Spreadsheets Drift Apart Before They Get Slow

The Sheet Edits Values Without Saying So

Excel does not leave what you enter alone. Microsoft lists four automatic conversions in its own documentation: removing leading zeros from numerical text and converting it to a number, truncating numerical data to 15 digits of precision and displaying it in scientific notation, converting numerical data surrounding the letter E to scientific notation, and converting a continuous string of letters and numbers to a date. The same company states the ceiling plainly. Precision stops at 15 significant digits, so in any number of 16 digits or more, everything past the fifteenth is rounded down to zero.

Which values get caught is predictable once you look. ZIP codes and phone numbers that open with a zero, order numbers padded to a fixed width, sixteen-digit card numbers, part codes that mix letters and digits. The conversions run when a CSV is opened, when a column is pasted in from another screen, and when find and replace sweeps through.

None of this looks like an error, which is the whole difficulty. The cell holds a tidy number and the totals still add up. Months later the file gets moved into a system, the altered values move with it, and now the new build is the thing that looks broken. So the first step in a migration is not moving anything. It is comparison: pull a few rows for every field that still has an original somewhere else, and check them by eye.

Hitting the Ceiling Is Rarely the Reason

Google’s help pages put a spreadsheet at up to 10 million cells or 18,278 columns. A single Excel worksheet holds 1,048,576 rows by 16,384 columns. At a hundred orders a day, one row each, reaching either number takes decades. Capacity is seldom what pushes a company off spreadsheets.

What actually stalls is not one file but the space between several. The test is short. Do two files answer the same question differently? Last month’s revenue, the stock left on the shelf today, this customer’s phone number. If the answer depends on which file you open, that is the moment.

Copies are not made out of laziness. Someone edits alongside a colleague, a row disappears, and from then on everyone keeps a private version. Or one person should see part of the data and there is no way to show part of it, so their slice gets cut out into a file of its own. A sheet opens as a whole file; you cannot hide a single column from one viewer. Companies with a drawer full of copies have usually met one of those two situations first.

Moving a Sheet As-Is Leaves the Rules Behind

A spreadsheet runs because a person fills its gaps by eye. Two rows with similar names get recognised as one customer. An empty cell gets read as not yet arrived, or as not applicable, depending on the column. A system cannot fill those gaps. Instead it asks first, and without answers the build does not start.

• What is one row? One order, or one product inside that order?
• Which field makes the same customer the same customer? Names collide and phone numbers change.
• Which cells are allowed to be empty, and which are not?
• Can past records be edited, and if they are, does the change leave a trace of who and when?
• Who enters data and who only reads it?

Once those five are settled the rest is a question of tools. Left unsettled, the copies come back within two months. People return to spreadsheets not because the system is slow but because something arrives every day that the system refuses to accept.

Decide What Moves and What Stays

Anything several people edit together, where yesterday’s change matters, moves. Orders, stock, customers, bookings. Anything one person works out for themselves is better left in a sheet: what happens if the price changes like this, how many records match that filter. Build those calculations into screens and the scope doubles while the person still cannot slice the data the way they wanted.

That is why the export stays after the move. Filter inside the system, pull a CSV, do the arithmetic in a sheet. The point was never to abolish spreadsheets. The point is that a spreadsheet stops being the original.

Nor does the switch happen in a day. There will be a stretch when both places are being filled in, and that stretch needs a dated line saying which one is authoritative. If both are, you are back where you started.

Start Without Giving Up the Sheets

Two or three of the files you use now are a whole first conversation. Weple’s data collection and cleanup work starts by laying those files side by side and reconciling the column names. Where values have already been altered, whether one row means the same thing in every file, which file is currently acting as the original: we sort that out before anything is built. How the work is counted is written on the pricing page. If you are weighing automation before the move, the order in Delete the Tasks Automation Should Never Touch is worth reading first.