Restaurant cash handling, from the opening float to a variance with a reason

A drawer that lands exactly right every single night is unusual, and an exact match was never the point of counting. Change gets given in a hurry. A ten-rupee coin goes into the wrong tray. Somebody pays ₹450 for vegetables out of the till at eleven in the morning and means to write it down after the lunch rush. The routine exists so the small gaps stay small and visible, and so the one night with a real gap looks different from every other night.

Three gaps are ordinary. Rounding and change errors, in both directions across a week. A paid-out nobody recorded: restaurantdaily.ai, an Indian daily-operations product built squarely on this problem, puts it plainly, that a rupee out of the drawer with no entry is untracked variance by definition (restaurantdaily.ai, read 2026-09-09). And a counterfeit note, which is a one-sided loss: under RBI's Master Direction on Counterfeit Notes 2024, a fake tendered at your bank is not credited, it is stamped and impounded, and it never comes back (read via Taxguru, 2026-09-09).

There is no published rupee threshold, in India or anywhere useful. Xenia's retail cash page says a variance over the store's own threshold without ever naming one, and Petpooja's shift-change guide marks the ₹160 in its worked example as invented. Pick your own band, and hold the rule that does the work: every gap gets a reason.

Cash is also not going away in a restaurant. RBI research finds cash falling from about 80% of consumer spending in 2021 to under 60% in 2024, while still carrying 50% to 60% of household expenditure (Bhuyan, RBI Bulletin October 2024, via Data For India, read 2026-09-09), and UPI ran 24,508.96 million transactions worth ₹29,82,355.95 crore in August 2026 alone (NPCI product statistics, read 2026-09-09). What share of your own sales is cash, nobody can tell you: two Indian vendors publish figures that contradict each other with nothing behind either. Count a week of your own and you will know.

The float is the change you start the shift with, and its only job is to be the same number every day. Decide it once, count it in at the start of the shift, and never quietly top it up from the day's takings. The moment the float moves without a record, the expected figure moves with it and the variance stops meaning anything.

How much? No survey exists for India. One Indian POS vendor suggests ₹3,000 to ₹5,000 and publishes nothing behind the number, so treat it as a starting guess rather than a benchmark. The better test is your own first hour: enough small notes and coins to give change until the takings start funding themselves. Denominations matter more than the total, and a float of ₹3,000 in five-hundred-rupee notes is not a float.

On the register the float is the first thing the cash box asks for, and every row it writes afterwards is a ledger event carrying the signed-in staff and the till it happened on, never the PIN digits. Anyone signed in as owner, manager or cashier can run the box, which matters at 11pm when the owner has gone home.

Money leaves a restaurant drawer all day. ₹450 of vegetables, a gas cylinder, the auto fare for a delivery that could not wait, ₹200 to the plumber. Every one of them is legitimate and every one of them makes the drawer short by exactly that much. The rule is one line, written at the moment the note leaves the drawer, saying what it was for. Almost everything that goes wrong here goes wrong because somebody meant to write it later.

Money coming the other way gets the same treatment. A partner putting ₹2,000 into the till for change is a Cash in slip, not a mystery surplus at midnight, and it lifts the expected figure by exactly that much. Keep the paper voucher too if that is your habit, but the entry that reaches the expected figure is the one on the register.

A paid-out is also an expense, and it belongs in the week as well as in the drawer. That is what the weekly close is for.

A blind count means the person counting does not know what the till expects. They count the notes and coins, write the total down, and only then see the expected figure. If the counter can already see that the register expects ₹18,450, the count comes to ₹18,450, and nobody has lied to you. That is simply how people count against a target.

The term comes from American and global retail writing rather than Indian practice, and no Indian source we could find uses it by name. Xenia names it as the count made without telling the counter the expected total, and Solink sets it beside separation of duties in the same list (both read 2026-09-09). Their dollar thresholds do not travel to a Ratnagiri counter. The discipline does, and Petpooja's day-end guide already describes it in Indian words: count the drawer by denomination and write the total down before you touch the report (Petpooja, 17 July 2026).

The desktop register's cash box hides the cash sales figure and the expected total until the count has been typed in. Expected, Counted and Variance then appear together, and the variance is written in words: balanced, short by ₹100, over by ₹40. One honest caveat: the phone register has had a cash box since August and shows the expected figure before the count, so the phone is not blind.

Who counts matters as much as when. The person who took the money should not be the only one who counts it, and at a handover the drawer gets counted and signed over rather than carried through two shifts. Petpooja's shift-change guide makes the sharpest version of that point: a shortfall found at 11:30pm belongs to nobody in particular when two people handled the drawer and neither signed for it.

Expected minus counted is the whole arithmetic: opening float, plus cash sales, plus anything put in, minus paid-outs is what should be in the drawer, and the difference from what is actually there is the variance. A variance on its own is a number nobody acts on. A variance with a sentence next to it is a control.

• Small, and in both directions across the week. Change and rounding. Leave it alone and watch the weekly total instead of the nightly one.
• Small, always short, always the same shift. This is the one to act on. Watch the pattern rather than the single night: clustered shortages on one shift sit on the same list as excessive voids and cancellations clustered on cash (StudioMatrx, 25 July 2026). The staff theft post covers what to do next.
• Large and one-off. Look for the missing slip before you look at a person. It is nearly always a paid-out that never got written down, or a float somebody topped up.

On the register the cash box sits at the top of the day close, so the counting and the leftover checks happen in one sitting: the day close guide is the button-by-button version, and the day-end report post is the eight lines the whole close produces. After the shift closes, an hourly sweep sends the day-close cash-up email with gross, refunds, net, opening float, expected, counted and variance, so an owner who was not there reads the same numbers over the first chai.

A 12:40am bill belongs to last night's count. The trading day starts at 4am and you can move it (dashboard Settings), so a restaurant that shuts at 1am never has its night split across two dates.

A UPI or card payment never touches the drawer, so it plays no part in the count, and it still has to be checked every morning. What the till says you collected and what your bank shows are different numbers on any given day, because card and UPI settle on a cycle. That is the point Petpooja's day-end guide makes: collected and in the bank are not the same figure on the same date.

So the daily check is yesterday's tender split against today's credits, not today's against today's. Do it while the day is still recoverable. A missing card batch found on Tuesday is a phone call; the same batch found at the end of the month is an argument.

One thing to be plain about, because it changes how you read the UPI line: EasyKOT has no payment gateway. Cash is settled with change maths, and UPI and card are marked by the cashier as the tender the guest used. The till is a record of what your staff said happened, which is why the bank check is compulsory, and why a personal QR standing beside the fixed merchant one is the oldest leak in the business. One QR, fixed to the counter, no second standee.

None of this needs the internet at the shop. The register keeps selling offline and the cash box, the day close and the ledger rows behind them all work on the till's own database, syncing when the line comes back. If you want the routine on your own counter, EasyKOT is ₹299 a month or ₹2,990 a year, plus 18% GST as its own line, with three months free and no card. Tomorrow morning costs nothing though: decide the float, put a pad next to the drawer for the slips, and count blind before you look at the report.