AI in controlling: variance commentary

First the figures, then their explanation

For the first trial, one cost category is enough: the planned amount, the actual amount and the approved notes of the people responsible. Keep the export date and the plan version. Comparing actuals with the original budget answers a different question from comparing them with a forecast updated in the middle of the month.

Check gaps in the data with the tool the team already uses. Microsoft describes profiling column quality, distribution and values in Power Query. By default it covers the first 1,000 rows; you have to choose analysis of the whole data set separately. So a green preview of a sample says nothing about the quality of the remaining records. Microsoft documentation, checked on 9 October 2026.

What follows from a fictional variance of €12,000?

For the exercise, take a plan of €100,000 and actuals of €112,000. The difference is €12,000, or 12% of the plan. In a note, the cost owner confirmed an extra service for €7,000. They have not yet explained the remaining €5,000. These are made-up figures with no link to any client's result.

The draft commentary should keep that gap: costs exceeded the plan by €12,000, of which €7,000 is assigned to the named service and €5,000 needs an explanation. A sentence about „rising supplier prices” would be a guess until a source confirms that cause.

The controller can ask AI to shorten the description or prepare a question for the owner. Keep the calculation of the difference and the percentage in the spreadsheet, so the reader can reproduce it apart from the text. With a plan of zero, the usual percentage variance formula will not work; how to present such a line needs its own rule.

How do you check the commentary before sending it?

Read the sentences about figures and the sentences about causes separately. For a figure, point to the cell or the row of the report. For a cause, find the approved note, the document or the owner's answer. When the source contains a supposition, the summary should keep its uncertainty.

In the next exercise, change the actual amount and keep the old note. Check whether the prepared commentary reveals the mismatch. Also add a line with no explanation. Judge the number of statements that need correcting and the time it takes to check the whole text, because the generation time alone says little. This guide proposes such a trial; it reports no result of a model test.

When is a workshop enough, and when do you need connected systems?

If the data is ready and the team needs practice in writing and checking commentaries, you can agree Syntalith team training. The process review at Syntalith is free, automations, apps and agents start from €3,000 net, and team training from €1,150 net a day (pricing). We set the programme from the work of the controllers and of the readers of the report. It is a workshop on working with AI. Responsibility for approving financial data stays with your team.

When most of the time goes on collecting files and chasing corrections, you need a separate automation scope. We describe the wider workflow in the guide to financial reporting for the CFO. For the first call, a description of the report, its sources and the place where it usually comes back for corrections is enough.

What it costs

Where to start

We run the workshop as part of team training, from €1,150 net a day. For the free review, a description of one task the team does every week and the tools it uses is enough. You receive the written training programme within 2 working days.