ChecklistBy Khaled Hawari

Writing Variance Commentary Somebody Will Actually Read

Commentary that restates the number in words is the most common filler in a management pack and the easiest thing to fix.

Every comment on a variance page has to do three things. Name a cause. Attach a size to it. Say what happens next. A comment that does all three is worth the space it takes. A comment that does none of them is a caption, it is telling the reader something they can already see in the column beside it, and it should be deleted rather than improved.

That is the standard, and applying it strictly deletes more of the page than anybody expects on the first pass. The page gets shorter and gets read, which is a better trade than it sounds.

Where the commentary is drafted, and by whom, sits inside the close and is covered in the close calendar. This is about what the sentences have to contain once you sit down to write them.

The three tests, and what to do when a comment fails one

Failing all three is easy to deal with. Delete it. The interesting cases are the ones that fail exactly one, because each has a different remedy and people reach for the wrong one.

What the comment has What it is What to do
Size, no cause A caption. Comfortably the most common failure Find the cause or carry it as an open question with a name and a date. Do not write a plausible cause
Cause, no size An anecdote. Often true, and unusable, because the reader cannot tell whether to care Attach the number or drop it
Cause and size, no consequence History Say what happens next, or say explicitly that nothing does. “This does not recur” is a consequence and a useful one
All three A comment Leave it alone

The second row is worth sitting with. Anecdotes feel like good commentary because they contain real information about the business, and a preparer who has spoken to the branch manager naturally wants to relay what they heard. Without a size the reader has no way to rank it against the other seven things on the page, so it competes with them for attention and wins or loses at random.

What earns a comment

Three triggers. Nothing else on the page gets one.

  1. The movement is past the threshold. Set the threshold at the start of the year, in writing, and set it as a pair rather than a single number. A percentage alone produces noise from small accounts, where a modest sum looks dramatic. An absolute alone produces silence from large ones, where a real change in behaviour is a small proportion of a big number. A movement earns a comment when it clears both.

  2. It is the second or third consecutive month moving the same way, whether or not it clears the threshold. Drift is exactly what a threshold cannot see, and it is the pattern that matters most, because by the time drift clears a threshold it has been running for a quarter. This trigger requires somebody to look at three months rather than two, which is the entire reason it gets skipped.

  3. The reader asked about the line last month. Whether or not it moved. If it did not move, say that it did not, in one line. A question that gets answered once and then goes silent teaches the reader that asking is a way of generating a single response rather than a standing answer.

And one anti-trigger. A movement you explained last month, that behaved exactly as you said it would, gets one line confirming it did and nothing else. Re-explaining it in full is how a commentary page grows every month while saying less.

Worked examples

Every figure below is illustrative.

Written The problem Rewritten
Freight costs were 18,400 against 12,100 last month. Caption. The reader has both numbers already Freight rose 6,300 because eleven orders shipped out of the second warehouse while the racking was out of service, on lanes that cost roughly double the usual ones. The racking returns to service in the third week, so about half the increase persists into next month and none of it after that.
Administrative wages were up due to timing. A banned word with nothing attached to it Administrative wages carry a 41,000 bonus accrual catch-up posted in one month rather than spread, which is a presentation choice made in the autumn. It does not recur, and the year to date figure is unaffected.
Revenue was below budget owing to market conditions. Unfalsifiable, unactionable, and it invites the reader to discount everything else on the page Revenue was 96,000 below budget. Two of the four projects due to start in the month moved to the following one at the customer’s request, worth 88,000 between them, both now confirmed with dates. The remaining 8,000 is not yet explained and is carried as an open question.
Gross contribution on the services line improved, partly offsetting the above. Written to make the page feel balanced. It has no cause and no consequence Delete it. If the improvement is real and continuing, it is its own comment with its own cause and it does not need to be attached to somebody else’s bad news.

The last row is the one preparers resist and it is the one that most improves a page. Offsetting comments exist to manage the reader’s mood rather than to inform a decision, and a reader who has been on the receiving end of a few of them starts reading the whole page for tone instead of content.

Traceability

Every comment points at something: a transaction, a signed decision, a document, or a person who said it. Not at a category.

The reference lives in the working file rather than on the page, because a page cluttered with invoice numbers stops being readable, but it has to exist. The test is whether somebody who was not in the room could get from the sentence on the page to the evidence in under a minute. If the answer is that they would have to ask you, the comment is not traceable, it is remembered, and it will not survive your next holiday.

The words that mean “I do not know”

These are the words people reach for when the cause has not been found. Each is permitted only with a specific clause attached, and one of them is not permitted at all.

Word What it usually means alone What makes it permitted
Timing It will reverse, probably, or I have not looked The transaction it refers to, and the month it reverses
Phasing Timing, dressed better The same requirement as timing
Seasonality It happens every year, I think The same month in the prior two years, with the figures shown
One-off I hope it is What it actually was, and why it cannot recur
Mix Something moved between categories Which two categories, and how much each moved
Volume The number went down because the number went down Whatever drove the volume, which is the question the reader was asking
Market conditions Nothing at all Nothing. There is no permitted use of this phrase in a management pack

Market conditions gets its own ruling because the alternative is worse than an empty explanation. A pack that offers a view on a market has started making claims it cannot support from its own records, and once one sentence in a document is unsupported the reader has to start deciding which of the others are. Keep the pack to what the company’s own systems can evidence.

Timing claims are predictions, so check them

A timing explanation says a specific amount will reverse in a specific month. That makes it the only kind of comment on the page that can be marked right or wrong, which is a gift and almost nobody uses it.

Keep a timing register. One line per timing claim: the month it was made, the amount, the month it is due to reverse, and whether it did. Read the register before writing the next month’s commentary, and report any claim that failed to reverse in the section of the written page that deals with what was got wrong last month.

Do this for two months and the use of the word drops sharply, which is the point. Without it, “timing” becomes the word that means “I do not know”, the reader works that out faster than the preparer expects, and it then contaminates the genuine timing differences, which are common and completely legitimate.

Two tests before it goes out

Cover the numbers and read only the commentary. If you cannot tell what happened to the business from the words alone, they were captions and they were written by reading the columns rather than by understanding the month.

Look at the length of each comment. If they are all roughly the same, they were written to a shape. A real month is lopsided: one comment runs six lines because something genuinely happened, three run one line each, and two lines that moved got no comment because nothing caused them worth reporting.

What is deliberately left out

Lines the reader cannot act on. A movement in an account nobody controls, with no decision attached, belongs in the file rather than on the page.

Anything already said on the written page at the front of the pack. Say it once, in the place the reader gets to first, and let the variance page carry the detail underneath it rather than a second version of the summary. The standing pack is built so those two do not overlap.

Judgements that are not labelled as judgements. If you believe the margin loss is a pricing problem rather than a run of bad projects, say so and say it is your reading, with what it is based on. A judgement presented as a fact is the thing that costs a finance function its credibility, and it usually only costs it once.

And padding. Some months produce three comments. Write three. A page that carries twelve every month regardless has told the reader that the count is fixed, and the month there genuinely are twelve will look exactly like all the others.

What we do not do

We do not write a cause we have not traced. An open question with a name and a date on it is a better thing to hand a reader than a confident sentence that turns out to be wrong, and it costs the preparer far less than it feels like it will.

We do not carry the same explanation into a third month without escalating it. Two months is an explanation. Three is a condition, and it belongs on the written page as a decision rather than on the variance page as a comment.

And we do not soften a comment because of who is in the meeting. The commentary is written before the meeting, by one person, and it says the same thing whoever is going to read it.

MoreOther working documents

If this keeps failing in the same place.

A document that has to be re-explained every period is a process problem rather than a documentation problem. That is the point at which handing the function over is cheaper than fixing it again.