The numbers your Amazon software invented

Your inventory screen says: Order by 14 March. 1,400 units. A date and a quantity. Clean, specific, actionable — the kind of output you pay software for.

Now ask it one question. Where did the lead time in that calculation come from?

In most tools, the answer is that it came from nowhere. It's 90 days, because 90 days is what the field was set to when the account was created, and nobody ever changed it. Your actual supplier takes 58.

The date is wrong by a month. Nothing on the screen suggests it might be.

The problem isn't that your software doesn't know things. It's that it doesn't say so.

This is the most common defect in Amazon analytics, it's in nearly every tool on the market including the expensive ones, and almost nobody talks about it — because talking about it means admitting to it.

01The plausible default

Here's the pattern. A calculation needs an input. The input isn't available — you never entered it, or the data doesn't contain it, or the report that would have carried it wasn't uploaded.

The software has three options: leave the output blank, say what's missing, or substitute something reasonable and carry on.

It almost always picks the third. And the substitute is well chosen — 90 days is a defensible lead time, 15% is a defensible TACOS, zero is a defensible freight cost if you have no freight records. Nothing looks broken. The screen fills in. You act on it.

Call it what it is: a plausible default. A number that exists to make a screen look complete, presented identically to a number that came from your business.

The dangerous part isn't the substitution. It's the identical presentation. Your measured velocity and an invented lead time sit in the same row, same weight, same font — and the output that combines them inherits the confidence of the better half.

02Why every tool does this, including the good ones

It's worth being fair about the motive, because it isn't malice and it isn't laziness.

It's the free trial.

A new user connects their account on a Tuesday evening. If the software is honest, most of the screen is empty — no costs entered, so no margin; no purchase orders, so no lead time; no freight records, so no landed cost. It's an accurate picture of what the tool knows about them, and it looks like a product that doesn't work.

Fill those gaps with plausible defaults and the same user sees a rich dashboard within ninety seconds. Charts, percentages, recommendations. That converts. Every product team that has ever run the experiment has found this.

So defaults are a growth decision, made rationally, by people who aren't trying to mislead anyone.

They just optimise the first five minutes of the relationship at the expense of every decision after it. And the cost lands on the customer who stayed — the one placing a $40,000 order against a lead time nobody measured.

03Six tells: how to find the invented numbers in the tool you use now

You don't need access to anyone's source code. Invented numbers leave fingerprints, and you can check for all six in about ten minutes.

• 01Suspiciously round values90 days. 30%. 15%. Real measurements are 58 days and 27.3%. A round number in a field you never filled is a default wearing a disguise.
• 02The same value on every SKUSort by lead time, or by any cost assumption. If forty products share one figure, that figure describes the software's configuration, not your supply chain.
• 03An output that survives a missing inputThe strongest test. Delete a cost, or check a product you've never entered costs for. If a margin still appears, the tool manufactured one — and it will do the same everywhere else you're not looking.
• 04No date on the dataAsk what day the numbers run through. If no screen tells you, rates are probably computed against the calendar rather than the data — which quietly deflates every velocity when reports run late.
• 05Nothing ever says "unknown"Click through a full session. If not one screen admits to missing something, that isn't completeness. Nobody's data is complete. It means the gaps are being filled where you can't see them.
• 06Precision it can't possibly haveA margin of 27.43% built on a freight figure that was allocated by a rule and a return rate from a historical average. The decimals are decoration, and they're doing persuasion work the data can't support.

Run these on whatever you're using today. Most sellers find at least three, and the third one — an output that survives a missing input — is usually the moment the exercise stops being academic.

04Why a blank is worth more than a guess

The instinct is that an approximate answer beats no answer. For this class of problem, it's backwards, and the reason is worth being precise about.

A blank field is a known unknown. It costs you five minutes of irritation and points at a specific, cheap action: go and find the lead time. You do it once, and every decision downstream improves permanently.

A default is an unknown unknown. It costs you nothing today and takes away your ability to tell which of your decisions were made on real information. That's not a smaller version of the same problem. It's a different problem, and a worse one, because there's no signal to act on — you can't fix what never announced itself.

And the errors don't stay put. A lead time that's a month wrong doesn't just move a date. It moves your safety stock, your reorder quantity, the cash you commit and when you commit it — and when the stockout arrives you'll diagnose it as a demand forecasting problem, because that's where it surfaced.

05What a screen looks like when it admits what it doesn't know

The objection to all this is that honesty makes for an ugly product. It doesn't. It makes for a longer one — and the extra length is the part you needed.

Same SKU, same moment, two ways of showing it:

The left card is more confident and less useful. The right one is the same decision with its working shown — and it tells you which part to distrust and which part to act on anyway.

Three behaviours produce that second card, and they're not expensive to build. They're just unflattering during a trial.

• Refuse rather than default. "Lead time not measured — add a received date to any past order" beats a 90 that nobody chose. An empty state that names what's missing is a to-do list, and a useful one.
• Label the provenance. Measured, allocated, or assumed, on the number itself. Same figure, three very different weights — and the assumed ones are exactly where your setup work should go next.
• Warn at the decision, not at the field. The hardest and most valuable. A footnote on the margin field is useless, because nobody visits the margin field. The warning has to appear on the card where you're about to place a $40,000 order.

Note what the honest card does not do: it doesn't refuse to help. It still gives a date and a quantity. It just declines to pretend that all four inputs are equally solid — and in doing so, it tells you that the freight problem affects your margin but not your reorder date, which is the single most useful sentence on either card.

06"I don't want caveats. I want a number."

Fair, and the honest card gives you one. Order by 14 March, 1,400 units. It's right there, in the same place, in the same size.

What it adds is the answer to the question you'd otherwise have to ask a human: how much should I trust this? For a routine restock you'll never read past the headline. For the order that's twice your normal size, you'll read every line — and that's precisely the order where a 90-day default costs you a quarter.

The alternative isn't a simpler screen. It's the same screen with the doubts removed and the risk left in.

There's a version of this in every profession that handles consequences. A lab result carries its reference range. A surveyor's report says which measurements were taken and which were estimated. A structural engineer signs off on load assumptions, not just the load. None of them find this burdensome; they find it the minimum standard for advice someone will act on.

Software that tells you to spend $40,000 should be held to the same bar.

Frequently asked