Why do our reports disagree?

Almost never because somebody has made a mistake. Ask three people in your business for last month’s sales and you will usually get three numbers, all of them correct. Each is answering a slightly different question: sales of what, counted when, net of which returns, from which copy of the data. The disagreement is a definition problem wearing an arithmetic costume.

That matters because of how it gets treated. A disagreement that looks like an error gets handed to whoever is best with a spreadsheet, and they will find nothing, because there is nothing of that kind to find.

How do you prove this to yourself?

Pick a figure everybody argues about and a period that is closed. Ask three people who produce numbers for it separately, on the same morning, without telling them why or that you are asking anyone else.

Write down the three answers. Then go back to each person and ask a second question, which is the one that matters: what did you include, and what did you leave out?

You will almost never be shown an error. You will be shown three people answering three different questions correctly, and none of them will have thought the question was ambiguous, because in their part of the business it never has been.

What does “sales” actually mean?

This is the largest single cause, and the one nobody looks for.

Take one word and count the decisions hiding inside it. Is a sale counted when the order is placed, when the money arrives, when the invoice is raised, when the goods leave the godown, or when the customer has them? Is it before discount or after? Before returns or after? Does a cancelled order that was never dispatched count as a sale that then went away, or as something that never happened? On a marketplace, is the sale what the customer paid or what the marketplace will eventually pay you, and where did the commission go?

Every one of those is a real answer to a real question. Finance needs invoiced and net of returns, because that is what becomes money. The buying team needs ordered and gross, because that is what tells them demand. A store manager needs dispatched, because that is what left the shelf. All three are right, and all three will quote you “sales” without qualification, because in their own work there is only ever one meaning.

Nothing gets better until each of those is a separate, named number with its rule written down. Two people asking the same question should get the same answer, and that requires the question to have been written down once.

When is “last month” not last month?

The second cause is time, and it is the one that produces the most heat, because it makes a number you already signed off look wrong later.

Your past moves. A return comes in weeks after the sale. A credit note is raised against an invoice from a closed month. A marketplace settles late and restates its own figures. So the report you run today for a month that ended some time ago is genuinely not the report you ran on the last day of that month. Neither is wrong. They are as of different moments.

A number without an “as of” is not a number, it is a rumour with a decimal point. The fix is unglamorous: every figure carries the moment it was true, and a report that restates a closed period says what changed and why, rather than quietly producing a different total than the one in the minutes.

Which filter is nobody telling you about?

The third cause is exclusions, and it hides better than the other two.

Every report ever built has a rule about what it leaves out, and almost none of them have that rule written anywhere a reader can see. Cancelled orders. Test orders from when the channel was being set up. The staff purchase account. The store that joined halfway through the year. The channel that was added later and was never backfilled. The one customer whose returns are handled by an arrangement nobody wants to explain.

Each exclusion was sensible on the day it was added, and each was added by somebody who has since stopped thinking about it. Together they are why two reports from the same source disagree, which is the case that makes people doubt the system itself rather than the questions.

And the copies?

This is the cause everybody names first, and it is real: one report pulled from the billing system on the day, one from a spreadsheet updated on Fridays, one from an export somebody cleaned by hand. Each copy is a little stale and a little edited, and the differences compound.

It is also the easiest of the four to fix, which is why it gets all the attention. It is worth noticing that each of those copies exists because somebody had to move data by hand between two systems that do not talk. Every hand-off you count at the desk is also a place where two numbers get the chance to drift apart.

So is the answer a dashboard?

No, and this is where money gets spent badly.

Put a reporting tool on top of three sources that disagree and you have bought a fourth number. It is worse than the other three, because it arrives with charts and an air of authority and nobody can see which of the three it inherited its definitions from. The meeting still stops; it just stops later, with better typography.

The tool is not the fix. The written definition is the fix, and it is a piece of paper before it is a piece of software.

What does the fix actually look like?

Four things, in this order.

One place every number comes from, in your own accounts, that every report reads. Not a replacement for what you run: the billing system stays the book of record, the spreadsheet stays for looking at a number from a new angle. What stops is the copying.

A written rule for each number, in the words the business uses rather than the words a database uses. Sales, as finance means it. Sales, as buying means it. Both, named differently, so nobody has to guess which one is on the screen in front of them.

An “as of” and a filter on every figure, shown rather than documented somewhere nobody reads. If a number excludes staff purchases, the report says so where the number is.

And a parallel run before anything is switched off. The old report and the new one run side by side, on the same period, until either they agree or the difference is explained and signed off by the person who owns that number. A reconciliation that reports no difference every single time is not being run; a good one shows you the gap and tells you what it consists of.

What do you actually get out of it?

Something that is not a report, and it is the part that lasts.

At the end you own a document that says what each of your numbers means. That is what survives an analyst leaving, which is the event that quietly resets most businesses’ reporting to zero. A new person can be told what a figure means, instead of reconstructing it from a spreadsheet whose author has gone.

The fashion brand is the worked version. Eleven stores, two marketplaces and a website, each with its own idea of when a sale counts and what happens to a return, and a month-end that took most of a week and still produced numbers finance and buying each quietly distrusted. What changed was not that anybody started doing arithmetic better. It was that sales came to mean one thing, written down, and every report started from the same place.

So before anyone is asked to find the error, go and ask three people what they counted. The answer is usually sitting in plain view, in the gap between three reasonable people who were never asked to agree on the question. That gap is what reporting and data work closes, and it closes on paper before it closes in code.