> For the complete documentation index, see [llms.txt](https://docs.rootcause.ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.rootcause.ai/user-guide/simulations.md).

# Run Simulations

With a trained Digital Twin, you can ask questions that standard analytics cannot answer: not just what happened, but why — and what would happen if you acted differently. Simulations are how you do that.

Each simulation type is designed for a different kind of question. You can describe what you want in plain language and let RootCause generate the configuration, or choose a simulation type directly and configure it yourself.

For the technical background, see [Digital Twin & Simulations](/core-technologies/digital-twin-and-simulations.md).

***

## Starting a simulation

From your Digital Twin, click the **Simulations** row (or **Run Simulation** in the toolbar), then **New Simulation**.

<figure><img src="https://1662811113-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FBXg3gZLR0e2Q8SzeQmql%2Fuploads%2Fgit-blob-0d7f5fe726cf02999a9c1e6ff11006770b265946%2Frc-new-simulation-launcher.png?alt=media" alt="New Simulation screen showing the Generate from Query field, the Quick Start section with a Generate suggestions button, and the first simulation type cards, with the run list on the right"><figcaption><p>The New Simulation panel. Type a plain-language question to auto-generate a configuration, generate Quick Start suggestions, or choose a simulation type directly.</p></figcaption></figure>

**Generate from Query** — type a question ("What happens to churn if we give away free tech support?") and click **Generate Scenario**. RootCause maps your question to a simulation type and pre-fills the configuration. Review and adjust before running.

**Quick Start** — click **Generate suggestions** and RootCause proposes ready-to-run simulation ideas based on the relationships in the model. Pick one to load it as a pre-filled configuration.

***

## Simulation types

<figure><img src="https://1662811113-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FBXg3gZLR0e2Q8SzeQmql%2Fuploads%2Fgit-blob-6e98a1df61590bc59405c16e297b2bd74c2adfd9%2Frc-simulation-types.png?alt=media" alt="The OR CHOOSE A TYPE grid with the seven static simulation type cards: Prediction, Intervention, Optimization, Best Action, Explanation, Root Cause Analysis, and Anomaly Scan &#x26; Diagnosis"><figcaption><p>The seven simulation types available on a static twin.</p></figcaption></figure>

| Type                                                                                          | Question it answers                                                 |
| --------------------------------------------------------------------------------------------- | ------------------------------------------------------------------- |
| [**Prediction**](/more-details/digital-twin/simulation-types/prediction.md)                   | What outcome is most likely for a specific case?                    |
| [**Intervention**](/more-details/digital-twin/simulation-types/intervention.md)               | What happens if we change variable X?                               |
| [**Optimization**](/more-details/digital-twin/simulation-types/optimization.md)               | What combination of inputs best achieves our goal?                  |
| [**Best Action**](/more-details/digital-twin/simulation-types/best-action.md)                 | What is the minimum change needed to reach a target outcome?        |
| [**Explanation**](/more-details/digital-twin/simulation-types/explanation.md)                 | What drives this outcome, and how much does each driver contribute? |
| [**Root Cause Analysis**](/more-details/digital-twin/simulation-types/root-cause-analysis.md) | What caused this specific observed outcome?                         |
| [**Anomaly Scan & Diagnosis**](/more-details/digital-twin/simulation-types/anomaly-scan.md)   | Which variables are behaving anomalously, and why?                  |
| **Forecast** *(temporal twins only)*                                                          | How will this variable evolve over time?                            |
| **Temporal Intervention** *(temporal twins only)*                                             | How does a time-bounded intervention affect outcomes over time?     |
| **Causal Health Monitor** *(temporal twins only)*                                             | What emerging issues should I be alerted to before they escalate?   |

On temporal twins the picker swaps in a time-aware family: a counterpart to every static type except Prediction, whose place **Forecast** takes, plus the temporal-only **Causal Health Monitor**. See [Temporal Simulations](/more-details/digital-twin/temporal-simulations.md).

Panel (multi-environment) twins get per-environment **Panel** variants of the available types. Each runs the analysis across the twin's environments, so you can see where an effect holds and where it differs.

***

## Intervention

Tests "what if" scenarios. Set a change to one or more variables; the simulation propagates that change through the causal graph and shows the effect on your outcomes.

**How to run:**

1. Select **Intervention**
2. Add interventions — choose a variable, pick the change type, and set the new value. The options depend on the variable: a numeric one offers **Target value**, **Relative change** and, on a sweep-capable scenario, **Range (sweep)**; a category or boolean one offers **Target value**, **Adjust probability** and **Set probability**
3. Optionally add conditions ("only for premium customers") to control which records the change applies to
4. Define metrics to measure — built in natural language, a visual builder, or SQL (see the [Intervention](/more-details/digital-twin/simulation-types/intervention.md) page for the metric builder)
5. Click **Run Simulation**

**Results:** Side-by-side baseline vs. intervention comparison with confidence intervals and an effect breakdown by causal path. Full walkthrough: [Intervention](/more-details/digital-twin/simulation-types/intervention.md).

***

## Optimization

Finds the best combination of inputs to maximize or minimize an objective, given constraints you define.

**How to run:**

1. Select **Optimization**
2. Set the objective — variable to optimize, direction (maximize/minimize), measurement
3. Define decision variables (what the optimizer can change)
4. Set constraints (limits that must be respected)
5. Click **Run Simulation**

**Results:** Recommended values for each decision variable, expected outcome at the optimum, and trade-off analysis if you have multiple objectives.

***

## Best Action

Finds the minimum change to a specific case that would flip the predicted outcome. Useful for individual-level decisions: what is the smallest intervention that would prevent this customer from churning?

**How to run:**

1. Select **Best Action**
2. Set the target outcomes you want to achieve, and for a categorical target how sure the model has to be before a record counts as having reached it
3. Choose who to find actions for: the records below the line in a completed Prediction of that target, or records you supply yourself
4. Tick the variables the solver may change, and narrow the values those variables may take
5. Set a maximum number of changes to keep recommendations practical, and choose how each record's plan is picked
6. Click **Run Simulation**

**Results:** Each record's status against the target, one recommended plan for every record that needs a change, the other options behind it, and the changes that get the most records there.

***

## Explanation

Identifies the drivers of an outcome and quantifies how much each contributes. Three modes:

* **Discovery** — "What influences outcome B?" Finds all causes of a specific variable.
* **Directional** — "How does A affect B?" Traces the specific causal path between two variables.
* **Impact** — "What does A affect?" Finds all downstream effects of a specific variable.

<figure><img src="https://1662811113-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FBXg3gZLR0e2Q8SzeQmql%2Fuploads%2Fgit-blob-393332bc8581a5b8c8be752195b30b31775ffa61%2Frc-simulation-key-drivers.png?alt=media" alt="A completed Explanation run for the question What are the main drivers of customer churn: the Key Driver Analysis bar chart of relative causal contribution to Churn, with Contract highest, and the ranked driver table beneath it with contribution, range, and confidence per driver"><figcaption><p>A completed Explanation. The Key Driver Analysis ranks each variable's causal contribution to churn: Contract leads at 44.4%, then InternetService (25.4%) and OnlineSecurity (17.0%), each with a range and a confidence rating.</p></figcaption></figure>

**How to run:**

1. Select **Explanation**
2. Choose the mode
3. Select source and/or target variables
4. Optionally add segment filters
5. Click **Run Simulation**

**Results:** Causal paths with contribution weights, ranked driver table, and segmented breakdowns if requested.

***

## Prediction

Generates a predicted outcome for a specific case, with uncertainty estimates.

**How to run:**

1. Select **Prediction**
2. Select the target variables to predict
3. Enter input data for the remaining driver variables — the columns offered are filtered to those with a causal path to your targets
4. Set the confidence level
5. Click **Run Simulation**

**Results:** Most likely outcome per target variable, confidence intervals, and full probability distribution. A categorical target also opens with **Who's at risk**, which ranks the records by their chance of a value you pick and hands the ones below the line to Best Action.

***

## Root Cause Analysis

Traces a specific observed outcome backward through the causal graph to identify its underlying causes. Different from Explanation, which identifies general drivers — Root Cause Analysis focuses on why a particular outcome occurred. Full walkthrough: [Root Cause Analysis](/more-details/digital-twin/simulation-types/root-cause-analysis.md).

***

## Anomaly Scan & Diagnosis

Scans all variables for anomalous behavior and uses the causal graph to diagnose which upstream variables are responsible. Useful for monitoring and incident investigation. Full walkthrough: [Anomaly Scan & Diagnosis](/more-details/digital-twin/simulation-types/anomaly-scan.md).

***

## Forecast *(temporal twins only)*

Projects variables forward in time using causal relationships and temporal patterns.

**How to run:**

1. Select **Forecast**
2. Select target variables
3. Set the forecast horizon (number of periods ahead)
4. Set confidence level for uncertainty bands
5. Click **Run Simulation**

**Results:** Time series of projected values with widening confidence bands.

***

## Temporal Intervention *(temporal twins only)*

Scripts an intervention that happens within a specific time window, and shows how effects build, peak, and decay over time.

***

## Causal Health Monitor *(temporal twins only)*

Watches a time series — either observed data you supply or the twin's own forecast — and scores every variable at every timestep against the model's expectations. The result is a health timeline with per-variable alerts, a causal ordering of which alerts drove which, and (optionally) an automatic Root Cause Analysis for what it flags. Use it to catch emerging issues before they escalate.

***

## Natural language queries

For any simulation type, you can describe the scenario in plain language and let RootCause generate the configuration. Click **Generate Scenario**, review the interpretation, adjust if needed, and run.

***

## Reading results

All simulations include confidence intervals — wider intervals mean more uncertainty. Intervention and optimization results always compare against a baseline, showing the marginal effect of your action rather than the absolute outcome.

Results can be exported as a report, and the report to PDF or saved for reference. Saved simulations appear in your Digital Twin's simulations list and can be re-run or included in Reports.

***

## Finding past simulations

The **Simulations** list keeps every run. To find one:

* Filter by outcome (**All**, **Succeeded**, **Failed**), by simulation **Type**, by **Variable**, or by time range.
* The **Ran by** toggle starts on **All** and lists both; narrow it to runs started by **People** or by **Apps** (see [Applications](/more-details/applications.md)).
* When a search or a filter is holding runs back, the header reads `1 of 2 · 1 hidden`; **Show all** clears the search and every filter.
* Tick the checkbox on one or more runs to select them for report export.

***

## Next step

Simulations produce findings. The next step is to turn those findings into a document you can share.

Next step: [Step 7: Produce Reports](/user-guide/reports.md)