February 2026 marked a notable step change – it was the first time AI tools really had the capability to build strong financial models, explains Ian Schnoor, executive director of the Financial Modeling Institute, the world’s only financial modelling accreditation organisation. “Not perfect, but very good. Before that, AI tools were lousy,” he says.

Before this shift, AI tended to perform best when given a structured framework and clear guidance. It was better at completing templates than constructing a model from scratch. However, newer systems are increasingly able to produce a workable model from a broad prompt, even if the quality varies, as Schnoor discussed on a webinar in May 2026. 

“There’s no magic to getting AI to build your model,” says Schnoor. “You can say ‘look at the historicals and build me a model.’ But that doesn’t mean it will be a good one.”

Building a model with AI is still an iterative process. The user has to direct, refine and challenge what the model produces; that only works if they understand modelling fundamentals. “Building a model with AI is like building it yourself but faster,” says Schnoor. “You’re telling it what you would do anyway – ‘let’s build revenue this way, no I don’t like that, let’s modify.’ You will not be able to get it to build a good financial model unless you already know how to.”

Considerations before use

The first consideration before using AI to create a financial model is whether you already understand what a good model should look like. It means knowing the ingredients and structure, including core components such as assumptions, outputs and scenario analysis, before you write any prompt. AI should always be treated as a collaborator, not an authority.

It also means having the confidence to challenge the output. “At the end of the day, a human is going to want to talk to a human. They’re still going to ask you questions,” says Schnoor. 

Claude currently appears to be one of the leading options for financial modelling work, with other major large language models (LLMs) improving quickly. Schnoor says, “I believe all the LLMs will have strong modelling skills in time. But they will still make mistakes that need to be checked.” 

The most effective way to review an AI-generated model is to treat it as a draft that must be checked.

Checklist for reviewing an AI-generated financial model

• Check that all core sections are present: cover page, summary page, assumptions page, scenario management page and clean schedules. Sometimes, when prompted, models do not include all of these.
• Check for hidden sheets, hidden rows and hidden columns. Ask the AI if it has added any but then confirm manually.
• Check for external links to other files and remove them if they are not intentionally required. You can ask the AI to confirm but you should not rely on it. Schnoor says: “Sometimes I have asked it to check on something and then it gives me a wrong answer. ‘There ARE links to external files.’ ‘You’re so right. How did I miss that?’ You need to check these things. And then double check.”
• Check that inputs and outputs are clearly distinguished, for example, through consistent formatting or colour coding.
• Check for hard-coded numbers inside formulas – “dead numbers” that stop the model updating properly.
• Check that formulas are consistent across months or years. AI is notorious for changing formulas as it goes across a forecast.
• Check for overly long or overly complex formulas in key statements and keep them clean and readable.
• Check that the balance sheet genuinely balances and has not been forced to balance with a plug, where someone has back-solved just to make it work.
• Check that things like debt schedules are complete and that debt balances do not go negative by mistake, including short-term debt used to deal with future cash shortfalls.
• Check that capacity constraints and other operational limits are reflected in the forecast, for example, that revenues do not imply a company can manufacture an unlimited volume.
• Check that depreciation is reasonable and that assets are not over-depreciated.
• Check that major assets and liabilities are not becoming negative without a valid reason.
• Check that the model includes internal checks and that those checks work across every forecast period, not just the first year. 

Other common errors include oversimplifying and inserting dead numbers where a formula should be, building inflexible formulas that are hard to update, creating unnecessarily long formulas, stating incorrect information confidently and generating different answers to the same request on different attempts. One useful tactic is to ask the AI the same thing again in a slightly different way and comparing the outputs.

Used well, AI can speed up financial modelling and free you up to spend more time on presentation, analysis and communication. But the core requirement remains the same, says Schnoor: “The key to future success for finance professionals is that you still need to understand all the ingredients and pieces and tools used. Having strong modelling skills is important.”