Flux analysis, and the thresholds that stop it being noise

Flux analysis is the comparison, at close, of each account's balance or movement against a chosen base — last month, the same month last year, budget, or a rolling average — with a written explanation required wherever the difference matters. It is a completeness control: not a hunt for errors, but a demand that every material movement have a named cause before the period is signed.

What decides whether it works is the threshold, and almost everyone implements it as one number. One number is always wrong, in one of 2 predictable directions.

Four comparison bases, and when each is honest

• Prior month. Catches coding faults fastest, useless for anything seasonal — a landscaper's March always alarms beside February.
• Same month last year. Handles seasonality, hides drift: a cost creeping up 4% a month arrives as one annual change with one explanation.
• Budget or forecast. The only base testing a decision rather than a bookkeeping event; worthless if nobody revisited the budget after Q1.
• Rolling average of the last 3 periods. The best default where there is no budget, absorbing one-off months without hiding a trend.

One threshold always fails, in the same two ways

An absolute-only threshold means every large account clears it every month, so the reviewer works a list of movements that are simply the business operating, while a small expense account tripling goes unseen. A relative-only threshold reverses it: small accounts flood on trivial swings, and the largest account moves a substantial sum without reaching the percentage.

So flag only when both floors are crossed, then add the rules the pair alone misses. Worked below against an absolute floor of 5,000, a relative floor of 5%, and a hard ceiling of 50,000.

The third row justifies the whole arrangement. A 4.9% movement on the largest account outweighs everything else on the page combined, and the paired threshold suppresses it. The ceiling exists for that account.

A threshold that flags the same accounts every month is a subscription to a report nobody reads.

What a commentary must contain to count as explained

"Higher due to increased activity" explains nothing and is the modal answer. An accepted commentary needs 4 things; the third makes the control real.

1. A named driver. Not a direction but a cause: 3 sites onboarded mid-month, a supplier price change effective on the 12th, an annual premium landing here.
2. An amount attributed to that driver, so the explanation's arithmetic is visible rather than asserted.
3. A residual — whatever the drivers miss, stated explicitly. Requiring drivers to cover 80% of the movement stops a plausible sentence standing in for an investigation.
4. A recurrence flag: whether to expect this again next period, which is what makes the commentary usable by a forecast.

Residuals are where the interesting failures surface. Unexplained movements in income and expense of similar size, in one month, are usually a client transferring money between their own accounts — the detection rule is in transfers coded as income and as expense. A jump in fee expense with no matching sales movement usually means a settlement booked net, the gross-up problem in payouts that never match the sales behind them. A variance in an already-closed period points at duplicated data, as in the feed re-delivering transactions after a reconnect.

Who signs it, and what the signature asserts

The preparer writes the commentary; a second person accepts it. The reviewer is not re-checking arithmetic the system already did — they test the attribution, and may reject a commentary that is merely true. Both names and timestamps belong on the record, and nothing carries forward: a recurring variance is re-asserted each period, or the control becomes a subscription.

This is a close control, not a reporting pack. Advisory dashboards sit elsewhere; what flux owes the close is a defensible answer to why each material movement happened, before anyone relies on the figures — including whoever carries them onto a filing, where a wrong number produces a return that comes straight back rejected.

Implementing it is mostly threshold plumbing and a commentary store with versions and owners — the sort of thing we build under internal tools and ops, inside the reconciliation and close topic for accounting and tax practices.

• flux analysis
• close
• variance
• controls