Flux & Variance
Why flux commentary gets sent back
Almost always for one of three reasons, and none of them are about the numbers being wrong.
1. It restates the number instead of naming a driver
This is the most common one by a wide margin.
"Spend is pacing above target." "Costs increased versus prior month." "Revenue is unfavorable to plan." These describe the variance in words. They contain no information the reviewer didn't already have from the column to the left.
The test is simple: does the explanation name something you could go verify?
| Restatement | Driver |
|---|---|
| Professional fees increased significantly | Two advisory engagements that had been deferred were invoiced in the same period |
| Freight is running above prior month | Higher receipt volume plus a shift toward expedited shipping on late orders |
| Payroll is unfavorable | Two open roles filled mid-month, plus the additional pay period falling in this month |
The right-hand column can be checked. Someone can pull the invoices, look at the shipping mix, or count the pay dates. The left-hand column can only be agreed with or disagreed with.
This is also the specific failure mode of AI-drafted commentary. A language model given only the variance and the account name will produce fluent, professional-sounding restatement — because restatement is all that's available to it. It doesn't know the engagement was deferred. If the draft describes the movement without naming a cause, it isn't finished, and its fluency makes that easy to miss.
2. It doesn't separate timing from run rate
A large share of what gets labeled favorable is a bill that hasn't landed yet.
Not every movement falls into one of these two buckets — plenty of accounts move for reasons that are neither. But where the distinction does apply, merging them produces commentary that's technically accurate and practically misleading:
- Timing. The cost exists. It will appear next period. Nothing structural has changed.
- Run rate. The underlying level of spend or revenue has actually moved.
If you don't make the distinction, you'll be asked to make it in the meeting — usually by someone who noticed the same line reversed last quarter. One clause pre-empts it: favorable on timing; invoice expected next period reads very differently from favorable.
3. It isn't complete
This is the one preparers find hardest to catch, because it's structurally invisible from where they sit.
The preparer works forward: pull the schedule, sort by variance, explain everything above threshold. The scope of the work is defined by what appears on the report.
The reviewer works backward, asking whether the explanation set is complete. That's a harder question, because completeness requires knowing what should be there — not just checking what is.
Two specific gaps a variance report will never flag for you:
Errors that didn't move anything. A threshold on movement can't surface an entry coded to the wrong account within the same family, or an accrual whose composition changed while its balance didn't. Narrow, but worth knowing where it sits. More on this here.
Inconsistent threshold application. If you can't state the rule that decided which lines got explained, the schedule will look arbitrary to someone reading it cold — and the line you skipped is the one they'll ask about.
A short pre-send check for your variance commentary
- Does every explanation name a cause someone could verify?
- Is anything called favorable actually just late?
- Was the same threshold rule applied down the whole schedule?
- Did any account stay flat that shouldn't have?
- Would each line survive one follow-up question?
That last one is the whole test, compressed. If you can't answer the obvious follow-up, the reviewer will ask it.
If you work this in Excel: the free flux template flags what needs explaining and shows the residual your drivers don't cover. The CloseOps Flux & Variance System ($79) starts from your trial balance instead: confirm each account's classification and it builds the income statement and balance sheet flux statements, then ranks what to investigate.
Related: what flux analysis is, if you're new to it. The whole method in month-end flux, start to finish. Also most of this is computable — four of the five properties of good commentary can be tested by formula. And why the bar producing these rejections was never written down, and why agent-drafted commentary still gets sent back. Or the Reviewer Red-Flag Checklist — free.
Found something wrong or missing? Tell me here — anonymous, thirty seconds.