AI meets probabilistic modeling: How to build and audit Monte Carlo simulations with confidence

Why the future of risk analysis is AI-assisted, human-led, and built in Excel

AI tools can now scaffold a Monte Carlo simulation model faster than any analyst working alone. They suggest distribution types, propose input ranges, and explain simulation outputs in plain language—without fatigue and at speed.

They can also introduce errors that look entirely plausible and are hard to detect without the right audit tools.

This is not a reason to avoid AI in the modelling workflow. It is a reason to understand exactly where it helps, where it fails, and what to do about it. The analyst who masters that distinction will build better models, faster, and defend them with more confidence than ever before.

Excel is still the right tool

Before discussing AI, a word about the platform.

Excel remains the most widely used environment for probabilistic risk modelling—and for good reason. It is transparent, auditable, and understood by every stakeholder in the room. When a CFO, a project director, or an investment committee member asks “how did you get that number?”, an Excel-based model can show the answer cell by cell. That auditability is not a legacy limitation. It is a professional asset.

@RISK by Lumivero extends Excel into a full Monte Carlo simulation environment without changing that transparency. Distributions replace point estimates. Thousands of iterations replace a single scenario. Tornado charts, scenario analysis, and correlation matrices sit alongside the model, readable by anyone who needs to understand where the uncertainty comes from. The analyst stays in control of the structure, the assumptions, and the output. Excel stays the medium through which that expertise is communicated.

Adding AI to that environment does not change the platform. It changes the speed at which the analyst can work within it.

The case: An offshore wind investment decision

A developer is approaching Final Investment Decision on a 250MW offshore wind project. CAPEX is estimated at €900 million. The project carries a long-term revenue contract followed by a period of exposure to market electricity prices. The investment committee wants a distribution of NPV outcomes—not a point estimate—and a clear view of what is driving the uncertainty.

This is a common situation in energy and infrastructure project finance. The inputs are numerous. The relationships between them are complex. A single-point estimate of NPV cannot capture the range of outcomes that the investment committee actually needs to understand. A Monte Carlo simulation can—and with @RISK Agent, it can be built and interrogated faster than ever before.

Stage 1: Building with AI

@RISK Agent embeds AI directly into the @RISK Excel environment through Lumivero’s Model Context Protocol layer. The analyst describes the asset and the decision context in plain language. The agent proposes a model structure, suggests appropriate distribution types for each uncertain input, and identifies which variables carry the most uncertainty. The scaffolding that would previously have taken hours to build from scratch is ready to run in a fraction of that time.

This is where AI earns its place in the workflow. Not by replacing the analyst’s judgment, but by removing the mechanical work that precedes it. The analyst reviews what AI proposes, overrides where necessary, and brings domain knowledge to bear on the assumptions that a general-purpose AI cannot know—the specific contract terms, the local market conditions, the asset’s operational history.

The model is built. The distributions are assigned. It is ready to run.

Stage 2: Verifying with @RISK

Running 10,000 iterations, the tornado chart ranks the drivers of NPV variance immediately. In this model, revenue uncertainty dominates, followed by capital cost and operating cost variability. The chart does not just confirm what the analyst expected—it shows the relative weight of each variable and flags where the model’s assumptions deserve the most scrutiny.

This is the audit step. And it is where the analyst’s expertise becomes decisive.

AI builds models efficiently. It does not always build them correctly. In this case, the AI treated two correlated input variables as independent—a common and consequential error that the tornado chart exposes immediately. Correcting that single assumption shifts the P10 NPV materially. The project remains viable at the central estimate, but the downside scenario looks different. That is a different conversation with the investment committee.

No prompt engineering surfaces that finding. The simulation engine does. The tornado chart, the scenario analysis, the correlation matrix—these are the tools that catch what AI gets wrong. They do not slow the workflow down. They make the output defensible.

Stage 3: Interrogating the output

With the model corrected, @RISK Agent’s natural language interface allows the analyst to interrogate the results directly. What drives NPV below zero? How does the upside change under a more favourable revenue scenario? At what point does the project fall below the required return threshold?

The answers come from the simulation data—grounded in the actual model output through the MCP layer, not in the AI’s general knowledge of energy finance. The responses are specific, traceable, and ready to use in a board presentation or investment memorandum.

The analyst is not removed from the process. The analyst is freed to focus on the decisions that require human judgment—because the mechanical work has already been done.

The modeller stays in control

This point deserves to be stated plainly.

AI accelerates. It does not decide. Every distribution in the model reflects a judgment call about how uncertain an input is and what shape that uncertainty takes. Every correlation reflects knowledge of how variables move together in the real world. Every scenario reflects a view about what could go wrong and how bad it could get. Those judgments belong to the analyst. They always will.

What @RISK Agent changes is the cost of iteration. Testing a different distribution assumption, rerunning the simulation, reading the new tornado chart—that cycle now takes a fraction of the time it previously required. The analyst can interrogate more assumptions, stress more scenarios, and arrive at a more robust output in the same amount of time. That is not a threat to professional judgment—it is an amplifier of it.

The workflow in three steps

• Build with AI. Describe the asset and the decision context. Let @RISK Agent propose the structure, the distributions, and the parameter ranges. Review and override where your expertise differs from the AI’s suggestions.
• Verify with @RISK. Run the simulation. Read the tornado chart before you read the NPV number. Audit the correlation matrix. This is where AI is most likely to have made a silent assumption that needs correction.
• Defend with both. Use @RISK Agent’s natural language Q&A to interrogate the corrected model. The answers are grounded in simulation data. The output is traceable, auditable, and ready to defend to any stakeholder.

Excel provides the structure. @RISK provides the simulation engine. AI provides the speed. The analyst provides the judgment that makes the output worth defending.

Build with AI. Verify with @RISK. Defend with both.

Ready to enhance your risk workflow? Download the example offshore wind investment model used in this article and try the workflow yourself—or watch Manuel walk through it in our on-demand webinar.

Note: You'll need @RISK to run the model. If you don’t have it yet, buy @RISK today.