Create a simulation using AI

Learn about best practices.

  1. On the left navigation bar, click Simulation workshop.

  2. On the Simulations tab, click Add.

    The new simulation form - start of form

  3. Enter a minimum of one to three sentences that describe the general area of brand-to-consumer interaction and several related queries or issues.

    Example: An online shopper contacting customer service about common post-purchase issues. Please include scenarios such as: a delayed shipment that hasn't arrived by the promised date, a return request for an item that didn't match the description, and a promo code that failed to apply at checkout.

  4. Click Generate.

    The new simulation form - middle of form

    The system generates a name and a description for the simulation. It also finds or generates scenarios and personas as follows:

    Scenarios: The system uses the description that you provided in step 3 to generate 3 - 5 scenarios. Then it discards any AI-generated scenarios that are similar enough to your existing ones (based on title). This isn't limited to exact matches. Scenarios with closely worded titles (e.g., singular/plural variants, minor rewording) may also be treated as duplicates and discarded, even if their descriptions differ.

    Personas: The system randomly selects up to 5 personas from your account based on availability. These include both the predefined personas that are provided by default and any custom personas that you've created.

    These scenario and persona "candidates" are then shown to you for review and selection.

  5. Select the scenarios and personas to include in the simulation. You must select at least one of each to finish the simulation setup.

    Within each AI-generated scenario, you can optionally select the routing target that gets this scenario to the proper live agents or AI agents. (This requires that you select a contact center first, as routing targets are specific to contact centers.) You can also select identities to use if the scenario relies on them.

  6. Click Finish setup.

    This saves your changes thus far. You now configure the simulation as described below.

    The new simulation form - end of form

  7. Simulation type: Choose the type of simulation. While this field doesn't currently affect simulation behavior, make an accurate selection, so your data is meaningful when options based on this value are introduced. Types include:

    • Test AI agents - This is for testing AI agents or conventional bots.
    • Train live agents - The process is transparent and for skill-building and practice.
    • Audit live agents - You're discreetly evaluating agents' performance without their knowledge.

  8. Name: If desired, change the name for the simulation.

    We recommend you surface in the name a short descriptor for the target AI agent or live agent group. Also consider including your team or line of business, the specific use case, and a version or date. This makes it easy to identify your simulations at a glance and avoid confusion when multiple teams are working in the same environment. Strategies like these help you to remember why you ran the simulation. You might also want to enter a date and time to make the simulation name unique; uniqueness of names isn't enforced.

  9. Description: If desired, change the short description of the simulation's objective or purpose. This information isn't sent to the LLM; it's just to facilitate understanding by you and others at your business.
  10. Scenarios: Make changes to the selected scenarios if desired.

  11. Customer identity: Choose the type of identity to use:

    • Generate random identity: The identities of all synthetic customers in the simulation are generated randomly. This is the case even if you have predefined identities specified in the scenarios; those predefined identities are not used.
    • Make customer anonymous: The synthetic customer is assigned only a random consumer ID.
    • Use predefined identity: The synthetic customer is assigned one of the predefined identities that's specified in the scenario. (Learn how and why identities are synchronized automatically to ensure each predefined identity is used in only one active conversation at a time.)

    If any selected scenario has no predefined identity at all, a dialog appears when you run the simulation asking whether to proceed using a random identity for that scenario instead, or go back and add one.

  12. Personas: Make any changes to the selected personas if desired.
  13. Ignore response speed: Turn this toggle on if you want the synthetic customer to send responses immediately, without regard for the Response speed set on the persona. This toggle is off by default, so the synthetic customer respects the persona's configured response speed unless you turn it on.
  14. Contact center: Select the contact center against which to run the simulation. You can select only one.

  15. Routing details: Choose how to route the conversations:

    • Start each conversation on scenario's routing target: This keeps the initial routing more targeted.
    • Start all conversations on same routing target: This overrides the scenario's setup. It's useful for end-to-end testing or training, where conversations should start on an initial routing target. If you select this, you must choose the routing target, which is specific to the selected contact center.
  16. Total conversations: Specify the total number of conversations in the simulation.

  17. Max. concurrent conversations: Specify the maximum number of concurrent conversations overall in the simulation, not per agent. The upper limit is determined by the Total conversations setting or the predefined system maximum, whichever is lower.

    If the number of unique identities available across the selected scenarios is smaller than the value that you set for Max. concurrent conversations, then actual parallel execution will be limited to the number of available identities, since each simultaneous conversation needs its own identity. Unlike the identity dialog back in step 11, this appears as an inline note and doesn't block you from running the simulation. Add more identities to your scenarios, or lower your concurrency expectation to match available identities.

    To assign one concurrent conversation to every active agent, set this value to match your active agent count.

  18. Max. conversation turns: Specify the maximum number of turns in each conversation. If this limit is reached, the conversation is closed by the system.

    A turn is a series of responses by the same participant; it ends when the participant changes.

  19. Tags: Optionally associate tags with the simulation. Tags facilitate rapid filtering and organization, allowing you to quickly identify simulations based on specific technical topics, difficulty levels, teams or departments, and more.
  20. Click one of the following:
    • Save as draft: You and others can see and edit the draft simulation, but no one can run it. Keep a simulation in draft state until its configuration is complete.
    • Save & publish: You and others can see, edit, and run the published simulation. Publishing can't be undone.

Create a simulation manually

This process is similar to the AI-assisted process except that the system doesn't use AI to generate any scenarios for you, or to preselect any of your existing personas.

Learn about best practices.

  1. On the left navigation bar, click Simulation workshop.
  2. On the Simulations tab, click Add.
  3. In the lower-left corner, click Skip and create manually.
  4. Use the fields provided to configure the simulation. For detailed information on each field, see the section above on the AI-assisted process.

Edit a simulation

Learn about best practices.

  1. On the left navigation bar, click Simulation workshop.
  2. On the Simulations tab, select the simulation, and click the Edit button that appears.
  3. Make the changes.
  4. Do one of the following:
    • If the simulation is in Draft state, click Save as draft to keep it in Draft state, or click Save & publish to make it runnable by you or others. Publishing can't be undone.
    • If the simulation is already in Published state, click Save. This is your only option.

Publish a simulation

Publishing a simulation is the process of moving it out of Draft state, so you can run the simulation. Do this once you've finished making changes and are ready to start testing AI agents and training live agents using the configuration. Publishing is a process that can't be undone.

  1. On the left navigation bar, click Simulation workshop.
  2. On the Simulations tab, select the simulation, and click Publish.
  3. Click Publish again to confirm the action.

Archive a simulation

Archiving a simulation makes it uneditable and unrunnable. You can still display it using the Filter options. You can also unarchive a simulation at any time.

  1. On the left navigation bar, click Simulation workshop.
  2. On the Simulations tab, select the simulation that you want to archive, and click Archive on the right.
  3. Also archive completed runs and reports: Check this if you also want to archive the associated runs and their reports.
  4. Click Archive again to confirm the action.

Unarchive a simulation

Unarchiving a simulation makes it editable and runnable again. However, it doesn't unarchive associated simulation runs and their reports; you must do this separately on the Reports page. (Tip: Use the Filter options to display items based on their state.)

  1. On the left navigation bar, click Simulation workshop.
  2. On the Simulations tab, use the Filter options in the upper-right corner to show the archived simulations.
  3. Select the simulation that you want to unarchive, and click Unarchive on the right.

Start a run immediately

Before running a simulation, ensure it is published, i.e., no longer in draft state.

Learn about best practices.

  1. On the left navigation bar, click Simulation workshop.

  2. On the Simulations tab, select the simulation, and click Run.

    After a simulation is completed, it takes a few minutes for the assessment to appear.

Stop a run

You can stop a simulation run that's in progress for any reason. For example, you might realize there's a mistake in the configuration, and you don't want the simulation to continue. Stopping a run closes all in-progress conversations in the agent workspace.

A manual stop doesn't mean losing visibility into what happened. You get actionable, multi-level insights from every run, not just the ones that were completed end-to-end.

  • Transcripts: When you stop a run, you have access to the conversation transcripts for all conversations that were completed before the run was stopped. However, transcripts for any conversations that were in progress at the time of the stop are discarded and are not available.
  • Assessments: Agent assessments and the simulation assessment reflect only the completed conversations, so the results stay reliable when the simulation doesn't finish as planned.

To stop a run

  1. On the left navigation bar, click Simulation workshop.
  2. Click the Running tab.
  3. Identify the simulation run that you want to stop, and click Stop on the right.
  4. Click Stop again to confirm the action.