Skip to content
allanninal.dev/build
Collaborate

Part 4 of 7 · Budget variance reporter series ~6 min read

How timing gets filtered out

This is the part that makes the report short enough to read. Without it, a variance report is a list of lines that are over or under, most of which will be fine next month, and a reader who learns that within two months stops reading.

Key takeaways

  • Four tests: the missing regular, the early arrival, the year-to-date check, and the reversal.
  • Between them they typically remove two thirds of the lines that clear the thresholds.
  • A filtered variance is listed as filtered, in one line, never silently dropped.
  • The year-to-date test is the strongest and the simplest: is the year still on budget?
  • A variance filtered three months running is promoted, because a consistent filter is a finding.

The four tests

The four tests that separate timing differences from real varianceA vertical chain of five steps entered by a box labelled A variance over threshold, for one line in one month. Step one asks whether a regular transaction is missing, checking the last twelve periods of transaction history to see whether a supplier who bills monthly did not this month; if so it exits to Timing, because the invoice is not in yet. Step two asks whether something arrived early against the budget shape for calendar-fixed lines; if so it exits to Timing, noting that next month will be under. Step three asks whether the year to date is within threshold; if so it exits to Timing, because the year is fine. Step four asks whether the movement is a reversal such as an accrual or a credit note; if so it exits to Timing as a correction landing. Step five is A real variance, where the transactions are pulled. A note says every exit is listed in the report as filtered, with one line saying why.AWS ACCOUNTA variance over thresholdmonth, one lineA regular is missing?bills monthly,not this monthTransaction historylast 12 periodsTiminginvoice not in yetyesSomething arrived early?budgeted for laterBudget shapecalendar-fixed linesTimingnext month will be underyesYear to date on budget?within thresholdTimingthe year is fineyesA reversal?accrual, credit noteTiminga correction landingyesA real variancepull the transactionsEvery exit is listed in the report as filtered, with one line saying why.
Fig 1. The four timing tests in order of how often each one is the answer. Nothing is hidden: a filtered line appears in the report as filtered, with its reason.
  • App integration
  • Management
  • Analytics

The missing regular

The most common cause of an underspend and the easiest to detect. A supplier who has invoiced in eleven of the last twelve periods, for a similar amount, and did not this period, is almost certainly late rather than absent. The line is under budget by roughly their usual amount, and it will be over by the same amount next month.

The test is simply: are there suppliers on this account code with a monthly cadence who are missing this period, and does their usual total account for most of the variance? If yes, timing, and the report says which supplier.

The early arrival

The mirror image, and it only works because of the budget shaping from Part 3. A calendar-fixed line budgeted for August that gets paid in July produces a large July overspend and a large August underspend. Knowing the line is calendar-fixed and knowing which month it was due turns that from two anomalies into one recognisable event.

Year to date

The strongest test and the simplest. A line that is twenty per cent over for the month and within two per cent for the year has almost certainly just moved money between months. It is worth one line in the report and no attention.

The one caveat is that this test weakens as the year progresses. In month eleven, a year-to-date figure is so large that a substantial monthly variance barely moves it, so the threshold for this test tightens over the year rather than staying fixed.

Reversals

An accrual posted last month and reversed this month, a credit note against a prior period’s invoice, a misposting corrected. All of them produce a variance in a month where nothing happened, and all of them are identifiable from the transaction descriptions and the paired amounts.

Filtered, not hidden

Every filtered variance appears in the report, in a single line at the bottom, with its reason. This is a small thing and it is the difference between a report people trust and one they suspect.

How eleven over-threshold variances become four reported onesA horizontal row of five boxes. Over threshold: eleven lines. Missing regular: three filtered. Early arrival: two filtered. Year to date fine: two filtered. Reported: four lines. A note says the seven filtered lines are listed at the bottom of the report with reasons.ELEVEN LINES BECOME FOUROver threshold11 linesMissing regular3 filteredEarly arrival2 filteredYTD fine2 filteredReported4 linesThe seven filtered lines are listed at the bottom of the report, with reasons.
Fig 2. The filter’s arithmetic for one month. The seven that were removed are still on the page, which is what lets a reader disagree with the filter rather than distrust it.
  • Machine learning
  • Management
  • Analytics

Promotion after three months

A variance that is filtered as timing three months running is not timing. A supplier who has been late three months in a row has changed their billing, or has stopped supplying, or is disputing something. A line whose year to date has been fine three months running while every month is over is a trend that the year-to-date test is hiding.

So a filter reason that repeats is escalated: the line moves into the reported section with a note saying it has been filtered three times and why that is now itself the finding. It is the one place where the system overrides its own rule, and it exists because the most dangerous thing a filter can do is be consistently right about something that is quietly becoming wrong.

Next: what the report actually says.

All posts