Duplicates, mixed formats, hidden totals, silent blanks and mislabelled columns: how to spot each one in a spreadsheet in a few minutes.
Why five minutes of cleaning pays off
An analysis can only be as good as the file it reads. Most bad results do not come from the AI getting the maths wrong. They come from a file that says something slightly different from what you meant. Here are the five problems we see most, each with a quick check you can run in Excel or Google Sheets before you upload.
1. Duplicate rows
Exports that combine two reports, or a sheet someone pasted into twice, double-count everything. Totals look healthy, averages barely move, and the trend is wrong.
Check: sort by an ID or date column and scan for repeats, or use your spreadsheet's "remove duplicates" on a copy and compare the row count.
2. Mixed formats in one column
A date column with "03/04/2026", "2026-04-03" and "April 3" in it is three formats, and "03/04" could be March or April. Numbers stored as text ("1,200" with a comma, "€50") are just as bad: they may be skipped or misread.
Check: filter the column and look at the distinct values. Pick one format (ISO dates like 2026-04-03 are the safest) and convert everything to it. Remove currency symbols and thousand separators from number columns.
3. Total and subtotal rows
Exports from accounting tools often add a "Total" row at the bottom, or subtotal rows between groups. To an analysis they look like one enormous extra customer or month.
Check: look at the last rows of the file and at any row where the ID column is empty or says "Total".
4. Silent blanks
An empty cell, a "0", "N/A" and "-" can all mean "we don't know". Mixed together, they make averages wrong in ways that are hard to see.
Check: decide what "unknown" looks like in your file. Usually that's an empty cell, and replace the others with it. Never use 0 for unknown in a number column; 0 is a real value.
5. Column names that don't say what they mean
"Value", "Amount 2" and "Col_F" force the analysis to guess. A guess about what a column means is the fastest way to a confident but wrong recommendation.
Check: rename each column to a plain description with its unit: "Order value (EUR)", "Signup date", "Region". If two columns look alike, say how they differ: "Revenue (gross)" and "Revenue (after refunds)".
Then upload
DataScope takes CSV, Excel (.xlsx and .xls) and JSON files. Once your file passes these five checks, upload it and choose the analysis type that fits your question. If the data is important and still messy, an Expert Touch review adds a human check on top of the AI's findings.