The Before State
Five hours of writing for five minutes of reading
Priya managed FP&A for a B2B software company in Singapore. Every month, she spent four to five hours writing management accounts commentary — the narrative section that explains what the numbers mean, what drove the variances, and what leadership should pay attention to. Her board pack executive summary took another three hours.
The maddening part: the executives read it for about five minutes and moved on. She was spending a working day every month producing content that was consumed in a fraction of the time.
She'd heard about other finance professionals using ChatGPT but had real concerns. "I was worried about putting client data into a third-party tool. I'd read the horror stories. And I genuinely didn't understand how AI could help with something as specific as management accounts — wouldn't it just hallucinate numbers?"
The Real Concern
"What if I put confidential data into ChatGPT?"
This is the most common concern finance professionals have — and it's a legitimate one. The Finance kit addressed it directly: AI should never receive raw financial data with client identifiers, MNPI, or anything covered by an NDA. The correct approach is to describe the trend in plain terms, then ask AI to write the commentary. The AI never sees actual figures — only your description of what they mean.
What She Did
One technique changed everything
After reading the Finance kit core guide on a Saturday, Priya understood the fundamental rule: AI writes the words. You control the numbers. She would describe a trend to ChatGPT — never the underlying figure — and ask for commentary language.
Instead of writing "Revenue was £4.2m against £3.7m budget" into ChatGPT, she'd write: "Revenue came in 13% ahead of budget, driven by enterprise — write three versions of a concise management commentary sentence, each with a different tone: factual, confident, and forward-looking."
The output was immediately usable as a first draft. She reviewed it for accuracy, tweaked the tone, added the actual numbers, and moved on. A commentary section that previously took 45 minutes now took 12.
She also used Claude for longer work — giving it the previous month's board pack and asking it to suggest structural improvements, tighten the exec summary language, and flag any inconsistencies in tone. Claude's 200,000-token context window meant it could hold the entire document in one conversation.
The After State
30 days later
Management accounts commentary: 5 hours → 90 minutes
Board pack executive summary: 3 hours → 45 minutes
Variance analysis narrative: drafted in one pass instead of three revisions
Stakeholder update emails: 40 minutes → 10 minutes
Macroeconomic context research: Perplexity AI replaces 2 hours of manual reading
Zero sensitive data entered into any AI tool — full compliance confidence
"Once I understood that AI is for language, not calculations, I stopped being nervous about it. I never put an actual number into ChatGPT. I describe what the number means — and it writes the sentence."