This is the weekly rhythm that makes every number trustworthy: count by voice, scan invoices as they arrive, and read your food cost while you can still do something about it. Get the loop right and variance, reorder points, and plate costs all fall in line. This chapter is counting and cost tracking in one place, because in practice they are one habit.
Chapter III
III
Run the loop
Count your walk-in in 10 minutes by voice
Counting the walk-in shouldn't take an hour. If yours does, you're counting the wrong things in the wrong order. Or you're writing instead of talking.
Our take
Most independent operators count their walk-in cooler once a week and hate every minute of it. We think weekly is right and one hour is wrong. A 2-door walk-in with about 80 items should take 10 minutes by voice, every Sunday night, in the same order, every time. The reason you count: to catch the 15 or so items that move enough money to matter, and to count them often enough that your numbers stay close to reality. Counting every can of tomatoes down to the exact unit is wasted effort.
How to count your walk-in
Start at the door, top shelf, left. Always. Stockcount records items in whatever order you say them, so the consistency is up to you: pick a route and walk it the same way every week. A predictable route is what keeps a count fast and complete.
Open the Chat tab, tap Speak, then say the item and the count. Something like "Romaine, two cases." Stockcount does the unit math for you, turning "two cases" into however you track romaine. The first time it hears a pack size it doesn't recognize, it asks how big the pack is, then remembers for next time.
Move shelf by shelf without backtracking. If you missed an item, keep going. You can add it at the end. Looping back doubles your count time and your error rate.
Weigh proteins, eyeball produce, count cases. Anything over $40/lb gets the scale. Anything under, just eyeball it. The math (see below) says perfection on cheap items is wasted minutes.
End at the door, bottom shelf, right. Each count shows up as a chip in the chat the moment you say it. There's no separate sign-off screen: name the last item and you're done.
If Stockcount questions a count, answer it. Don't recount everything. If you say a number more than 10 times higher than what's currently on hand, or you zero out an item that should be well stocked, Stockcount pauses to check before saving. Look at that one item, say yes or give the right number, and move on. It's a quick sanity check, not an accusation.
Why this works
Three things make voice counting faster than paper or tapping numbers into a screen:
Your hands stay free. A walk-in count is also a freshness check. You should be turning packages to read the dates. With a clipboard, you put it down and pick it up 30 times. With voice, you don't.
A fixed order is easy to keep. Inventory accuracy comes from counting the same things in the same order every time. Stockcount doesn't force a route on you, but voice makes the habit easy: you walk the walk-in with both hands free instead of hunting for the next line on a clipboard. Skipping around is how items get missed.
The app does the unit math, not you. "Two cases of romaine" is what your brain knows. "12 heads" is what your invoice knows. Stockcount translates between them. You stop doing mental arithmetic at 9pm.
It feels strange the first time. Talking to your phone in a cold room, alone, sounds odd until you've done it twice. By the third count it's faster than you'd believe, and the clipboard is the part that feels strange. Give it two full counts before you decide.
The math
Don't do this
What good looks like
A finished Sunday walk-in count for a 38-seat cafe usually lands at 78–84 items in 9–11 minutes, with zero or one count Stockcount stops to question. The whole count, freezer included, is done before 10pm and ready to shape the Tuesday produce order.
If your number is double that (20 minutes), your list of items is out of date. Walk it once with a manager and cut anything you don't stock anymore.
Inventory accuracy doesn't come from counting harder. It comes from counting the same things in the same order, every single time. That order is a habit — and the habit is yours to build.
Our take
Stockcount won't force a count route on you. It records items in whatever order you say them. That's honest, not a gap: a route is a physical habit — a path through cold rooms — and no setting can walk it for you. What Stockcount can do is hold the habit in place. Group your items by the area they actually live in, tell it how often each one needs counting, and the Today tab will show you what's due. The path you still walk yourself, the same way, every week. Set it up once with intent, and every count after that gets faster.
How to set your count route
Walk your space once and write the order down. Door of the walk-in, top shelf, left, then shelf by shelf. Then the freezer, dry storage, the bar, paper goods. Number the stops. This list is your route.
Tell Stockcount your storage areas. Describe your space to Stockcount in chat — walk-in, reach-in, dry shelves, the bar — and which items live where. It sets them up. Now your items are grouped the way your building is.
Count one area at a time. On the Stock List, filter to a single storage area, count it, move to the next. Filtering by area is how you keep a long count from turning into a scavenger hunt.
Tell Stockcount how often each item needs counting. Set a count cadence on each one: weekly for fast movers and expensive items, monthly for slow dry goods. You don't count everything every week — you count what's due.
Let the Today tab drive. It shows what's overdue and what's worth counting next, ranked by value and urgency. Open it, then count what it lists, in your route order.
Teach every counter the same path. A route only works if it's identical from person to person. Walk yours with each new counter.
Why this works
A count has two enemies: missed items and double-counted items. Both come from wandering. When you walk the same path every week, your hands and eyes learn it — you notice the gap where the romaine should be without checking a list. Consistency, not effort, is what makes a count trustworthy.
Cadences keep the route short. Counting every item every week is how a count balloons to an hour. Most dry goods barely move; counting them monthly is plenty. Let the fast, expensive items come up often and the slow, cheap ones come up rarely, and the weekly count stays lean.
Don't do this
What good looks like
A 38-seat cafe's storage areas in Stockcount mirror the building: walk-in, freezer, two dry shelves, the bar. Every item is assigned to one. Fast movers are set to weekly, dry goods to monthly. On Sunday the closer opens the Today tab, sees what's due, filters the Stock List to the walk-in, and counts it top-left to bottom-right — the same path the owner walked, the same path the new hire was taught. The count takes eleven minutes and nothing gets missed, because nobody is improvising.
Stockcount isn't a chatbot you perform for. During a count it's a second pair of eyes: it logs what you say, tells you what's left, and questions a number that looks wrong.
Our take
Stockcount earns its place during a count by being narrow. It does a few things well: it records every count you speak, it tells you what still needs counting, and it stops you when a number looks off. It is not a general assistant. Ask it about your inventory and it's fast and genuinely useful. Ask it to track a delivery or run a checklist and you'll just be disappointed. This guide is the line between the two.
How to use Stockcount during a count
Count by talking. Tap Speak, say the item and the amount. Each count lands as a chip in the chat the moment you say it — your running record, with no sign-off screen.
Ask what to count next."What should I count tonight?" or "What's overdue?" Stockcount answers with a list ranked by value and urgency. Good for starting, good for "am I missing anything."
Ask whether you've already counted something."Did I count the heavy cream yet?" Stockcount pulls your recent counts, so you don't have to remember or backtrack.
Answer its sanity checks. Say a number ten times higher than expected, or zero out a well-stocked item, and Stockcount pauses to confirm. That's the second pair of eyes working. Look at the shelf, answer, move on.
If you're a manager or owner, ask the deeper questions."What's the par on romaine?""When did we last count chicken thighs?""Explain the variance on heavy cream." Par is the amount you aim to keep on hand; variance is the gap between what you used and what your sales expected. These questions pull current stock, par levels, and cost history — manager-level answers, in the same conversation as your count.
Why this works
Stockcount is useful mid-count for one structural reason: it's already in the loop. Your counts go through the chat, so asking a question isn't a context switch — it's the same conversation. You're not opening a report. You're talking to the thing that just logged your last twenty counts.
And narrow beats broad here. A chatbot that pretends to know everything makes you guess what's real. A tool that does five things — log, list, recall, check, explain — lets you trust every answer, because you know what's behind it.
Don't do this
What good looks like
Mid-count, Sunday night. The closer is halfway through the walk-in and can't remember whether the oat milk got counted. "Did I count oat milk?" — Stockcount says yes, two minutes ago. A little later it pauses on a case of tomatoes counted as "forty" when last week was four; the closer looks, sees it really is four cases, corrects it, and moves on. At the end, "What's still overdue?" turns up two items in dry storage that got skipped. Total interruptions: under a minute. The count is complete, and nothing was entered twice.
The single best habit in a Stockcount workflow is scanning invoices the day they arrive. Two minutes at the loading dock saves you two hours at month-end.
Our take
Most food-cost pain is really an invoice problem. Operators count weekly but only deal with invoices once a month, so the costs in the app run 3–4 weeks behind reality. Scan the invoice the moment the driver hands it over, and your dish costs reflect today's prices by tonight. The math stays honest in real time.
How to scan
Open chat (on your phone or the web).
Tap the attachment icon, then take a photo or upload the PDF. It doesn't need to be a perfect photo. Stockcount handles glare and creases. What matters is getting the whole invoice in the frame.
Give it a few seconds. Stockcount reads the invoice, works out which vendor it's from, and matches each line to your inventory, showing its progress as it goes.
Answer any questions. Stockcount only asks about things it can't sort out on its own — usually an item that's not in your catalog yet ("Burrata isn't in your list. Add as new?"), or a line it couldn't read clearly. Everything it's sure about, it fills in for you.
Confirm. The invoice is saved, and the new costs flow through everywhere.
What happens to your costs
Every ingredient on the invoice gets a fresh cost — a blended average of what you already had on hand and what you just bought, weighted by how much of each. No "cost updated from invoice X" notification pops up. The new cost just becomes current everywhere it's used: dish costs, your food-cost totals, every number built on them.
Open any ingredient or recipe later and you'll see today's price already in it. The whole system moves together.
The receipts review queue
Sometimes a scan turns up a vendor product Stockcount hasn't seen before, or one where it isn't sure of the pack size (how many units come in a case). That product lands on the Receipts page on stockcount.io, which only managers and owners can open. Each row shows:
The product, with the vendor's own wording for it
The ingredient it's linked to
The pack size Stockcount worked out (for example, 12 each per case), and the last price per unit
A confirm button
Confirming locks in the pack size, so future invoices from that vendor match automatically. Clear the queue when you have a few minutes. Until you do, the cost on those products is Stockcount's best guess.
Don't do this
What good looks like
A typical week for a mid-volume cafe: 4 vendor invoices scanned as they arrive, 2–3 review queue items sorted out within 24 hours, nothing left over at the end of the week. Sunday close-out has nothing to do, because the work already happened at the dock.
If your queue has 30+ unresolved items at month-end, the habit hasn't taken hold yet. Scan one invoice a day for a week. That's enough to build the muscle.
Where this connects
Receiving a delivery: the moment the driver shows up, broader than just the scan.
The driver's double-parked, the invoice has 40 lines, and he needs your signature now. That signature is the last moment you can still push back. Use it.
Our take
Receiving is a money moment disguised as a chore. The produce truck shows up at 6am, the driver is three stops behind, and the easy thing is to sign the sheet and stack the boxes. But your signature says "everything on this invoice arrived, in good shape, at these prices" — and once you've signed, every argument is over. Receiving deserves ten honest minutes: count the truck against the invoice before you sign, then scan the invoice into Stockcount the same morning, while the short case is still fresh in your memory. Do that and a delivery never becomes a month-end mystery.
How to receive a delivery
The scenario: Sysco rolls in Tuesday at 6am with the produce order. Here's the loop, start to finish.
Count the truck against the invoice before you sign. Match cases to line items. This is the only moment the driver, the product, and the paperwork are all in one place. Five minutes here saves an hour of phone calls later.
Mark anything wrong right on the paper invoice. Three cases of romaine short — cross it out, write "3," have the driver initial it. Damaged, wrong item, melted — note it. Then sign.
Scan the marked-up invoice into chat the same morning. Open chat, tap the attachment icon, photograph the invoice. Glare and creases are fine. Stockcount reads every line, works out which vendor it's from, and matches each line to your items.
Answer only what Stockcount asks. It checks with you on the things it can't resolve — an item not in your catalog yet, or a line it couldn't read. Everything it's sure of, it fills in.
Correct any line that doesn't match what arrived. This is the step that matters. If the invoice billed six cases of romaine but three showed up, the line should say three. If Stockcount asks you about that line, give it the real number there. If it doesn't ask — because the invoice said six clearly — open the receipt on your phone afterward and correct that line. What you "bought" should always match what actually came off the truck.
Confirm. The receipt is saved. Your costs update across every dish that uses those ingredients.
Record any credit the vendor agreed to. If Sysco is crediting you for the short cases or a flat of bruised tomatoes, tell Stockcount in chat: "Sysco credited us $96 for three short cases of romaine." That credit lowers your food cost, the same way the purchase raised it.
Why this works
Here's the part most operators get wrong, so read it twice: scanning a delivery records what you bought and what it cost. It does not change the on-hand number on your Stock List.
Those are two different jobs. A delivery answers "what did I buy, and what did it cost me?" A count answers "what is physically on my shelf right now?" Stockcount keeps them separate on purpose. If a delivery automatically bumped your on-hand count, then a delivery and a count on the same day would fight each other, and you'd never know which number to trust. So the rule is plain: invoices update your costs, counts update your stock. You need both, and you do them as separate habits.
That's also why correcting the short case matters. If you let the invoice's "six cases" stand when three arrived, Stockcount believes you bought $96 more romaine than you did. That phantom $96 doesn't vanish. It resurfaces weeks later as a food cost higher than it should be, with no obvious cause. Correcting the line at receiving is how you keep that from ever becoming a mystery.
And the signature: once you've signed, the conversation is over. The driver's company will stand behind that piece of paper. Counting the truck first isn't about distrust. It's about making sure the paper and the pallet say the same thing before you make the paper official.
The math
Don't do this
What good looks like
Tuesday, 6:10am. The Sysco driver is in a hurry, as always. The receiver works down the invoice with a pen, cases stacked by line item. Forty lines, two problems: three cases of romaine never made the truck, and one flat of tomatoes is bruised through. Both get crossed out and corrected on the paper, the driver initials them, the invoice gets signed.
By 6:25am the marked-up invoice is photographed into chat. Stockcount reads it, asks one question about a new specialty item, and matches the rest. The receiver corrects the romaine line to three cases and confirms. A quick message — "Sysco is crediting us $96 for the short romaine and $18 for the bruised tomatoes" — and the credit is logged.
Total time: under fifteen minutes, most of it spent counting the truck. When Sunday's close-out comes around, there's nothing to untangle. The delivery already told the truth.
If your Sunday close-out involves Excel formulas and a stack of unscanned invoices, you're doing the bookkeeper's job. The point of Stockcount is that the close is already done by Sunday: the only work left is reading the answer.
Our take
Most operators find out their food cost on the 15th of the next month, when the bookkeeper sends over the P&L (the profit-and-loss report). That's six weeks late on a number that should reset every week. Stockcount is built on the opposite idea: counts and invoices go in as the week happens, so Sunday night is just reading and reacting, not data entry.
The whole routine, done right, takes 20 minutes.
The 20-minute close-out
A complete week, step by step:
Catch up any unscanned invoices. Open chat and upload anything that piled up. (Aim for zero. See Scan an invoice.)
Clear the receipts queue. Open the Receipts page on stockcount.io and sort out anything Stockcount couldn't match cleanly. It should be 0–3 items if the daily scan habit held.
Open the Analytics page. Read net COGS (your food cost), the COGS %, and the trend.
Open the Variance breakdown from the Analytics page (the variance card links straight to it) — this is where what you used and what you bought don't line up. Scan for anything off by more than ±10%. Ask Stockcount about it: "Explain the variance on tomatoes this week."
Tell Stockcount your sales."We did $14,200 in sales last week."
That's the close. Anything left for month-end is on the accounting side, not the operating side.
Your money picture, in plain English
The Analytics page leads with two numbers, a trend, and a coverage badge:
Net COGS: last week's food cost, in dollars. (COGS stands for "cost of goods sold" — restaurant-speak for what your food and drink actually cost you.)
COGS %: net COGS ÷ revenue — your food cost as a share of sales. It shows a dash until you connect your sales (through Square) or type in a sales total yourself.
COGS trend: the line underneath, tracking the last several weeks, so you can see which way things are heading — not just where they sit today.
Coverage badge: something like "81% counted · 19% imputed." This is Stockcount telling you straight how much of the number it actually measured versus had to estimate. ("Imputed" just means estimated.)
Reading variance without panicking
Variance is the gap between what your counts say you used and what your invoices say you bought. Some variance is normal. A big one is a question, not an accusation.
Stockcount splits every ingredient's variance into two parts:
Quantity variance: you used more (or less) than expected. Usually portioning, waste, theft, or a count error.
Price variance: the cost per unit moved. Usually a supplier change you didn't notice.
Anything off by more than ±10% on an item you use a lot is worth a look. Stockcount has the full context. Ask "Explain the variance on chicken thighs this week" and it pulls the cost history, recent waste, and any comps (meals given away free) into one answer. Both the variance and COGS pages can export to a CSV file (a spreadsheet) if you want the raw rows somewhere else.
Don't do this
What good looks like
A 38-seat cafe, Sunday 8:30pm: the count finishes at 8:41pm, the receipts queue clears at 8:45pm, the Analytics page is open by 8:47pm, two variance items are looked into by 8:50pm. Total: 20 minutes. Ready for Monday with food cost % locked at 28.4% (last week was 29.1%, three weeks ago was 31.2% — a steady drop that started when olive oil prices settled down).
That's the work. No spreadsheet was opened. No bookkeeper was called. The whole point of running the loop weekly is that nothing is a surprise.
What's not here yet
No PDF export of reports. Export is to CSV (spreadsheet) only.
No real-time alert when food cost spikes. There's a weekly COGS notification you can switch on, but a mid-week jump you'll catch at Sunday close.
No combined close-out across locations. Each location closes on its own.
If any of those are blocking your workflow, ping us. We move on real operator demand.
Why is my food cost 38% when I think it should be 30%
You ran the numbers and they say 38%. Your gut says 30%. Until you know which one is lying, you can't fix anything.
Our take
A food cost that lands 8 points above where you expect it is almost never one big leak. It's five small ones, and four of them are invisible until you go looking. The worst move is to react to the 38% by cutting portions across the board or leaning on the kitchen. You don't know where the money is going yet. The number is a symptom, not a diagnosis. This guide is the diagnosis: the five places the 8% hides, the order worth checking them in, and how to read what Stockcount already tracks so you fix the real leak instead of guessing.
How to track down the 8%
Work these in order. The first three rule out a number that's wrong for boring reasons. The last three find the actual money.
Check the measurement before you trust it. Open the Analytics page and look at the coverage badge under net COGS (your food cost in dollars). It reads something like "82% counted · 18% imputed" — imputed meaning estimated. If that badge is amber or red, part of your 38% is a guess, not a measurement. Count one more shelf each week until it turns green, then trust the number.
Catch your invoices up. Stale prices are the single most common reason the app and your gut disagree. If invoices have been sitting in a folder, scan them now. A dish priced on three-week-old costs is fiction, and so is the food cost built on it.
Make sure your recipes are real. Stockcount works out what you should have used by multiplying your sales against your recipes. If a best-selling dish has no recipe, or a prep recipe still has its yield — the number of servings it makes — set to 1, that math is wrong before it starts. Open Recipe Coverage to see which menu items still have no recipe, and build your top sellers first.
Open the variance breakdown and find the biggest dollar gaps. Variance is the gap between what your invoices say you bought and what your sales say you should have used. Skip the percentages for a moment and look at the lines with the largest dollar gaps. The top five are where your 8% lives. Everything under them is noise.
Read the price variance — that's your vendors. Stockcount splits each line into price variance and quantity variance. Price variance means the cost per unit moved. This is vendor creep: produce crept up forty cents a case, dairy moved twice, and nobody emailed you. Ask Stockcount in chat: "Explain the price variance on heavy cream this month."
Read the quantity variance — that's your kitchen. Quantity variance means you used more than your sales predicted. It has three usual causes: heavy hands on the line, waste and spoilage, and comps (meals given away free) or staff meals nobody recorded. Ask Stockcount to explain the worst one. Six comped entrees and thirty staff meals aren't theft and aren't a mystery — but only if someone logged them.
Why this works
Stockcount holds two numbers for every ingredient and every dish. One is theoretical usage: what your sales say you should have gone through, based on your recipes. The other is actual usage: what your counts and invoices say you really went through. Food cost is honest only when both are built on honest inputs — current prices, real recipes, and counts frequent enough that little is estimated.
That's why the 8% almost always splits across the same five places:
Stale costs. Invoices entered weeks late, so dishes are priced on last month's market.
Phantom recipes. Dishes with no recipe, or a prep recipe whose yield was never set, so theoretical usage is wrong.
Price creep. Vendors raised prices a little at a time, and small raises compound.
Heavy hands. The line over-portions. A grain bowl built 15% bigger than the recipe is a 15% loss on every one sold.
Unlogged giveaways. Comps, staff meals, and waste are real food cost. Unrecorded, they look identical to theft.
None of the five is dramatic. That's exactly why they survive. A 38% food cost isn't a fire. It's five slow drips, and the only way to find a slow drip is to look in all five places on purpose.
The math
Don't do this
What good looks like
A 45-seat cafe opens the Analytics page and sees 38.2%. Instead of calling a meeting, the owner walks the six steps. Coverage is amber at 71%, so the real number is softer than it looks. Two weeks of unscanned invoices go in. The marinara prep recipe still has its yield at 1, which had been quietly halving the cost of every pasta dish — fixed. The variance breakdown shows $380 of price variance on dairy (the vendor moved twice) and $710 of quantity variance on the grain bowl (the line was building it big).
Nothing here was theft. Nothing here was one disaster. Six weeks later — coverage green, invoices current, recipes real, the grain bowl re-portioned, the dairy line renegotiated — food cost reads 31.4%. The 8% wasn't a mystery. It was five ordinary leaks that nobody had looked for in the same hour.