At month-end, calculating is often finished faster than writing. Explaining variances, formulating key messages, putting together a report for management or the supervisory board – that takes hours. AI assistants can turn this into a well-structured first draft. The trick is that the numbers come from Excel and the AI only writes the text.
The ground rule
The workflow in five steps
- Prepare the analysis: a compact table with actual, plan, prior year and variance per line. Variances already calculated in Excel, not by the AI.
- Add background knowledge: what you know about the causes – keywords are enough (“steel price +8%, large order moved to December”).
- Prompt with a clear structure: audience, length, outline, tone, materiality threshold.
- Check the draft: compare every number in the text against Excel, check causes for plausibility.
- Transfer to Word: paste into the report template or work directly with an assistant integrated into Word (e.g. Copilot in Microsoft 365).
A prompt template
Variance = Actual − Plan | Var. % = Variance ÷ Plan| A | B | C | D | E | |
|---|---|---|---|---|---|
| 1 | Line | Actual | Plan | Var. | Var. % |
| 2 | Revenue | 4,210 | 4,080 | +130 | +3.2% |
| 3 | Material costs | −1,905 | −1,780 | −125 | −7.0% |
| 4 | Personnel costs | −1,240 | −1,250 | +10 | +0.8% |
| 5 | EBIT | 412 | 409 | +3 | +0.7% |
A typical AI draft then reads something like: “Material costs of €1,905k are €125k (7.0%) above plan. The cause is the steel price, which has risen by 8% since April. [clarify cause: volume effect?] From December the price adjustment clause in two key-account contracts applies.” – The placeholders show you exactly where you still need to follow up.
How to recognise a good AI draft
- Every number appears exactly as in your Excel table.
- The key message fits into three sentences and would be understood without tables.
- No generic phrases like “seasonal” without justification.
- Open points are marked rather than invented.
Further uses
- Multilingual reports: translate the finished commentary for a foreign parent company.
- Different audiences: turn one report into a short version for the board and a detailed version for division heads.
- Presentations: derive slide bullet points from the Word report.
- Consistency check: have the AI check whether text and tables contradict each other.
Common mistakes
- Letting the AI calculate the variances itself.
- Pasting entire raw data exports instead of a condensed table.
- Sending the draft unchecked – responsibility for the report stays with controlling.
Conclusion
AI is an excellent writer for reporting – as long as it doesn't calculate. With a checked Excel table, clear rules in the prompt and a careful final check you save many hours at month-end. More on building good reports in Monthly reporting for management.
Frequently asked questions
Can AI write my whole monthly report?
It can deliver a very good draft. Numbers, causes and conclusions, however, have to be checked by you – responsibility for the report stays with controlling.
How do I prevent invented numbers?
Provide a fully calculated table, forbid own calculations in the prompt and have missing causes marked as placeholders. Then compare every number in the text with Excel.
Which AI tools are suitable?
General AI assistants such as Microsoft Copilot, ChatGPT, Claude or Gemini, as well as assistants integrated into Word and Excel. What matters is that the tool is approved by your company for the data.