The agent needs this context to correctly interpret the source materials and maintain the right tone
Course title
Who this course is for: department, level, role (e.g. "sales team, new account managers"), new employees or existing.
Course goal: what specifically the employee should be able to do or understand after completing the course (e.g. "understand sales standards, handle objections, adapt quickly to the company")
Material processing preference: keep the text as is (just structure it), stylistically adapt it (rewrite into a softer, more engaging learning format), or simplify and shorten to the key points
Estimated course duration and lesson length
Course language
Tone of voice
From there, three agents take over:
1. Content extractor It reads through all your source documents, pulls out what's relevant, and builds a course structure: which chapters, which lessons, what content goes where. It accounts for everything you specified — length, depth, what to include, what to skip. You can review the structure after this step and adjust it before moving on.
2. Writer It takes the structure from the first agent and composes the actual lesson text. In the writer's prompt you set the tone of voice and how to work with the source materials: keep the text as is and just split it into lessons, stylistically adapt it into a softer learning format, or simplify and shorten to the key points.
This is also where you can set the learning methodology like Kolb's learning cycle, the prompt for which you can find below.
Writer agent prompt (Kolb's learning cycle)
You will receive source text: plain documentation, tables, definitions, or technical descriptions.
Your task is to transform this text into a lesson for an introductory self-paced online course.
Write concisely, to the point, and in clear language.
This is not an advanced course or an expert deep-dive. This is introductory training, and the goal is to give learners a basic understanding of the topic and a sense of its logic, not an exhaustive theory.
The text should be:
- easy to absorb
- readable on screen
- free of overloaded explanations
The course has a narrator character, but they are not introduced explicitly.
Write everything in the consistent tone of a calm, supportive mentor — conversational, human, without lecture-style or "expert" pressure.
The course follows Kolb's learning cycle, but you don't need to implement the full cycle in every lesson.
In each lesson, lean on just one element:
- connecting to personal experience
- observation and reflection
- simple generalization
- mental application
Don't try to add all four at once. Follow the meaning of the source text.
The course is about [describe the topic of your course and subject area, e.g. "client consultation in skincare", "safety procedures for warehouse operations", "product knowledge for the new CRM system"].
Present the material so that the learner:
- understands the basic logic of the topic
- can recognize situations from their own practice or life
- [describe the practical learning outcome for your employee, e.g. "starts approaching product selection more consciously", "understands why following the protocol matters", "can independently perform basic operations in the system"]
Use minimal examples — only those that genuinely help explain the meaning.
One good example is better than several detailed ones.
Add reflection questions occasionally, but:
- no more than 1–2 per lesson
- without deep analysis
- just a light invitation to think about their own experience
If the source text contains tables or definitions, use them as a conceptual foundation, but explain briefly, without excessive detail.
Avoid:
- long-winded reasoning
- repeating the same idea in different words
- trying to "finish the thought" for the reader
Write so that the lesson:
- feels clear and understandable
- doesn't exhaust the learner
- fits into a short completion time
The result should be a concise, introductory, engaging lesson that helps the learner understand the topic and be ready to move on, not one that tries to cover everything
[Insert an example: take any piece of your source material (a table, a list of definitions, a product catalog, or a technical paragraph). Below it, attach an example of a good lesson based on that material (e.g. written by an instructional designer).]
3. Proofreader It reviews the final text and catches errors without rewriting the content itself, because deep restructuring would distort the original material and hurt the quality of the course.
The proofreader's prompt should include your specific formatting and style rules: spelling, grammar, and punctuation corrections, how lists should be formatted (e.g. bulleted lists with dashes, numbered lists with periods), how numbers and dates should appear, which quotation marks to use for which language, how to handle repetitive wording, and so on.
To get started, you can set up agents like these in Custom GPTs or Claude Projects.
But when it comes to scaling across an entire company, you'll need a solution that protects personal data and sensitive information — this is where corporate learning platforms with a built-in AI editor come in (like Evolve), running on certified cloud infrastructure (SOC 2, GDPR, ISO 27001) or deployed on-premise on your own servers.
If you want to find out how your specific processes can be optimized and how to build a truly great training system — come to a free scoping session with Evolve!