You are about to reorder. Two thousand units, same product as last time, the one that works. You know your numbers on this one — 26% cost of goods, healthy margin, it has been carrying the account for a year and a half.
You place the order. That decision was made with a number you calculated eighteen months ago, and it was correct when you calculated it.
01Everyone does the math. Almost nobody does it twice
Before you bought this product the first time, you did the work properly. You got a quote from the supplier. You got a freight quote. You looked up the duty rate. You checked the FBA fee in the calculator. You put it in a spreadsheet and it came out at $8.84 a unit landed, on a $34 product — 26% of price.
That was real work, done carefully, and it was true.
Here is what nobody ever schedules: the day you go back and replace those quotes with what you actually paid.
The freight quote was $1.35 a unit. The invoice, when the container cleared, was $2.15. The supplier raised the unit price on the second order and again on the third — $6.90 became $7.60. The duty came in at $0.71, not $0.59. Amazon raised the fulfilment fee. The product slowed slightly over summer and started paying more storage. Returns settled at around 4%, which was never in the spreadsheet at all, because on the day you built it you had no returns.
None of that was hidden. Every one of those numbers arrived in your inbox, in an invoice, with a date on it. They just never went back into the 26%.
Your costing was a forecast. You have been using it as a record. Nobody ever decides to make that substitution; it happens quietly, and it happens to everyone.
A seller with one product has one spreadsheet, built before launch, that slowly becomes fiction. A seller with forty products has forty costings done at forty different moments by forty different versions of themselves, and no way to tell which ones still hold. Same disease. The second one is worse, because it is invisible at scale.
02What the number actually is
Here is the product as costed, and the same product as it actually runs.
| What you calculated | Per unit |
|---|---|
| Price | $34.00 |
| Landed cost | $8.84 |
| Amazon referral | $5.10 |
| FBA fulfilment | $5.45 |
| Storage | $0.35 |
| Advertising at 15% | $5.10 |
| Left over | $9.16 — 27% |
| What it costs now | Per unit |
|---|---|
| Price | $34.00 |
| Landed cost | $10.46 |
| Amazon referral | $5.10 |
| FBA fulfilment | $5.78 |
| Storage | $0.61 |
| Returns at 4% | $0.95 |
| Advertising at 18% | $6.12 |
| Left over | $4.98 — 15% |
Twenty-seven to fifteen. Not a rounding difference — nearly two thirds of what the product was leaving you.
Every one of those changes reduced what the product leaves you. That is not a law of nature, and it would be easy to overclaim here. Costs do fall: freight rates drop, suppliers cut the unit price when you order more, a fee occasionally moves the right way.
But an un-revisited costing has a bias built into it, and the bias is structural. The things you leave out are always costs. Nobody ever forgets to include a discount. Returns, storage on a product that slowed down, a fee change, the gap between a freight quote and a freight invoice — these are the items that get discovered late, and every one of them is a subtraction.
And when the drift runs the other way, it is expensive in a different currency. If your landed cost quietly fell to $7.90 and you are still planning at $8.84, you are underbuying a product that could absorb more money, and putting that capital somewhere worse. A stale number does not only make you overcommit. It also makes you miss.
Only one of those two mistakes ever shows up in your bank account. So only one of them ever gets noticed.
03The order you just placed
You placed that order expecting it to tie up $17,680 — two thousand units at $8.84.
It ties up $20,920. Two thousand units at $10.46.
So the first surprise is $3,240 of capital you had not planned to spend, on an order you have already committed to. And you will find that one out, eventually, when the supplier’s invoice arrives and the number is simply larger than the one in your head. Annoying, survivable, too late to change.
The second surprise you will never find out at all.
At the number you were working from, each unit leaves you $9.16 after everything that varies with the sale — about $18,300 across the order. At the number it has actually become, each unit leaves $4.98 — about $9,960. Same two thousand units, same $34 price, a little over half the return.
The order is not a disaster. The product still makes money. But you committed more capital than you meant to, for a little over half of what you expected back, and you decided it on the strength of a spreadsheet you last opened before your first container shipped. It has quietly become the second-best thing you could have done with the money.
04The same wrong number, at four different moments
This is the part that matters, and it is not really about accuracy. It is about when.
Before you place the order. Everything is open. Order 900 instead of 2,000 and watch how it sells at the real margin. Go back to the supplier with three orders of history and ask for a better price. Move to $39 and see what conversion does. Skip it entirely and put the money into the product making 31%. Every option is available, and the number has cost you nothing yet, because it is still a decision.
After you order, before it lands. Narrower, but real. You cannot un-buy the units, but you can restructure the balance payment. You can plan the cash around it instead of being surprised by it. You can start the price test now rather than in four months, so the stock lands into a better number.
After it lands. You own two thousand units. Price and advertising are what you have left. You can steer, barely. Mostly you sell through and accept what it gives you.
After it sells through. Nothing. There is no decision left to make.
When the consequence finally arrives, it does not arrive labelled. It shows up in November as a month that felt tight, a supplier you had to ask for a few more days, a reorder you skipped because the cash was not there. Nothing on any screen says this is the March order, costed in a spreadsheet from two years ago.
So you do not even get the lesson. You get a bad month with no cause attached — which means you will do it again, with the next product, using the next number you calculated once and never revisited.
05So how accurate does a number have to be?
Not perfect. That is the part most advice about knowing your numbers gets wrong, and it is why most sellers ignore the advice.
A cost does not need to be right. It needs to be right enough for what you are about to do with it.
The same $8.84 is three different things depending on the decision in front of you. Reading last month’s performance? A stale cost is a minor annoyance; the trend is the same either way. Deciding whether to run a lightning deal? It matters, a bit. Committing twenty-one thousand dollars of inventory for four months? It is the entire input, and being 18% wrong about it is the difference between the best and the second-best use of your capital.
One number. Three completely different tolerances.
The useful question was never whether your data is clean. It is whether this number is good enough for this particular decision — and that is only answerable if you can see how old it is and what it is made of. A figure with no history attached cannot be judged at all. You either trust it completely or you do not, and both of those are guesses about a guess.
06The work you are only postponing
It is more work than typing 26% and moving on. That is true, and worth saying plainly rather than pretending otherwise.
But the work does not disappear when you skip it. It relocates.
It turns up at year end, when your accountant wants cost of goods sold and you are rebuilding it from a folder of PDFs and a bank statement. It turns up at the reorder, where it becomes a guess with twenty-one thousand dollars attached. It turns up the morning your supplier asks for the balance and you genuinely do not know whether the money will be there before Amazon’s next payout.
Two minutes when the invoice arrives, or two hours in April from memory, under pressure, with the decision already made. Same work. Only the price changes.
And it is smaller than it sounds, because most of it is not yours to do.
- What you recordthe purchase order already open in front of you
- Which fieldssupplier, quantity, unit price, freight, duty, payment dates
- How oftenonce per order, a few times a month
- What gets measured for yousupplier lead times, from the dates on your own orders
- Also measuredhow long Amazon actually takes to pay you, from your own settlements
- Also measuredhow your payments are usually structured, from your own history
You are not being asked to build a database. You are being asked to finish a calculation you already started, and to stop discarding the confirmation you already received.
07Why a good tool warns you instead of stopping you
There is an obvious fix software could apply here: refuse. Make the cost field mandatory. Block the reorder screen until the freight invoice is entered. Stop the seller from making the mistake.
That is the wrong answer, for three reasons.
You did not do anything wrong. The 26% was honest work, correctly done, and accurate for months. Blocking treats a good-faith estimate as an error. A label accuses nobody — this cost is from March, before two freight invoices is a fact, and you can hear it without having to defend yourself.
Only you know which decisions are worth two minutes. The same number is trivial in one place and decisive in another, and software cannot tell those apart. It can tell you how old a figure is and what is about to rest on it. The judgement is yours, and it should stay yours.
A required field does not produce a true number — it produces a number. Every mandatory cost box in every tool has been cleared by somebody typing whatever gets them to the next screen. That figure then sits in the database looking exactly like a measured value, indistinguishable from one forever.
Blocking does not manufacture accuracy. It manufactures compliance, and compliance and accuracy look identical in a column. So the gate is not the cure for numbers nobody verified. The gate is where they come from.
The right behaviour is narrower and quieter. Never ask at signup. Never ask for all of it. Ask once, attached to a decision, at the moment the answer stops being free: can you afford this order? Not yet answerable — the freight on your last container was never recorded, and it is 18% of this month’s outflow.
One field. One minute. Then the numbers move in front of you: the landed cost, the break-even, the trough, two campaigns that flip from scale to cut. That is the only version of this anyone ever does twice — the one where the work pays out the same minute it is done.
Frequently asked
How often should I update my cost of goods?
Not on a schedule. When the real figure arrives — the supplier invoice, the freight bill, the duty entry — and before any decision that depends on it. In practice that means once per purchase order, at the point the paperwork lands.
What costs do sellers most commonly leave out?
Freight variance between the quote and the invoice, duty, returns, storage on anything that has slowed down, and fee changes. They share one property: each one makes the product look better than it is when it is missing.
My estimate is close enough. Does the difference really matter?
It depends entirely on what you are about to do with it. The same 18% error is a rounding issue when you are reading a report and a five-figure misallocation when you are sizing an order. Ask what the number is for before asking how good it needs to be.
What if I do not have the freight invoice yet?
Then the figure is an estimate, and it should say so. An estimate marked as an estimate is a useful number. An estimate that looks like a fact is the problem this whole article is about.
Do I have to enter every purchase order?
No. You will get a number either way. You will also get told how much of it you can rely on — and if you are about to commit twenty thousand dollars on the strength of it, you will probably want the two minutes first.
Why is my margin lower than my spreadsheet says?
Usually because the spreadsheet holds what you expected to pay rather than what you paid. Compare the four inputs that move most — unit price, freight, fees and returns — against the last invoice for each, and the gap will normally be sitting in one or two of them.
Final thoughts
None of this means you need a clean system before you are allowed to see anything.
Type 26%. Get your view. Nothing is blocked, nothing is withheld, the software works. What you should be able to see is which figures were measured and which were estimated — and which of the estimates is about to be spent.
Because the question was never whether your number was right. It was right when you calculated it. The question is whether you find out it has stopped being right while there is still something you can do about it.
Every tool can tell you what you told it. The useful ones tell you which parts were measured, which were estimated, and which estimate is about to decide something.
A number does not become dangerous because it is an estimate. It becomes dangerous when you forget that it is one.