Generative AI for Finance: Commentary, Research, and Policy Knowledge

An FP&A analyst uses a commercial AI tool to draft the monthly variance commentary for the board pack. The AI produces fluent, professional-sounding paragraphs that explain why revenue missed by $2M. The problem: the AI attributed the miss to 'seasonal demand softening' because that's a common pattern in its training data. The actual reason was a delayed contract signing that pushed $3M in recognized revenue to Q2 — which means the company is actually $1M ahead of plan on an adjusted basis. The AI wrote a confident explanation of a problem that doesn't exist. The CFO catches it during review. A less experienced CFO might not have. Generic AI is dangerous in finance because it generates plausible-sounding explanations that may be completely wrong for this company's specific situation.

Generative AI that works in finance requires grounded retrieval against the company's actual financial data and context. The chart of accounts structure. The accounting policies. The variance history for this specific account and business unit. The known items that explain this period's variances (the delayed contract, the one-time charge, the FX impact). With these grounded, the AI drafts commentary that's factually correct and specific to the company. Without them, it generates plausible fiction. Done with this discipline, generative AI saves the FP&A team hours of commentary writing per close. Done casually, it creates errors that undermine trust in the finance function.