Skip to content
allanninal.dev/build
Collaborate

Part 3 of 7 · Cash drawer reconciler series ~6 min read

How a variance gets explained

This is the part that everybody skips, because doing it by hand takes fifteen minutes per shift and there is a queue at the door. It is also the part that turns an accusation into an explanation. The explainer’s job is to spend those fifteen minutes, on every single shift, in about eight hundred milliseconds.

Key takeaways

  • Five lookups, run in order of how often each one turns out to be the answer.
  • A cash refund rung as card is the single most common cause, by a wide margin.
  • An explanation must match on both amount and direction, within a small tolerance.
  • Partial explanations are kept. Explaining £16.50 of a £22 variance is most of the work.
  • The explainer never invents a cause. Unexplained is a legitimate and common outcome.

Five lookups, in order

The five lookups that try to explain a cash varianceA vertical chain of six steps entered by a box labelled Raw variance, meaning counted minus expected. Step one asks whether a cash refund was rung as card, matching on amount and direction against the shift's refunds from the point-of-sale report; a match exits to Explained, noting the refund id. Step two asks whether a no-sale happened near the variance amount, checking the no-sale log for time and drawer; a close match exits to Likely explained, flagged as probable. Step three asks whether the float was changed, comparing the opening float with the expected one from the thresholds tab; a difference exits to Explained, recording the new float. Step four asks whether a safe drop happened after the count, comparing timestamps; a drop after the count exits to Explained, since it is an ordering problem rather than a loss. Step five asks whether a tip-out was taken, which applies in hospitality only; a logged tip-out exits to Explained. Step six is the Residual variance, what is genuinely left. A note says order matters, and the first lookup explains more shifts than the other four combined.AWS ACCOUNTRaw variancecounted minus expectedCash refund, card key?amount and directionRefunds that shiftfrom the POS reportExplainednote the refund idmatchNo-sale near the amount?within toleranceNo-sale logtime and drawerLikely explainedflag as probablecloseFloat changed?opening vs expectedFloat recordfrom the thresholds tabExplainedrecord the new floatdiffersSafe drop after count?timestamp orderDrop logamount and timeExplainedordering, not lossafterTip-out taken?hospitality onlyExplainedand paid, not lostloggedResidual variancewhat is genuinely leftOrder matters: the first lookup explains more shifts than the other four combined.
Fig 1. The five explanations, tried in order of how often each one is the answer. Anything left after all five is the residual, and only the residual is ever watched.
  • App integration
  • Machine learning
  • Security & identity
  • Management
  • Analytics

Cash refunds rung as card

This is the answer more often than everything else put together, and the reason is mundane. A customer returns something, the member of staff refunds them in cash because that is what the customer wants, and rings it through as a card refund because that is what the original sale was. The till now believes there is money in the drawer that somebody handed over an hour ago.

Matching is on amount and direction: a variance short by exactly the value of a card refund from the same shift is that refund, with very high confidence. The system notes the refund id in the explanation, which has a useful side effect — the same refund id appearing in explanations week after week is a training signal about one till or one procedure.

Why tolerances differ by lookup

  • Refunds: exact. A refund is a specific number that appeared in a specific report. If the variance is not that number to the penny, this is not the explanation, and loosening it here would explain away real problems.
  • No-sales: loose. A no-sale is a drawer opening with no transaction, and the system never knows what was taken out or put in. A no-sale close in time to a variance of a plausible size is evidence, not proof, and it is recorded as probable rather than explained.
  • Float changes: exact. The float is a number somebody wrote down. Either it differs from expected by the variance amount or it does not.
  • Safe drops: exact, with a time test. A drop is an amount and a timestamp. The interesting case is a drop recorded after the count time, which is not a loss at all — it is two records made in the wrong order, and it is common at the end of a busy shift.

Partial explanations count

A £22 variance with a £16.50 cash-refund match is not unexplained. It is £16.50 explained and £5.50 residual, and the residual is what goes into the window. Systems that treat explanation as all-or-nothing throw away most of their own work and end up watching noise.

A single cash variance explained in partsA horizontal row of five boxes. Raw: twenty-two pounds short. Refund: sixteen pounds fifty of that is explained by a cash refund rung as card. No-sale: none near the amount. Float: as expected. Residual: five pounds fifty, which goes into the watching window. A note says three quarters of it had a cause sitting in the day's own records.ONE VARIANCE, EXPLAINED IN PARTSRaw£22.00 shortRefund-£16.50 explainedNo-salenone near itFloatas expectedResidual£5.50, into the windowThree quarters of it had a cause sitting in the day's own records.
Fig 2. One shift’s variance worked through the five lookups. Most of it is explained; the small remainder is the only number that goes anywhere.

What the explainer will not do

  • It does not invent a cause. Unexplained is a normal, frequent, completely acceptable outcome, and pretending otherwise is how a reconciliation system starts lying.
  • It does not chain explanations. Two refunds that happen to sum to the variance is a coincidence at small amounts, and treating it as an explanation would hide real losses.
  • It does not look at who was on shift. It has no staff data and does not request any, which removes an entire category of misuse.
  • It does not adjust the till. The variance is a fact about a shift and stays recorded even once it is fully explained.

Next: what happens to the residual, and the window that decides whether anybody hears about it.

All posts