One missing invoice

A freight forwarder sends an invoice for $2,180. It covers one container carrying three SKUs. Somebody files it in an email folder, means to allocate it later, and never does.

That's the whole event. One bill, one container, one afternoon of not getting around to it. If you run an Amazon business you have done this, probably this quarter.

Here's what makes it worth an article: that invoice does not stay missing in the place you lost it. It doesn't produce one wrong number in one report. It produces a series of confident, plausible, wrong answers in four other parts of the business — none of which look like they have anything to do with freight.

And the last thing it breaks is the thing that would have caught it.

01Where a missing $2,180 actually goes

Say the container held 12,000 units across three SKUs. That's about $0.18 a unit of freight that never made it into anyone's cost.

On a $7.00 unit, an $0.18 omission is 2.5%. Small. Now watch it move.

Nobody was careless. Each module did its arithmetic correctly on the input it was given. The system worked exactly as designed, and produced four bad decisions.

02Why gaps travel, and which ones don't

Here's the part worth internalising, because it lets you stop worrying about most missing data and start worrying about the right pieces.

Your operation isn't five separate systems. It's five sets of decisions drawing on a small number of shared facts. Advertising, inventory, finance and purchasing all read landed cost. Inventory, cash and advertising all read sales velocity. Nothing is as siloed as the software makes it look.

So there are two kinds of gap, and they behave completely differently:

• Shared-input gaps sit in a fact that many decisions read. They propagate — one omission, several wrong answers, in places that look unrelated to where you lost it.
• Leaf gaps sit in a fact only one thing reads. Which forwarder you used, the case-pack dimensions, your supplier's WeChat handle. Missing, annoying, contained.

This is also why the seepage is so hard to spot from inside. You don't notice a bad landed cost by looking at landed cost — nothing about $7.00 looks wrong. You notice it as an ad campaign that should be profitable and isn't, and you go and look at the campaign. Which is the one place the problem definitely isn't.

A propagated error always shows up somewhere other than where it was made. That's what makes it expensive to find and easy to misdiagnose.

03The six facts everything reads

Short list. If you know these six honestly, most of your numbers are trustworthy. If you don't know one of them, you have a leak, wherever it happens to surface.

• 01Landed unit costRead by: margin, break-even ACOS, restock ranking, inventory valuation, cash. The most widely-read fact you own — and the one most often left at factory price.
• 02Sales velocityRead by: days of cover, reorder date, ad budget, cash timing, forecast. Wrong if your data is stale, because a rate computed over days that haven't reported reads low.
• 03Stock — sellable, reserved, inboundRead by: reorder decisions, capital tied up, stockout risk. One number won't do; the three behave differently.
• 04Ad spend, per productRead by: per-SKU profitability, contribution, bid decisions, the P&L. Account-level ad spend is fine for the P&L and useless for everything else.
• 05Amazon fees, per saleRead by: margin at every level. Mostly measured for you on the settlement line — the ones that hurt are the account-level charges nobody assigns.
• 06Money out, with its dateRead by: cash forecast, commitment view, P&L. The date matters as much as the amount; a payment with no date is a payment you'll be surprised by.

Every one of these is read by at least three parts of the business. That's the definition that makes them worth your attention, and it's why "I'll sort the freight out at year end" isn't a filing decision — it's an operating decision about four other modules.

04The damage you don't see is the trust

The four wrong decisions in section one are the visible cost. They're recoverable — you overbid for a month, you restocked the second-best SKU. Annoying, survivable.

The real cost is the last step, and it's the one people underrate.

When your P&L doesn't tie to your bank account, you don't spend the afternoon reconciling. You do what everyone does: you stop consulting the P&L. You go back to the number you can see without arguing with it — Seller Central revenue — and you run the business on that.

Which is where you started, before you built any of this.

That's the loop worth naming. A gap doesn't just corrupt your numbers. It corrupts your willingness to look at them — and an operator who's stopped looking is guaranteed to accumulate the next gap too, because nothing will ever surface it.

It's the same reason a smoke alarm that chirps for a month ends up in a drawer. The failure isn't the false alarm. It's what the false alarm taught you to do.

05Five common gaps and what each one poisons

Two patterns run through all five. First, none of them announce themselves — the screen where the data is missing is usually the screen that still looks fine. Second, each one is a shared fact, which is exactly why the consequences land three modules away.

06The answer is not "track everything"

Having read the above, the natural response is to go and record everything. Don't. That's the failure mode on the other side, and it's just as common.

Perfect data has a cost, and it's paid in the currency you have least of: attention. A seller who spends six hours a week allocating every cost to two decimal places has bought accuracy in figures that were never going to change a decision, at the price of not looking at the ones that would.

There's also a quieter problem: the more chores a system demands, the sooner it gets abandoned. The most complete tracking system in the world is the one nobody has updated since March.

So rank the work honestly:

worth closing = (how many decisions read it)
              × (cost of getting those decisions wrong)
              ÷ (effort to close it)

Freight allocation scores high on all three: read by four modules, expensive when wrong, and closable in an hour a month at PO time. A per-unit allocation of your accounting retainer scores near zero on all three. Both are "missing data". Only one is worth a Tuesday.

07What a system should do with a gap it can't fill

Most tools handle missing data the same way: substitute something plausible and carry on. A default lead time of 90 days. Freight assumed at zero. Cost falling back to the last known price. The screen fills in, and nothing indicates that a decision is now resting on a number nobody entered.

That is the single worst option available, because it converts a gap you could have seen into a confident answer you can't. Three better behaviours:

• Refuse rather than default. "Freight not measured — EOQ unavailable" is more useful than an order quantity computed from a zero. An empty state that names what's missing tells you what to go and fix.
• Label the provenance. Every number should carry whether it was measured, allocated by a rule you chose, or assumed. Same figure, very different weight — and the assumed ones are your work queue.
• Disqualify the decision, not just the number. This is the one almost nobody does. If landed cost is incomplete, it isn't enough to footnote landed cost — the screen that ranks SKUs by return on capital should say that the ranking is unreliable, because that's where the damage happens.

That last point follows directly from the cascade. If errors surface somewhere other than where they're made, then warnings have to appear somewhere other than where the data is missing — at the decision, not at the field.

08Find your own gaps in twenty minutes

Take the six shared facts. For each one, three questions:

1. Where does this number come from? Measured from a document, allocated by a rule, or typed by someone once? If you can't answer in a sentence, that's the gap.
2. When was it last true? A lead time entered in 2024 isn't data any more. A velocity computed against stale sales isn't either.
3. What decision would flip if it were 20% off? This is the one that ranks the work. If nothing flips, leave it. If your restock order flips, close it this week.

Six facts, three questions, twenty minutes. Most sellers doing this the first time find two gaps they'd never have gone looking for, because both were surfacing as problems in a module that had nothing to do with the cause.

Then do the reconciliation you've been avoiding: does the profit your system reports match what actually arrived in your bank over the same period? Not to the cent — but every difference you can't name is a gap, and one of them is upstream of a decision you made this month.

Frequently asked