> ## Documentation Index
> Fetch the complete documentation index at: https://docs.rulebase.co/llms.txt
> Use this file to discover all available pages before exploring further.

# Coaching simulations

> Assign AI role-play exercises so agents can practice a scenario and get scored automatically.

A coaching simulation is an AI role-play exercise. You describe a scenario;
Rulebase generates a customer persona and a step-by-step scenario plan. You
assign it to agents, and each agent handles the conversation while the AI stays
in character as the customer. When they finish, Rulebase scores the attempt and
returns a report.

Use simulations to build a skill before agents face it live: a tricky refund,
a de-escalation, a compliance-sensitive verification.

## Before you start

* **Coaching** must be enabled for your organization ([Settings > Features](https://app.rulebase.co/settings/features)).
* You need a coach or admin role to author and assign simulations. Agents
  complete the simulations assigned to them.
* Decide which **scorecard** you want to grade against if you plan to score the
  attempt; simulations reuse your QA scorecard criteria.

## Create a simulation

1. Go to the [Coaching](https://app.rulebase.co/coaching) page and click **New**.
   Rulebase creates a draft and opens the editor.
2. Fill in the setup on the left:
   * **Simulation name** and **due date**.
   * **Duration** — the time limit shown to the agent as a countdown.
   * **Language** for the role-play.
   * **Scorecard** — the criteria the attempt is graded against (optional).
   * **Scenario type** — **chat** or **email**.
   * **Scenario** — describe the situation in plain language, or **upload a
     training document** to base it on instead.
3. Click **Generate**. Rulebase produces:
   * a **persona** (name, photo, and characteristics like "Long-time customer"
     or "Paid subscription") that you can edit or regenerate, and
   * a **scenario plan** — a titled scenario with ordered steps, and the tasks
     the agent is expected to complete at each step.

<img src="https://mintcdn.com/rulebase/r5Wr8jebJAzAsiac/images/coaching-simulation-create.png?fit=max&auto=format&n=r5Wr8jebJAzAsiac&q=85&s=aaedc3db9ac1eb1da5cb01fab5db36b9" alt="Creating a simulation: setup fields on the left; a generated persona and step-by-step scenario plan on the right" width="2880" height="2000" data-path="images/coaching-simulation-create.png" />

<Tip>
  Regenerate the persona or edit its characteristics until the customer feels
  realistic for your team. The scenario plan's steps and tasks become the
  yardstick the agent is measured against, so review them before publishing.
</Tip>

### Scoring options

* **Passing score** — the threshold an attempt must reach to pass (the default
  is 75%).
* **Reattempts** — whether agents can retry, and how many times. An assignment
  completes once the agent passes or runs out of attempts.

## Preview, assign, and publish

1. Use **Simulate** to preview the role-play yourself before it goes out.
2. On the **Assign** step, choose the agents who should complete it.
3. **Publish** the simulation. Drafts aren't live and can't be assigned;
   reverting a published simulation to draft clears its assignments.

<img src="https://mintcdn.com/rulebase/r5Wr8jebJAzAsiac/images/coaching-simulation-assign.png?fit=max&auto=format&n=r5Wr8jebJAzAsiac&q=85&s=f3b187a890f0be6fcec34459818638e8" alt="Assigning a simulation to agents on the Assign step" width="2880" height="2000" data-path="images/coaching-simulation-assign.png" />

## The agent experience

Assigned agents open the simulation from their learning view and work through a
live conversation:

* The **customer details** panel shows the persona and a running countdown for
  the duration.
* The AI opens and responds in character as the customer; the agent replies as
  the support rep.
* Depending on how your workspace is configured, agents may get in-conversation
  help: a **"what are you meant to do next"** panel with suggested actions,
  hints, or the ability to apply **macros**.

<img src="https://mintcdn.com/rulebase/Bm2BhWSW1pIM7u6i/images/coaching-simulation-chat.png?fit=max&auto=format&n=Bm2BhWSW1pIM7u6i&q=85&s=4d65d56d70bc6a2d2620b701bec485ad" alt="A simulation in progress: the AI customer persona and timer on the left, the role-play conversation and next-step hints on the right" width="1255" height="988" data-path="images/coaching-simulation-chat.png" />

## Scoring and the report

When the agent ends the conversation, Rulebase evaluates the attempt and
produces a report with:

* an overall **score** and pass/fail against the passing score,
* a **summary** of how the agent handled the scenario,
* **knowledge gaps** and **key issues** to work on, and
* a per-criterion **scorecard breakdown** when a scorecard was attached.

If reattempts are allowed and the agent didn't pass, they can try again until
they pass or exhaust their attempts.

## Reminders for unattempted simulations

For a published simulation with a due date, agents who haven't started an
assignment can be reminded before it's due. Configure how many days ahead to
remind on the simulation. Reminders stop automatically once the agent attempts
or completes the simulation, or if the assignment is removed.

<Note>
  Assignment emails, including due-soon reminders, depend on notifications being
  enabled for your organization. If your agents aren't receiving reminders, check
  with your Rulebase contact.
</Note>

## Related

* [Coaching overview](/guides/coaching/overview)
* [Run a coaching session](/guides/coaching/run-a-session)