Background
A US-based company was producing its monthly management accounts through a manually maintained Excel model. The core challenge was a format mismatch: ERP reports came out in a raw GL structure that didn't align with how management wanted to see the numbers. Every month, a finance team member would spend one to two days reformatting the data, applying formulas, manually mapping vendors, classes, and GL accounts to the management reporting structure, calculating cash-adjusted EBITDA, and writing an analytical review to assess performance against prior periods and budget.
The process worked, but it was slow, dependent on one person's knowledge of the model, and left little time for actual analysis.
What Was Built
Using Claude, the team first did the foundational mapping work: every vendor, class, and GL account in the ERP was mapped to its corresponding line in the management reporting format. This mapping became the persistent logic that Claude could apply automatically each month.
From there, the monthly process became simple. Download a GL dump from the ERP, pass it to Claude, and receive a fully formatted FP&A report — including the management reporting structure, cash-adjusted EBITDA calculation, variance analysis, and an analytical review commenting on performance.
Where Human Judgment Still Mattered
The process didn't eliminate human involvement — it redirected it. Each month, a finance team member would review the output, map any new vendors or GL accounts that had appeared since the last run, sense-check the analytical review against their knowledge of what had actually happened in the business, and add context that no model could infer — a one-time expense, a deal that closed late, a cost that was prepaid. This human layer made the output meaningful rather than just mechanical.
The Result
Monthly FP&A reporting went from one to two days down to one to two hours. The time saved wasn't just administrative — it shifted the finance team's focus from data processing to analysis and business partnering. The analytical review, previously a rushed summary at the end of a long manual process, became a more considered piece of work because the numbers were ready in minutes rather than days.
Key Takeaway
AI didn't replace the finance function — it removed the low-value mechanical work that was consuming most of the time. The human still owns the judgment calls: what's new, what's unusual, what needs explaining. But with the heavy lifting automated, those judgment calls now happen in a fraction of the time, and the output is better for it.