Should you use AI in financial modelling?

𝗜 𝗳𝗮𝗰𝗲𝗱 𝘁𝗵𝗲 𝗳𝗲𝗮𝗿 𝗮𝗻𝗱 𝗱𝗼𝘄𝗻𝗹𝗼𝗮𝗱𝗲𝗱 𝗖𝗹𝗮𝘂𝗱𝗲 𝗶𝗻 𝗘𝘅𝗰𝗲𝗹. 

Like everyone else I've been apprehensive about the impact of AI in financial modelling, plus I didn't think LLMs were really good enough, as I had tried CoPilot a few times, and the errors were glaring. But after hearing about a step change in performance in Claude on a podcast, I finally decided to find out for myself. 

In short, it's very impressive.

I could see myself using it for complex data analysis, building difficult formulas and checking some calculations. And yes, it can build a three-statement model quickly. But for the work I do - building investor-facing models for Founders, CFOs and businesses raising capital or making critical decisions - I'd still rather build it from scratch, because the value I bring isn't the spreadsheet, it's the understanding behind it:

- How the business actually scales

- Where the errors are hiding

- Which levers really drive valuation

- What the CEO or an Investor will challenge in the room 

I'd also have concerns about data confidentiality before putting anything sensitive near it. 

So what does this mean for financial modelling? 

For simple models and templates it's a legitimate time-saver. For complex, investor-facing work, you need genuine understanding behind the build. And if you're an analyst early in your career, I would still recommend learning the fundamentals first. Understand 3-statement modelling and the underlying accounting. Then use AI as a tool, not a shortcut. 

Because the questions that determine model quality aren't spreadsheet questions. They're strategic ones. 

- What are the key drivers of the business?

- Where are the sensitivities that move valuation?

- What story does this model tell?

- Will you/it survive scrutiny in an investor meeting? 

That takes an understanding of the model, the business, judgement and commercial awareness. Pattern recognition built over years of seeing what investors actually care about. 

AI is extraordinary and a useful tool. It's not a replacement for understanding.