What a chart of accounts should actually look like

Nobody sets out to build a bad chart of accounts. They get built one reasonable decision at a time: somebody wants to see software spend separately, so an account appears. Then one for a particular subscription. Then one for the annual conference, because that year it mattered. Three years later there are two hundred accounts, eleven of them are used twice a year, and four have names that only make sense to somebody who has left.

The cost is not tidiness. It is that the profit and loss statement stops being readable, categorisation becomes a series of judgement calls, and the same expense lands in three different places depending on who booked it. At that point the detail has made the numbers less reliable rather than more.

The test for a new account

Before adding one, ask what decision the separate line would change. If seeing it apart from everything else would make you act differently — spend less, renegotiate, price something else — it earns an account. If the honest answer is that it would be interesting to know, it does not.

The second test is whether it will still be used in a year. Accounts created for a single event stay on the report forever, showing zero, quietly making everything else harder to read.

Most small companies are well served by something in the range of forty to eighty active accounts. That is not a rule, and a business with genuinely distinct cost structures will need more. It is a signal: past a hundred, it is worth asking which of them anybody actually looks at.

Accounts answer what, dimensions answer where

This is the distinction that keeps a chart of accounts small. The account says what the money was — rent, software, contractor labour. Anything about which part of the business it belongs to is a different question, and accounting systems have a separate tool for it: classes, departments, locations, tags, depending on the software.

Once you start encoding the second question into account names, the chart multiplies. Three offices and twenty expense types become sixty accounts, and adding a fourth office means creating twenty more. Keep the twenty accounts, add the office as a dimension, and you can report either way without touching the structure.

The same applies to anything you are tempted to put in a name after a hyphen. If an account name contains a place, a person, a project, or a year, it is usually a dimension wearing an account costume.

Structure that pays for itself

• Direct costs separated from operating expenses, so gross margin is a real number rather than one somebody assembles in a spreadsheet.
• Payroll split enough to be useful — wages, employer taxes, benefits — but not one account per pay component.
• Owner activity in its own clearly named accounts, so draws, contributions, and personal spending on the company card are never mixed into operating results.
• A small number of parent accounts with sub-accounts underneath them, so the report can be read at either depth without being re-cut.
• Account names that describe the expense rather than the vendor, because vendors change and the report should not have to.

Cleaning one up without losing your history

The instinct is to delete, and it is the wrong first move. Deleting or merging accounts rewrites how prior periods are presented, and this year stops being comparable to last year — which is the main thing a small company uses its own history for.

Make accounts inactive rather than deleting them, so their history stays where it was and nothing new can land there. Merge only where two accounts genuinely mean the same thing and always meant it. And make the change at a period boundary, ideally the start of a year, with a note of what changed and when, so the step in the comparatives has an explanation attached.

Then leave it alone. A chart of accounts is infrastructure, and its value comes from being stable enough that this year and last year can be read side by side. The best one is not the most detailed — it is the one that answers your actual questions and has not been rebuilt in three years.