Column consistency
One thing per column, one format, one convention. ISO dates (2026-08-12), USD to two decimals, sentence-case channel names. Don't accept a mix of formats — accept one and rewrite. The pass is mechanical: pick a column, scan top to bottom, fix every slip, move to the next. The file should look like one person wrote it on one day, even if it took a week and three exports to land. A consistent format is what makes the formulas downstream correct, and makes the file readable in five minutes instead of fifty.
Formula errors
Scan every formula for #REF!, #DIV/0!, #VALUE!, #N/A, bare ?, and blank zeros. Each one is a sign the file was patched without review. Replace them with labelled placeholders — [ref-needed], [div-by-zero], [value-error] — and add a one-line note in a side column explaining what should go back in. Look for a SUM range that quietly swallowed a deleted row, a VLOOKUP pointing at a renamed sheet, a custom format that hides the error but not the value. After this pass the totals can be trusted, and any remaining red flag is deliberate, not undeclared.
Duplicate rows
Duplicate rows are the easiest thing to remove badly. Sort by canonical key — SKU plus date, invoice plus line, customer plus month — and look at the duplicates. Drop the ones that add no new fact. Keep the ones that carry distinct meaning: a refund, a rebill, an exchange, a partial fulfilment. Where you keep one, label the row in a side column so the next reader knows it isn't a duplicate, it's a reconciliation. A file that hides a refund behind a generic duplicate is worse than one that didn't audit at all.
Named ranges
Name every meaningful cell range after the column it represents. Unit prices live in a range called unit_price. SKUs in sku_range. Invoice totals in invoice_total. The single most useful upgrade: formulas now read as sentences. =SUM(invoice_total) over sku_range reads as a sentence — sum the invoice totals across the SKU column — instead of a maze of cell coordinates. A reviewer can spot a wrong range in one glance, and the next formula inserted reads on the same convention. The cost of this pass is ten minutes. The payoff lasts as long as the file does.