AI variance analysis that writes the commentary and never touches the numbers

Every figure shown here is invented for illustration. Nothing on this page is computed from anyone’s accounts.

In the product this reads as prose, not a table: a short written review of the month, section by section, read beside the dashboard. Every figure in it is a link to the formula and the ledgers behind it.

Asking a chatbot to “analyze the variances” in a pasted trial balance gets you fluent paragraphs with numbers the model produced the way it produces words. Some will be right. Magic MIS splits the job in two. A deterministic engine computes every variance and decides which are material. Only then does the model write — around placeholders, never digits — and the engine inserts the figures afterwards. A draft containing a number of its own fails a check and is never shown.

The same figures feed the workbook, the dashboard and the chat, so a follow-up question — “which customers?” — is answered from the numbers the commentary was written from. To write this kind of commentary yourself, the method is in how to write MIS commentary.

Flux analysis is the review of how much each account balance moved between two periods, usually this month against last, with an explanation for every movement over a set threshold. Controllers run it after the close to catch what the close missed — an accrual that was not reversed, a cost posted to the wrong account, a month of depreciation forgotten — before the numbers go to leadership. It covers the balance sheet as well as the income statement, because an error in one always has a twin in the other.

The arithmetic is identical: a figure, a baseline, the difference and a reason. What differs is the baseline. Flux analysis compares a period with an earlier period of actuals, and its purpose is control — is the month right? Variance analysis is often used for actuals against a budget or forecast, and its purpose is performance — did the month go to plan?

Magic MIS does both. Every line is set against last month, the same month last year and the year to date, computed from your trial balances and filtered to what is material: the comparison that both checks the close and explains the month. Enter a budget and every line is set against it too — on the board, in the commentary and in the suggestions on where to act — with each budget figure marked as your plan, never your books.

A month-over-month variance asks what changed since last month. It is quick to act on and noisy: a five-week month, a quarterly bill or a seasonal peak all show up as movement that means nothing. A year-over-year variance compares the month with the same month last year, which takes the season out and leaves the trend, but it reacts slowly to something that changed last week.

Read them together. A line that moved against both is worth a sentence; one that moved against last month only is usually timing; one that moved against last year only is the trend creeping. The commentary here is written against both, with the year to date beside them.