Automating Plan-vs-Actual Variance Commentary
Variance commentary is the writing FP&A does most and enjoys least. Every month you stare at a plan-vs-actual table and turn it into prose: what beat plan, what missed, why, and what it means going forward. The numbers are simple subtraction; the bottleneck is the writing and the judgment about which variances matter. This is exactly the kind of wordy, structured task AI handles well. This lesson builds a reliable system for AI-assisted variance commentary that stays accurate and forward-looking.
What You'll Learn
- How to produce plan-vs-actual commentary that explains rather than just restates
- A prompt template that enforces materiality thresholds and tone
- How to keep commentary forward-looking, not a rehash of the close
- How to build a reusable commentary engine for every monthly cycle
This lesson is about forward-looking management and board commentary on plan-vs-actual performance. It is not about the mechanics of closing the books, which is a separate discipline. Here, the close is already done; your job is to explain the result and what it means for the plan.
The difference between restating and explaining
Weak commentary restates the table in words: "Revenue was 320 against a plan of 350, a 30 unfavorable variance." The reader already has the number. Strong commentary explains and looks forward: "Revenue missed plan by 9%, driven entirely by two enterprise deals slipping into next quarter. The pipeline is intact, so we expect to recover most of this in Q3, but we are lowering the full-year forecast by 2% to reflect the timing risk."
Your prompts should always push AI toward the second style. The instruction that does this is simple: ask for the driver and the implication, not just the variance.
A reusable commentary prompt
Here is a template you can adapt every month. The numbers are calculated in your spreadsheet first; AI only writes.
You are my FP&A analyst writing plan-vs-actual commentary for the
monthly management report. Below is a table with budget, actual, and
variance already calculated by department.
Rules:
- Only comment on variances greater than 5% AND greater than $50k.
- For each, state the driver and the forward-looking implication,
not just the number.
- Group into favorable and unfavorable.
- Maximum two sentences per item, neutral and factual tone.
- If you do not know the driver, write "[driver to confirm]" rather
than guessing.
[paste plan-vs-actual table]
Three things make this work. The materiality thresholds stop the AI from commenting on noise. The "driver and implication" rule forces explanation. The "[driver to confirm]" instruction stops the model from inventing a reason, which is the single biggest risk in AI-written commentary.
Feeding the AI real drivers
The AI does not know why a number moved unless you tell it. The best commentary comes from giving it the drivers you already know and letting it write them up cleanly:
Here is the variance table plus my notes on what drove the big items:
- Marketing under budget: campaign delayed to Q4
- Services revenue over plan: one large implementation pulled forward
- Headcount under plan: two open roles unfilled
Write the commentary using these drivers. Where I have not given a
driver for a material variance, flag it for me to investigate rather
than inventing one.
This hybrid is the sweet spot. You supply the business truth; AI supplies fast, consistent, well-structured prose. You are never trusting the model to know your business, only to write about it well.
Keep it forward-looking
Management and boards care less about what happened and more about what it means. Add a closing instruction that pushes every comment toward the future:
After the variance bullets, add a three-sentence "forward view":
which variances are timing versus permanent, what it means for the
full-year forecast, and the one action we are taking in response.
This is what elevates commentary from a backward-looking report into a decision tool. Timing variances reverse; permanent ones change the forecast. Spelling that out is what leadership actually wants, and it is the natural bridge into your reforecast and your board deck.
Build a commentary engine
Once your prompt works, turn it into a repeatable asset. Save the full template, including your materiality thresholds and tone rules, as a reusable prompt or a custom GPT or Claude Project. Each month you paste the new table and your driver notes, and the structure stays identical. Consistent structure month over month is exactly what makes a management report easy to read, and it means a new analyst can produce on-house-style commentary from day one.
A simple monthly flow:
- Calculate variances in Excel against the plan.
- Add your known drivers as short notes.
- Run your saved commentary prompt.
- Edit for accuracy, confirm any flagged unknowns, and check no sensitive data leaked into the output.
- Drop the forward view into your reforecast and deck.
Accuracy guardrails
Two rules keep this safe. First, never let the AI compute the variances; that math is yours. Second, never ship a driver the AI invented. The "[driver to confirm]" and "flag rather than invent" instructions exist precisely so the model raises its hand instead of fabricating a plausible-sounding reason. A confident wrong explanation in a board report is worse than a blank, because someone will act on it.
Key Takeaways
- Strong variance commentary explains the driver and the forward implication, not just the number.
- Use materiality thresholds in your prompt so AI comments on what matters and ignores noise.
- Feed AI the drivers you know and instruct it to flag, not invent, any it does not have.
- Save your commentary prompt as a reusable engine and always keep the variance math in your spreadsheet.
اختبار قصير
التالي · الدرس 7 من 9
Scenario Modeling and Headcount Planning
قراءة 4 دقيقة
أنهيت 6 من 9 درسًا

اسأل Byte عن هذا الدرس
محادثة خاصة حول هذا الدرس. قرأ Byte الدرس كاملًا، ويجيب خلال ثوانٍ، ويُظهر القسم الذي تأتي منه كل إجابة. يمكنك السؤال بأي لغة.
الأسئلة والأجوبة
أسئلة عامة حول هذا الدرس. يجيب AI Tutor فورًا بلغتك، ويراجع فريق FreeAcademy الإجابات، ويمكن لكل المتعلمين قراءتها.

