The examples and architectures in this series are my personal educational, non-commercial experiments; they are not an offer of IT or consulting services.
A monthly average is a very polite way of not mentioning something. Post #16 argued that you can't see an odd payment until you know the normal one. Margin has the opposite problem: the average is so normal that it covers whatever sits below it.
The month below is a composite. The product lines, customers, revenue, margins and discounts are invented round numbers, illustrative only. They don't describe a real company, price list, customer or margin.
One month, three product lines
The report shows 19%, and that figure is arithmetically correct. It answers "how is the company doing?", not "is any part of it losing money?" Here is the same month split by line.
| Line | Revenue | Margin | Contribution |
|---|---|---|---|
| A | 500 | 30% | 150 |
| B | 350 | 15% | 52.5 |
| C | 150 | minus 10% | minus 15 |
| Total | 1,000 | about 19% | 187.5 |
Revenue and contribution are in made-up thousands. Line A carries the average, Line B is thin, and Line C is giving back part of what the others earn.
Illustrative composite from Post #17. The product lines, margins and amounts are made-up numbers and do not describe a specific company, product, customer or price.
What stacked up on Line C
There was no single bad decision behind Line C. In many teams the negative sign arrives as several small items, each of which is explainable on its own. In this illustration the largest one is six points.
| Item | Effect on Line C margin |
|---|---|
| Margin on product cost alone | plus 10 points |
| Discount for the largest customer | minus 6 points |
| Returns | minus 4 points |
| Freight | minus 5 points |
| Payment fees | minus 2 points |
| Sales mix: more of the lower-priced pack | minus 3 points |
| Margin on Line C | minus 10% |
The same arithmetic works per customer. In this illustration the largest customer of Line C buys 40% of the line at minus 16%, while the other customers sit around minus 6%. The line total of minus 10% hides both.
The sign depends on what you count
Margin on a product is a decision about which costs belong to it, not a fact lying in the ledger. The same Line C is plus 10% if you count only product cost, and minus 10% once discounts, returns, freight and payment fees are inside. A third version, with allocated overhead, would be lower still. None of them is "the true one". What matters is picking one basis and using it for every line, because otherwise you are comparing different things.
| What it looks like | What's actually true |
|---|---|
| Blended margin is 19%, so everything is fine | One line of three is below zero, and the average can hide it |
| Margin is a number | Margin is a number plus a decision about which costs count |
| A line below zero comes from one big mistake | It often builds up from several small items, none of them alarming alone |
| The average temperature of the building is comfortable | One room is freezing and another is overheating. The report is a very good thermostat for the building and a poor one for either room |
The same question as in Post #16
Post #16 asked "compared with what?" for a payment: you need the supplier's own normal. Here the question is the same, with the average in the role of the normal. Compared with the blended 19%, Line C looks like a small drag. Compared with zero, it is a loss. The check has to look at the part, not only at the whole.
Post #8 applies here too. A check that fires on every small dip teaches people to stop reading it. The threshold is a decision, and the reason attached to each alert is what makes it usable.
What a margin check has to contain
| Element | Why it matters |
|---|---|
| Margin per product line and per customer, not only in total | The total cannot show a part below zero |
| One cost basis for every row | Otherwise the rows aren't comparable |
| Month by month, not one snapshot | It shows the drift towards zero before the line crosses it |
| An alert when a line crosses zero or an agreed threshold, with the reason attached (discount, freight, returns, mix) | An alert without a reason is not actionable |
| A person confirms what happens next | The check changes no price and stops no sale |
One more thing sits underneath the table. Margin is only as current as the slowest unconfirmed document. The discount, the freight charge or the return often sits on an invoice or a credit note that nobody has confirmed yet, so the report looks complete and isn't.
The average answers "how are we doing?". The question that costs money is "which part is losing?"
A test you can run this week
Take last month. Calculate the margin on your ten largest customers or products, on one cost basis, and look at the lowest three. Did you know their sign before you calculated it?
I haven't put a script behind this one either. It is a table and a decision about the cost basis, and both are worth making before any tool is involved.
Follow the series
Post #18 looks at the gap between the tools: the invoice that is correct in every field and still waits, because nobody has said who confirms it, and what stays unsolved even when Excel, the ERP and an AI assistant are all in place.