AI Costing Prompts · By task
Draft board variance commentary without fabricated causes
Give an AI model your actual versus budget figures and get clean variance analysis plus a short, factual board commentary. The standout feature is what the model refuses to do: it will not invent a reason for the variance, it tells you the cause is not in the data and lists what it would need to explain it.
In short
Paste in actual and budget for revenue, gross profit, operating expenses and operating profit. The model calculates each variance in value and percentage, marks each as favourable or adverse, finds the real story by comparing gross margin percentages, and drafts a commentary of about 150 words. Where the cause of a variance is not contained in the numbers, it says so plainly and lists the data needed, rather than manufacturing a plausible-sounding explanation.
What the prompt is doing
Good variance commentary separates two things that are often blurred: what happened, which the numbers can tell you, and why it happened, which they usually cannot. The model can compute that gross profit fell while revenue rose, and it can show that the cause is margin compression rather than a top-line problem. What it cannot do honestly is name the reason for that compression, because price, volume, mix and input cost all live outside the four-line summary it was given.
This restraint is the whole value of the prompt for a finance business partner. A board pack that confidently states a wrong cause is worse than one that says "the cause is not yet identified, here is what we need to find it." That honesty is the same principle that underpins linking operational change to financial outcome in strategy execution and reporting it cleanly through a balanced scorecard: measure what you can defend, and be explicit about what you cannot yet explain.
The prompt
You are a finance business partner. Work only from the figures I give you. Do not invent any explanations or numbers. Where a cause is not in the data, say so and list what you would need to explain it. Actual vs budget this quarter (EUR): Revenue: 4,120,000 vs 4,000,000 Gross profit: 1,360,000 vs 1,440,000 Operating expenses: 910,000 vs 880,000 Operating profit: 450,000 vs 560,000 Do the following: 1. Calculate the variance and variance % for each line. 2. Mark each variance favourable or adverse. 3. State the key story (revenue is up but gross profit is down) and compute gross margin % actual vs budget. 4. Draft a short, factual board commentary of no more than 150 words, flagging clearly where a cause is not in the data. 5. List the data you would need to explain the margin drop.
A worked example
| Line | Actual (EUR) | Budget (EUR) | Variance (EUR) | Variance % | Verdict |
|---|---|---|---|---|---|
| Revenue | 4,120,000 | 4,000,000 | +120,000 | +3.0% | Favourable |
| Gross profit | 1,360,000 | 1,440,000 | -80,000 | -5.6% | Adverse |
| Operating expenses | 910,000 | 880,000 | +30,000 | +3.4% | Adverse |
| Operating profit | 450,000 | 560,000 | -110,000 | -19.6% | Adverse |
Gross margin fell from 36.0% budget to 33.0% actual, a drop of 3 points. The margin compression, not the revenue line, drives the shortfall in operating profit.
The model drafted a commentary of about 148 words that states these facts and then says explicitly that the cause of the margin drop is not identifiable from the data provided. It refused to guess. It then listed the data it would need to explain the drop:
- a revenue bridge splitting volume from price
- cost of goods sold broken down by category
- sales mix by margin band
- unit input costs versus budget
- discounting and rebate detail
- operating expense detail by line
That refusal is the point. The AI gave the board a true picture and an honest gap, instead of a confident answer that might have been wrong.
What it costs you to run
The input is around 260 tokens. A full run with the variance table, the margin analysis, the drafted commentary and the data wish-list costs a fraction of a cent on any current model. Running it every reporting cycle is effectively free relative to the time it saves drafting the first pass.
Token figures are approximate and vary by model and tokeniser.
The guardrail that matters
Work only from the figures I give you. Do not invent any explanations or numbers. Where a cause is not in the data, say so and list what you would need to explain it.
This is the most important guardrail in the whole hub. Board commentary is precisely where a fabricated cause does the most damage, because it gets repeated as fact. The instruction turns the model from a confident guesser into an honest analyst that names the gap. See how to stop AI inventing numbers.
When you need the real model
The prompt can tell the board that margin compressed, but it cannot tell them why, because the why lives in the cost detail. Tracing margin down to product, customer and channel, so the cause is in the data next time, is the work behind a real profitability model.