Why should I not explain every variance in the report?

Because a report where every line carries an explanation trains everyone to skim past all of them, and the few real signals drown in a sea of dutiful noise. Most variances are the normal month-to-month movement of a business that was never going to land exactly on budget, so writing a comment against them manufactures a cause that does not exist. Reporting by exception, meaning saying nothing at all about the lines that behaved normally, is not laziness. It is the thing that makes the exceptions visible. A 3 line narrative naming the real signals is worth more to a CFO than 40 lines that explain everything and reveal nothing.

How do I tell a real signal from normal noise in a variance report?

Run every variance through two filters rather than one. Materiality asks whether the gap is big enough in dollars to change an action you would actually take, with the threshold tied to decisions rather than to a percentage. Signal asks whether the variance falls outside the range that line normally moves within for ordinary reasons, estimated from 12 or more months of history. Only variances that clear both filters deserve investigation. In the worked month, revenue missed by 3.75% against a line that normally holds within 1.5%, which is the one line that earns a deep dive.

Should I investigate the biggest dollar variance first?

Usually not. The largest absolute variance is often just the largest line item doing its normal thing, because a big line with big natural volatility will always top the list. In the worked example COGS shows a 25,000 dollar miss, the second largest in the report, but it moved inside its normal 2.5% range, so it is noise. Revenue is the real signal despite being a similar size, because it broke a much tighter band. Size without context is not signal, which is why dollar screening alone is the wrong single filter.

What about a large percentage swing on a small line?

Percentage without dollars sends you chasing rounding. A 30% swing on a small cost line is eye-catching and usually immaterial, so it does not earn a full investigation. It does earn a note. In the worked month freight jumped 30% against a normal band of 8%, which is far outside its range but trivial in dollars, so the verdict is watch rather than investigate. Not because this month matters, but in case it is the leading edge of something that will become material later.

When does the two-filter screen break down?

It assumes each line has a stable history from which normal can be defined. It breaks wherever the baseline has genuinely shifted: a new pricing structure, an acquisition, a step change in volume, or a new cost line with no history at all. In those cases the old range is meaningless and should be reset deliberately rather than trusted. It also treats a single month as the unit of analysis, when the real signal is often the direction over several months, so a line that stays in range while drifting the same way for 4 months can matter more than a one-off breach. The screen decides where to spend attention. It does not replace reading the business behind the numbers.

Which Variances Deserve Your Attention

Core Argument

A monthly report lands on the CFO's desk with 40 line items, each with budget, actual, variance, and a dutiful comment. The CFO reads all 40 and comes away knowing no more than before about what needs attention. The report explained everything, which is the same as explaining nothing. Most variances are noise: the normal month-to-month movement of a business that was never going to land exactly on budget. The job is not to explain all 40. It is to find the 3 that matter and ignore the rest with confidence.

The core insight

A variance is not automatically information. Most of the time it is just the difference between a forecast that was never exact and a reality that is inherently variable.

The question a variance should trigger is not "why is this number different from budget?" Almost every number is different from budget. The question is "is this difference bigger than this line normally moves, and big enough to matter?" A variance earns your attention only when the answer to both is yes. If a line item swings within its normal range, explaining it is a fiction: you are inventing a story for random noise. If a line moves outside its normal range but the dollars are trivial, explaining it is a waste: you are spending scarce analytical attention where no decision hangs.

Most variance reports are screened, if at all, by dollar size alone. That is the wrong single filter, because it flags large lines that were always going to vary a lot and misses small lines that have broken from their pattern. You need two filters, not one.

"A report that flags 3 signals is more useful than one that explains 40 variances."

The framework: the two-filter variance screen

Before investigating any variance, run it through two questions.

1

Materiality: is it big enough to change an action?
Set a threshold in dollars, tied to decisions rather than to a percentage. A variance that cannot move any decision you would actually make does not deserve investigation no matter how large the percentage looks.
2

Signal: is it outside the line's normal range?
Every line item has a characteristic volatility, how much it swings month to month for ordinary reasons. Estimate that range from history. A variance inside the range is noise. A variance outside it is a signal that something changed.

Only variances that clear both filters deserve real investigation. That gives 4 outcomes:

	Within normal range	Outside normal range
Material dollars	Monitor. Large but normal.	Investigate. This is the signal.
Immaterial dollars	Ignore.	Note, low priority.

A real month run through the screen

Line	Budget	Actual	Variance	Normal range	Verdict
Revenue	$2,400,000	$2,310,000	−3.75%	±1.5%	Investigate
COGS	$1,440,000	$1,465,000	+1.7%	±2.5%	Monitor, in range
Payroll	$300,000	$302,000	+0.7%	±1%	Ignore
Marketing	$120,000	$138,000	+15%	±6%	Note, small dollars
Freight	$60,000	$78,000	+30%	±8%	Watch, early signal

Read by dollar size alone, COGS looks like the story: a $25,000 miss, the second largest in the report. But COGS moved inside its normal 2.5% range. It is noise, and writing a paragraph about it manufactures a cause that does not exist. Revenue is the real signal: a 3.75% miss against a line that normally holds within 1.5%, and material. That is the one line that should get the deep dive. Freight is tiny in dollars but has jumped far outside its band, so it earns a watch, not because of this month, but in case it is the leading edge of something.

The screen turns 40 comments into one investigation, one watch, and 38 lines you can leave alone with a clear conscience.

Why experienced managers get this wrong

1

They comment on everything.
A report where every line has an explanation trains everyone to skim past all of them. The real signals drown in a sea of dutiful noise. Reporting by exception, saying nothing about the lines that behaved normally, is not laziness. It is what makes the exceptions visible.
2

They chase the biggest dollar variance.
The largest absolute variance is often just the largest line item doing its normal thing. Size without context is not signal. A big line with big natural volatility will always top the list and will usually be nothing.
3

They react to big percentages on small lines.
A 30% swing on a small cost line is eye-catching and usually immaterial. Percentage without dollars sends you chasing rounding.
4

They treat one month as a trend.
A single variance is one data point. The signal in a business is usually the direction over several months, not the gap in any one of them. A line that is within range this month but has drifted the same way for 4 months can matter more than a one-off breach.
5

They have no baseline for normal.
Without knowing how much a line usually moves, you cannot tell signal from noise at all, so every variance looks equally suspicious and gets the same shrug or the same panic. The normal range is the whole foundation. Without it, the other filter is guessing.

Putting it to work tomorrow

Build a normal range for each major line, once. Pull 12 or more months of history and estimate how much each line typically swings for ordinary reasons. It does not need to be a formal control chart. A defensible high-low band is enough to separate a breach from routine movement.
Set a materiality threshold tied to decisions. Pick a dollar level below which a variance would not change anything you do, and stop investigating below it. The variance screen calculator applies that threshold and the normal ranges together, and returns the verdict for each line. Free the analytical time for the variances that could actually move a call.
Report by exception. Comment only on the lines that clear both filters. Let the silent lines stay silent. A 3 line variance narrative that names the real signals is worth more to a CFO than 40 lines that explain everything and reveal nothing.

Know the limits

The screen assumes each line has a stable history to define normal. It breaks where the baseline has genuinely shifted: a new pricing structure, an acquisition, a step change in volume, a new cost line with no history. In those cases the old range is meaningless and you reset it deliberately rather than trust it. And keep judgment for the small out-of-range variances: some of them, like the freight jump, are early warnings that matter long before they become material. The screen decides where to spend attention. It does not replace the reading of the business behind the numbers.

The principle worth keeping

The purpose of variance analysis is not to explain why every number differs from budget. Every number differs from budget.

"The purpose is to find the few gaps that are both real and worth acting on, and to say nothing about the rest."

Free Excel Model: FP&A Budget & Variance Template

An annual budgeting model with an assumptions hub, monthly P&L, actuals input and automatic variance reporting, so the screen above has something to run against.

Download Free ↓