At the end of a shift your cashier counts the drawer. The system expected $1,284.00. There is $1,280.00 in the drawer.
Four dollars. What do you do?
The wrong answers are the two obvious ones: accuse somebody, or type in $1,284.00 and go home. Both destroy the same thing — the only signal you had.
What the number is made of
A drawer variance is not a mystery. It is a subtraction, and each part of it is something a person did:
Opening float, counted by a person at the start. Plus cash sales, which the system knows exactly. Minus cash refunds, likewise. Plus pay-ins — change brought in from the safe. Minus pay-outs — the delivery driver, a box of bags, the window cleaner. Minus drops — takings moved to the safe mid-shift so there is not $3,000 in a drawer at six o’clock.
That gives you what should be there. The count gives you what is. The difference is the variance.
Notice what is not in that list: card payments, bank transfers, store credit. None of them touch the drawer. If your system includes card takings in the expected cash figure, every shift will look short by exactly the card total, and your staff will learn within a week that the variance means nothing. That is worse than not measuring it, because now you have a number everyone ignores.
Why you record it instead of fixing it
The instinct is to make it balance. The drawer says $1,280, so type $1,280 as both the expected and the counted figure, and the shift closes clean.
Do that and you have thrown away the single most useful number a register produces.
One short drawer is not information. Cashiers give the wrong change, a note sticks to another note, someone takes $5 for the parking meter and forgets to write it down. Retail is a physical business and physical businesses have friction. A shift that is out by a few dollars is a shift, not a crime.
The information is in the pattern, and you can only see a pattern if you kept the individual points. Three cashiers, forty shifts each:
- Cashier A: forty shifts, cumulative variance $+2.00. Out most days, either way, nets to nothing. Normal.
- Cashier B: forty shifts, cumulative variance −$3.00. Also normal.
- Cashier C: forty shifts, cumulative variance −$186.00, and never once over.
Cashier C is the finding. Not because they are stealing — they may be, or they may be consistently miscounting the float, or working the busiest register with no help, or using a drawer where the $20 slot is jammed. You do not know. What you know is that this is not random, because random does not run in one direction for forty shifts.
You could never have seen that from a system that made every shift balance.
What to do with a variance
Do not act on one. Look at it, note it, move on.
Do look at cumulative variance per person, monthly. Random errors cancel out. If someone’s total is drifting steadily one way, that is a process problem until proven otherwise.
Do compare people on the same register. A drawer that is short under three different cashiers is a drawer problem, or a procedure problem, not a person problem.
Do require a reason for pay-outs. Most “shortages” are undocumented pay-outs. If taking $25 out for the water delivery requires typing “water delivery”, the variance stops being a mystery and starts being a record.
Do count blind if you can. If the cashier can see what the system expects before they count, some of them will count until they get that number. The count should be what is in the drawer, not what makes the screen happy.
The design consequences
Once you accept that the variance is the point, some things in the software have to be deliberately less convenient.
Nothing may auto-close a shift. A scheduled job that closes registers at midnight has to invent a closing count, and an invented count makes the variance meaningless. If a shift is still open at nine the next morning, the right behaviour is to say so — loudly — and wait for a person.
The variance is stored, not applied. It sits on the shift as a fact. It does not adjust the cash account, it does not create a correcting entry, and there is no button that makes it go away.
Opening float is typed, not defaulted. Carrying yesterday’s closing figure forward as today’s opening float feels helpful and quietly guarantees that any error propagates until someone notices.
None of this makes closing a register faster. It makes the number mean something, which is the entire reason for counting.
The short version
A drawer variance is the only routine measurement in a shop that compares what the system believes against what physically exists. Everything else — stock levels, customer balances, revenue — is the system checking itself against itself.
That makes it worth protecting, even when it is inconvenient. Especially then.