Almost every merchant who understands their own economics learned them in a spreadsheet, and that is worth saying before criticising one. Building the sheet is what forces somebody to find out what a parcel really costs, which fee applies to which gateway and how much of a shelf price is tax. Nobody who has done it needs convincing that costs matter. The trouble is that the sheet is a snapshot of an understanding, maintained by hand, and the four errors it acquires are the same four in every store: stale costs, netted refunds, missed proration and quiet rounding.
- Keep the understanding, not the file. The knowledge in it is the valuable part.
- The four usual errors are stale costs, refunds netted backwards, fixed costs booked as lumps, and rounding applied too early.
- A sheet does not fail loudly. It drifts, and the drift is always in the flattering direction.
- The moment to stop is when maintaining it competes with acting on it, which is usually sooner than anyone admits.
Error one: one cost per product
A sheet holds one cost per product, so updating it after a supplier increase reprices every order in the file, including last quarter's. That is the opposite of what a platform's snapshot does, and it is why a hand-built margin history looks smooth: it is being continuously rewritten by its own maintenance.
The same field decides what a return credits back. Without dates, a unit sold in March and returned in June is credited at June's cost, and a price change between the two invents a small gain that no trade produced, which is exactly why a return should be credited at the cost that was effective then.
Error two: refunds netted backwards
Almost every sheet subtracts a refund from the month of its original order, because that feels tidy and keeps a product's history in one row. The consequence is that every past month improves whenever the present one bleeds, so no period is ever final and nobody can compare two months honestly.
Booking each refund on the day it was issued is noisier and correct, and it is the single change that makes a spreadsheet's history stop moving. Everything else follows from the same principle: a recognition date is simply the day an amount belongs to, and applying it consistently is most of what separates a trustworthy sheet from a decorative one.
Error three and four: lumps and early rounding
Fixed costs booked on the day the invoice arrives make one day catastrophic and the rest of the month flattering, which is why a daily sheet without proration is unreadable and a monthly one is merely lumpy. Spreading them across the days they cover is one column of arithmetic and it changes what the file can be used for.
Rounding is the quietest of the four. Any per-order cost that is a fraction of a cent, rounded per row, loses the same fraction every time in the same direction, and a column somebody once tidied to two decimal places will bias every total computed from it thereafter.
| Error | Effect | Direction |
|---|---|---|
| One cost per product | history reprices on every update | either, invisibly |
| Refunds netted backwards | past months improve as the present bleeds | flattering |
| Fixed costs as lumps | one terrible day, many flattering ones | both, per day |
| Rounding per row | the same fraction lost every time | flattering |
What to take with you
The cost knowledge, first: what a parcel really costs, which fee rules apply to which gateway, what the fixed block actually contains. That research is the expensive part and it transfers directly into any tool, or into a better sheet.
And the habit of looking. A merchant who checked their numbers weekly in a sheet will check them daily in a tool; one who never opened the file will not open a dashboard either. If the sheet has lapsed rather than failed, the honest first step is a daily P&L tracker with the four errors fixed, not new software.
Why the drift is always flattering
Three of the four errors move the answer in the same direction, and that is not a coincidence. Maintenance happens when something prompts it, and the prompts are asymmetric: a supplier increase gets entered late, a rounding tidy-up removes fractions rather than adding them, and refunds netted backwards make the past look better than the present.
So a sheet that has been maintained casually for a year is usually a little optimistic rather than randomly wrong, which is the harder failure to notice because nothing ever looks alarming. The bank account is the only thing that disagrees, quietly and slowly.
When to stop maintaining it
The signal is not size, it is where the time goes. When the hour a week spent re-fetching orders, re-checking refunds and re-pricing costs is longer than the time spent acting on what the sheet says, the sheet has become the work rather than the instrument.
The second signal is trust. When somebody asks a question about last quarter and the honest answer is that the file has changed since then, the history has stopped being a record. At that point the choice is either a disciplined rebuild with dates, or a tool that keeps the dates for you, and the method behind either is the same one set out in how to calculate your true profit.