How to set a materiality threshold you will actually stick to

The first monthly variance review you run will flag forty lines. The second will flag about thirty. There will not be a third.

This is the most common way the habit dies, and it is not a discipline problem. It is a threshold problem. If your rule for "worth explaining" is wrong, the review produces either a pile of noise or a page of nothing, and either way you stop opening the file.

So before anything else, set the rule.

What the threshold is actually for

Comparing two periods of a financial statement and writing down why every material line moved is a practice auditors call flux analysis, short for fluctuation. They do it at the start of an engagement for a practical reason: they cannot test every transaction, so they need to know where to look.

You have the same problem. You cannot review 800 transactions a month. You need a way to find the ten that matter. The threshold is that way, and it is the only part of the process you have to get right before the rest works.

One rule is never enough

Percentage on its own breaks on small line items.

Your bank fees go from $80 to $210. That is a 163% swing. It is also $130, and it means nothing. Flag it and you will spend your review chasing rounding, which is exactly how a review stops happening.

Dollars on their own break in the other direction.

Your payroll moves $9,000 on a $300,000 base. In absolute terms it is the biggest mover on the statement. In reality someone got a raise and a contractor invoiced late. It is a normal month.

The working rule is both at once. Flag a line only if it moved more than X percent and more than $Y. It has to clear both bars, not either one.

For an early stage company, 15% and $2,000 is a reasonable place to start. That is a starting point, not a standard.

Tune by output, not by argument

Do not try to reason your way to the correct threshold. Run it and count what comes out the other side.

You are aiming for eight to fifteen flagged lines a month.

Under eight, your rule is too loose. You are looking at a P&L that appears calm, and a calm P&L at an early stage company usually means the filter is eating real movement rather than that nothing happened.

Over fifteen, you will not finish. And an unfinished review is worse than no review, because you will assume the lines you never got to were fine.

Adjust one bar at a time. If small accounts keep surfacing, raise the dollar bar. If you are only ever seeing payroll and cloud costs, lower it. Two months of tuning is normal.

Revisit it when revenue changes materially. The dollar bar should scale with the size of the business; the percentage bar can usually stay where it is.

The balance sheet needs its own rule

Most people set one threshold, apply it to the P&L, and never get to the balance sheet. That is a shame, because the balance sheet is where the drivers are harder to see and the mistakes are more expensive.

It also needs different numbers. Percentage moves behave badly on balance sheet accounts: something that sat near zero last month goes to $40,000 and the percentage is either meaningless or undefined. A tighter dollar bar with a much looser percentage bar works better. For accounts that start at or near zero, just flag any movement above the dollar bar and skip the percentage test.

Then write the sentence

A flagged line is not the output. The written explanation is. And this is where most attempts fall apart, because people write down the number they already saw and call it commentary.

Weak:

Marketing expenses increased 38% month over month.

That is not an explanation. That is the same fact in sentence form. Anyone reading your board deck can see the number.

Strong:

Marketing increased $12,400 (38%). $11,000 of this is a one-time sponsorship of a conference in March, booked in full in the month of the event. Recurring paid spend was flat at approximately $20,000. Excluding the sponsorship, marketing was up 4%.

The second version tells you three things the first does not: what the driver was, whether it repeats, and what the trend looks like once you strip the noise out. That last part is the insight. Marketing spend did not go up 38%. It went up 4% and you bought a booth.

So the test for any line of commentary: does it say what moved, how much of the move each driver accounts for, and whether it happens again? If it fails any of the three, you have restated the number.

Start with two months

Open your last two closed months. Set 15% and $2,000. Count the lines that clear both bars. If you get four, drop the dollar bar to $1,000 and count again. If you get thirty, raise it.

That part takes about ten minutes and it is the only setup work there is.