The profit leaks hiding in your POS data
The costs you can see are rarely the expensive ones. Here is what reading 43,501 real orders actually surfaced — including why the system labeled its own best finding medium confidence.
By Greg Bush · Updated July 23, 2026 · 9 min read
The leaks are almost never where the conversation goes. They hide in items you sell constantly at a margin you last checked a year ago. On one bakery-café pilot, reading 43,501 real orders and 101,085 line items surfaced five high-volume items priced below the shop’s own average margin — roughly $1,900 a month of pricing opportunity. The system labeled that finding medium confidence by itself, because food costs were still owner estimates. That label matters as much as the number, and this article explains why.
Every operator already knows where they think the money is going. Labor was high last month. That one supplier keeps creeping. The delivery apps take their cut. Those are all real, and they are all the things you can already see — which is exactly why they are rarely the biggest leak.
The expensive problems are the ones with no symptom. An item that sells eighty times a week at a price set before your last two cost increases doesn’t announce itself. It looks like a best-seller. It shows up in every daily total as revenue. Nothing in your point-of-sale flags it, because from the POS’s point of view nothing is wrong — it sold.
Never confuse the machine being impressive with the machine being right.
Your POS is already keeping the record you need
A point-of-sale system is built to take money and move the line along. It is not built to explain your business to you. But in doing its job it quietly accumulates the most honest record you own: what actually sold, at what price, in what combination, at what hour, over and over, for years.
The gap is not data. It is that nobody has the hours to read 43,501 orders line by line, and no owner should have to. That reading is the part a machine is genuinely good at — and it is the only part we let it do unsupervised.
The four leaks that hide in plain sight
1. Items priced below your own average margin
Not below cost — below the margin the rest of your menu earns. These are the quiet ones. They sell well, so they feel like winners, and every sale drags your blended margin down a fraction. Volume is what makes them expensive: a small gap multiplied by a high count is a bigger number than a large gap on something you sell twice a week.
2. Cost drift since the last time you priced
Menu prices get set deliberately and then reviewed whenever there is time, which in practice means rarely. Ingredient costs move continuously. The distance between those two rates is a leak that grows on its own, and it is invisible unless something compares today’s cost against the assumption baked into the price.
3. Modifiers and combinations nobody priced
The base item was priced carefully. The substitution, the extra, the way half your regulars actually order it — often not. Line-item data shows what people really buy, which is frequently not the thing on the menu board.
4. Time-of-day patterns you staffed for once
Demand shifts season to season; schedules tend to be inherited from whenever they were last rebuilt. The orders themselves know when your business actually happens.
What that looked like on real data
The pilot is a family-owned bakery-café, running CoversIQ in production on its own point-of-sale export. Not a demonstration set, not a benchmark — the shop’s actual sales history.
pilot’s own point-of-sale
inside those orders
below-average-margin items
Five items, all high volume, all priced below what the rest of the menu was earning. None of them looked like a problem from the counter. Together they represented roughly $1,900 a month in pricing opportunity — on a menu the owner knew intimately and had priced himself.
Now the part most vendors leave out
That $1,900 is an opportunity identified, not money banked. The distinction is the whole article.
The system labeled its own finding medium confidence, unprompted, because the food costs it was reasoning from were owner estimates rather than invoice-verified figures. Change the cost inputs and the number moves. It said so rather than presenting a clean figure it hadn’t earned.
An analyst you can trust is one that tells you what it doesn’t know. A system that returns the same confident tone for a well-evidenced finding and a shaky one has removed the single most useful signal you had — which findings to act on today, and which to verify first.
Acting on it is also not automatic. Raising the price of five popular items is a judgment call about regulars, about what the shop across the street charges, about whether an item is a loss-leader on purpose. The system makes the case with evidence. The owner decides, every time.
What honest output looks like next to the usual pitch
| The usual dashboard | What we think it should do | |
|---|---|---|
| The finding | “Margin is down 3%.” | Names the five specific items, with their volume and the gap against your own average margin. |
| The number | A confident figure, tone identical whether the inputs were solid or estimated. | A figure plus a confidence label the system assigns itself — medium here, because food costs were owner-estimated. |
| The evidence | Trust the chart. | The orders and line items the conclusion came from, so you can check it against a week you remember. |
| The decision | Recommendations you are expected to accept. | A ranked case. The owner prices the menu — the system never changes anything. |
This is the same standard we ask you to hold us to. If a finding can’t show its evidence, it isn’t a finding yet.
How to run this check yourself, this week
You do not need us to start. If you export your POS data and can give an afternoon to a spreadsheet, you can find the first version of these leaks on your own:
The manual version of the same analysis
Export at least twelve months of order-level data — not the daily summary, the line items. The summary is where the leaks hide.
Rank every menu item by units sold. Work only on the top twenty; that is where volume makes small gaps expensive.
For each, put today’s real ingredient cost next to the price you charge. Today’s cost — not the one you used when you set the price.
Compute margin per item and compare each against your menu’s average. Anything high-volume and below average is a candidate.
Write down which cost figures were estimates rather than invoiced. That list is your confidence label — do that honestly and you are already ahead of most dashboards.
If that exercise finds nothing, you have learned something worth knowing and it cost you an afternoon. If it finds three candidates in your top twenty, the question becomes whether you want to repeat it every week by hand.
What this does not do
CoversIQ reads one restaurant’s operating data. That is a deliberate limit, not a roadmap gap. It does not roll several outlets into a combined view, and it does not produce a property-wide number across a hotel, a course, and a dining room — a resort using it is pointing it at the one restaurant inside the property, which is exactly the case it is built for.
It also does not touch your prices, your menu, or your schedule. It has no write access to anything. It reads, it ranks, it shows its evidence, and it stops — because the person who knows why an item is priced the way it is has never been the software.
And it does not replace anyone. On this engagement no staff were displaced; the owner stayed the decision-maker on every price, every menu change, every schedule.
Common questions
How much POS history do I need?
About twelve months is the useful floor, because it covers your seasonal swings. The pilot ran on 43,501 orders. Less than a year still works for pricing questions, but any claim about seasonality would be guessing, and should be labeled that way.
Does this work with my POS?
It works from an export rather than a live integration, which means the answer is usually yes — if you can get order-level data out of your system, it can be read. That also means nothing of ours connects to your payment flow.
Is $1,900 a month typical?
We have no basis to claim it is. It is one shop’s real figure on one real menu, and publishing it as a benchmark would be exactly the kind of number this article is arguing against. Your menu, volume and cost structure decide yours — which is why the check above is worth running before you believe anyone, including us.
What if my food costs are estimates too?
That is the common case, and it is fine — it just changes the confidence, not the usefulness. You will get candidates worth verifying rather than conclusions to act on blind. Invoice-verified costs on your top twenty items sharpen the whole picture faster than any software will.
Can it just fix the prices for me?
No, by design. Pricing is a judgment about your regulars and your street, and the system does not have that context. It builds the case; you make the call.
What to take away
The leaks with no symptom cost the most — high-volume items priced below your own average margin.
Your POS already holds the record; the missing piece is reading it line by line, which is the part a machine should do.
Real pilot result: five items, roughly $1,900/month of pricing opportunity, found across 43,501 orders.
It was labeled medium confidence because the cost inputs were estimates — the label is the feature, not a disclaimer.
Opportunity identified is not money banked. Anyone who blurs those two is selling you a demo.
Point it at a month you already understand.
Pick a period you know cold — a strong month, a bad one. We’ll read your own POS data and hand back what we find. If it can’t tell you things you already know, it hasn’t earned the ones you don’t.
Practical AI for small and mid-sized business. Live products, verified numbers, and your experts in charge of every decision.