Flux and Variance Analysis with AI
Flux analysis — explaining the period-over-period movement in each P&L and balance sheet line — is the single most repeated writing task in financial reporting. Every month, every quarter, every audit cycle, somebody at your firm is writing "revenue increased $X due to Y". This lesson teaches you to use AI as your first-draft writer for flux and variance, with the prompts and templates that produce audit-ready output.
What You'll Learn
- The structural difference between flux and variance and when to use each
- A master flux prompt that generates a defensible P&L explanation in 60 seconds
- The variance-bridge prompt that translates numbers into board-ready language
- How to handle non-recurring items, FX, and acquisitions cleanly
Flux vs Variance: A Working Definition
Flux analysis is period-over-period: this month vs last month, this quarter vs prior quarter. The goal is to explain what changed.
Variance analysis is actual vs budget or actual vs prior year. The goal is to explain why we did better or worse than expected.
Both have the same structure: a number, a comparison number, the difference, and the reason. AI is excellent at writing the reason once you give it the numbers and one or two drivers.
The mistake new users make is to ask AI to "explain the variance" without giving it the drivers. AI will hallucinate causes — it does not know your business. Give it the drivers and it will write the explanation beautifully.
The Master Flux Prompt
This is the prompt that lives at the top of your prompt library. Open ChatGPT Business or Claude Team and paste:
"Act as a Finance Manager writing flux analysis for management accounts. For each P&L line below, write a 2 to 3 sentence flux explanation versus prior month. Use the driver provided. Tone: factual, no marketing language, no speculation. Quantify each driver in dollars where given.
Revenue: current $4,820,000 (prior $4,510,000, change +$310,000). Driver: 4 new enterprise customers signed in the month, ARR impact $210,000 in-month; usage uplift from existing accounts $100,000.
Cost of goods sold: current $1,930,000 (prior $1,720,000, change +$210,000). Driver: cloud hosting up $60,000 from increased usage; customer success headcount up 2 FTE $80,000; partner commission accrual increase $70,000.
Gross profit: current $2,890,000 (prior $2,790,000, change +$100,000). Driver: derived; gross margin compression from 61.9% to 60.0% driven by hosting and headcount investments.
Operating expenses: current $2,150,000 (prior $2,090,000, change +$60,000). Driver: marketing spend timing on Q2 campaign $80,000; offset by lower professional fees $(30,000).
EBITDA: current $740,000 (prior $700,000, change +$40,000). Driver: derived from above.
Output as a list of 5 bullets, one per line item, ready to paste into the management accounts narrative."
The result is a list of clean, audit-defensible flux comments in the voice of a Finance Manager. You scan, edit any line that feels off, and paste into your management pack.
Why You Always Provide the Driver
If you skip the driver, AI will invent something plausible-sounding. That is the single biggest risk of AI-assisted flux. The driver is the truth from your team — usually a one-line answer from the budget owner or department head. AI's job is to translate that truth into clean prose.
If you do not have the driver, your flux analysis is not ready to write. Go back to the source. Ask the department head. Then come back to AI.
The Variance-Bridge Prompt
For variance to budget or prior year, the structure is slightly different because you usually want a bridge — a stack of drivers that adds up to the total variance.
"Act as a Finance Manager writing a variance bridge versus budget for [month] [year]. Total revenue variance: $[X] favourable / unfavourable.
Drivers:
- Volume: $[A] (units sold higher/lower than budget)
- Price: $[B] (average selling price higher/lower)
- Mix: $[C] (more/less of high-margin product Y)
- FX: $[D] (translation impact)
- One-off: $[E] (describe)
Write a 200-word variance bridge narrative. Open with the headline variance and direction. Then walk through each driver in order of size. Close with whether year-to-date variance is being recovered or worsening. Tone: confident, specific, no hedging."
This produces a board-ready paragraph. Test it on a real variance from last month and you will see why this prompt is worth saving permanently.
Handling Non-Recurring Items Cleanly
The cleanest narrative practice is to separate underlying performance from one-off items. AI follows this pattern naturally if you tell it to.
Prompt addition:
"Always separate underlying performance from one-off items. If a line variance includes a non-recurring impact, state the underlying movement first, then the one-off, then the reported number. Example: 'Revenue grew $200,000 underlying; reported growth was $350,000 including a $150,000 one-off contract milestone.'"
Add this clause to your master flux prompt. It instantly elevates the quality of the output to match what a senior FP&A manager would write.
FX, Acquisitions, and Other Comparability Killers
Two specific comparability issues come up every month for international and acquisitive companies.
FX: Always state movements in constant currency. Add this to your prompt:
"Where the company reports across multiple currencies, separately state the FX translation impact and the underlying constant-currency movement. Example: 'Revenue grew 12% reported, 9% in constant currency, with 3 percentage points of growth attributable to favourable FX translation.'"
Acquisitions: Always state organic vs inorganic.
"Where the period includes acquired revenue not in the prior period, state organic and inorganic contributions separately. Example: 'Revenue grew $2.1m, of which $1.3m is organic and $0.8m is from the Acme acquisition completed in March.'"
These two additions to your flux prompt give you the discipline of a top-quartile finance team without the extra writing time.
A Concrete Time-Saving Example
For a 25-line P&L with full flux commentary at 2 to 3 sentences per line, the manual writing time is around 90 minutes for an experienced finance manager. The AI-assisted version takes 8 to 12 minutes once you have your drivers in hand: 2 minutes pasting drivers into the prompt, 2 minutes for AI to draft, and 5 to 8 minutes editing and tightening.
The compounding gain is bigger than that. AI consistently produces the same structural quality across every line, so your variance commentary has even tone, even depth, and even formatting throughout the pack. That alone makes your CFO's review faster.
Building Your Flux Prompt Library
By the end of your second month using these prompts you should have:
- A master flux prompt with your firm-specific style baked in
- A variance bridge prompt with your driver categories
- Specific prompts for revenue, cost of sales, opex, and balance sheet items
- A separate prompt for cash flow flux (which has different driver categories)
Five prompts. Saved in a single document. Reused every month. This is what compound time savings looks like.
Key Takeaways
- Always give AI the driver — without it, AI will invent plausible-sounding explanations
- Use the master flux prompt for period-over-period, the variance-bridge prompt for actual-versus-budget
- Separate underlying from one-off items in every commentary line
- For multinationals, separate constant-currency from FX translation in every relevant line
- A 25-line P&L flux drops from 90 minutes to 12 minutes with no loss of quality