Walk into any managed pub group's head office and ask the operations director what percentage of wet sales go unrecorded each week. If they're honest, they'll tell you it's somewhere between eight and twelve per cent. If they're not honest, or they genuinely don't know, the number is probably higher.
That range — eight to twelve per cent — is not a guess. It comes from decades of stock auditing data across UK licensed premises, cross-referenced with EPOS transaction logs and supplier delivery records. The British Beer and Pub Association has referenced similar figures. Independent stock auditors like Venners and Stocktaking.com routinely confirm them when onboarding new clients. And yet most operators treat the number as an acceptable cost of doing business, because they cannot see exactly where it is happening.
Why 8–12% is the benchmark
The figure is a composite. It does not come from a single source of loss. Rather, it reflects the accumulated effect of multiple small mechanisms operating simultaneously across every shift, every bar, every week. Some of it is theft. Some of it is error. Some of it is culture. And importantly, these mechanisms interact with each other — they compound rather than simply add up.
When a venue first connects its EPOS, camera feeds, and stock data to an integrated monitoring platform, the typical discovery is not a single dramatic problem. It is dozens of small problems that collectively explain why GP% is three or four points below where it should be. The eight to twelve per cent benchmark holds because these problems are structural — they exist in nearly every venue, and they persist until they are made visible.
Mechanism one: cash bypassing the till
The most direct form of unrecorded transaction is a drink served and paid for in cash, where the transaction is never entered into the EPOS. The bartender takes the money, puts it in the till (or their pocket), and no record exists of the sale. In a busy Friday night environment with three bartenders working simultaneously, monitoring every single cash transaction visually is impossible for a manager who is also serving, dealing with customer complaints, and managing the floor.
The tell-tale sign in the data is a discrepancy between stock depletion and recorded sales. If your cellar records show you dispensed 120 pints of Amstel this week, but EPOS only recorded 108 sales, you have twelve unrecorded pints. At an average selling price of £5.80, that is £69.60 of revenue that went somewhere — but not through your books.
Scale that across every draught line, every spirit, every bottle of wine, and the numbers become significant very quickly. A venue doing £10,000 per week in wet sales losing even five per cent to cash bypass is haemorrhaging £500 per week, or £26,000 per year. And five per cent is conservative.
Mechanism two: error corrects and void transactions
Every EPOS system has a void or error-correct function. It exists for a legitimate reason — if a bartender accidentally rings up the wrong item, they need to be able to correct it. But the function can also be used to remove a completed transaction from the record after the cash has been collected.
Here is how it works in practice. A bartender serves a round totalling £18.40. The customer pays cash. The bartender rings the sale through the till, takes the cash, gives change, and the transaction is recorded. Ten minutes later, during a quieter moment, the bartender performs an error correct on that transaction. The EPOS removes it from the sales log. The cash is still in the till — but now there is an £18.40 surplus that the bartender can remove at the end of the shift without the till being down.
The pattern is detectable if you know what to look for. Venues with abnormally high error-correct rates — particularly concentrated on certain staff members or certain shifts — almost always have a problem. A healthy error-correct rate is below one per cent of total transactions. Anything above two per cent warrants investigation. Above five per cent, and you are almost certainly looking at systematic misuse.
Mechanism three: over-pouring and unmeasured serves
Not all unrecorded loss is deliberate. Over-pouring is the single largest source of unintentional stock loss in UK pubs. A bartender who free-pours spirits rather than using a measure consistently delivers 30–35ml instead of the legal 25ml serve. That is a 20–40% over-pour on every single spirit sale. Over a week of busy trade, the stock loss from over-pouring alone can account for three to four per cent of total wet sales.
The same principle applies to draught beer. A pint that is consistently over-filled by even half a centimetre means you are giving away free product on every pour. Multiply that across hundreds of pints per day, and the volume loss is substantial. The sale is recorded — the customer paid for a pint — but the stock depletion exceeds what the sales data would predict, because you dispensed more liquid than the sale accounts for.
What the data looks like when venues first connect
When a venue first connects to Minnie and we begin cross-referencing EPOS data, stock records, camera feeds, and delivery invoices, the first two weeks of data typically reveal a pattern that surprises even experienced operators. The gap between theoretical GP and actual GP is almost always wider than expected.
Theoretical GP is what your gross profit should be, based on the prices you charge and the cost of the products you sell. Actual GP is what you actually achieve after stock loss, over-pouring, unrecorded transactions, and waste. The gap between the two is your variance — and in most UK pubs, that variance sits between three and six percentage points.
A three-point GP variance on £10,000 weekly wet sales is £300 per week, or £15,600 per year. A six-point variance is £31,200 per year. These are not hypothetical numbers. They are what we see in practice, repeatedly, across venues of all sizes.
How AI camera cross-referencing catches it
The traditional approach to catching unrecorded transactions is the stock take — a periodic count of every bottle, keg, and case in the building, compared against sales data and deliveries. The problem is that stock takes happen weekly or fortnightly at best, and by the time you have the data, the losses have already occurred. You know you lost stock, but you do not know exactly when, how, or who was responsible.
AI camera cross-referencing changes the equation entirely. By analysing camera feeds in real time alongside EPOS transaction data, the system can detect when a drink is served but no corresponding transaction appears on the till within a reasonable window. It can identify when a pour happens without a preceding order being keyed in. It can flag when cash is handled at the bar but no sale is recorded.
This is not about watching staff with suspicion. It is about giving operators visibility into what is actually happening on their bar, shift by shift, in a way that was previously impossible without physically standing behind every bartender for every hour of every day.
What to do with the data
Establish a baseline. Before you make any changes, you need to know where you stand. Run the system for two full trading weeks without intervening. Let the data accumulate. You need a clear picture of your current variance before you can measure improvement.
Make structural changes. Once you have a baseline, address the largest sources of loss first. If error-correct rates are high, implement manager-authorisation requirements for all voids. If over-pouring is the primary issue, enforce measured pours and consider spirit dispensers. If cash bypass is the problem, move to cashless or reduce cash-handling touchpoints.
Monitor continuously. The biggest mistake operators make is treating loss prevention as a one-off project. You run a stock take, identify the problem, fix it, and move on. Six months later, the same problems have crept back. Continuous monitoring changes this — it keeps the data visible, keeps staff aware that the data is being watched, and catches regression before it becomes embedded again.
The eight to twelve per cent figure is not a fact of life. It is a consequence of operating without visibility. Once you have the data, the number comes down. In venues that have run Minnie for six months or more, the typical unrecorded transaction rate drops to two to four per cent — still not zero, but a dramatic improvement that translates directly into recovered revenue and a healthier bottom line.