Clean a Bank Statement CSV for QuickBooks / Xero / FreeAgent

QBO's required headers are lowercase: transaction date, memo, amount. The exact wording matters — this is the only block that clears the 'three required columns' error.

{
  "Date": "transaction date",
  "Description": "memo",
  "Amount": "amount"
}

Use when your bank splits debit/credit into separate columns (Lloyds Debit/Credit, HSBC Paid out / Paid in). QBO maps four columns max.

{
  "Date": "transaction date",
  "Description": "memo",
  "Debit Amount": "debit",
  "Credit Amount": "credit"
}

Xero requires only Date and Amount; Payee aids reconciliation. Xero accepts both -30.00 and (30.00), so Step 4 is optional for Xero-only output.

{
  "Date": "Date",
  "Description": "Payee",
  "Amount": "Amount"
}

FreeAgent uses title-case column names and a generous 5 MB cap, so date-range splitting is rarely needed.

{
  "Date": "Date",
  "Description": "Description",
  "Amount": "Amount"
}

HSBC and Lloyds prepend a multi-line summary above the real header. Set this on the csv-cleaner node so the header row lands at row 1. Tune the number to your bank's actual preamble depth.

{
  "skipRows": 8,
  "trimWhitespace": true,
  "normalizeSmartQuotes": true,
  "removeEmptyRows": true
}

The three csv-find-replace nodes only treat find as a regex when the node's useRegex toggle is on. The blueprint ships useRegex: true, but if a node is edited or re-added without it, patterns like [£$€¥] and \(([\d.,]+)\) are matched as literal strings and quietly replace nothing. Symptom: amounts still carry symbols/brackets after a clean run. Fix: confirm useRegex is on in every find-replace node.

Although the node accepts columns: Date,Amount,Description, the chained runner keys on the **first** match column and keeps the **first** occurrence regardless of the strategy: last setting. With Date first, two different transactions on the same date can collapse. Fix: put the most-unique column first (a Transaction ID or Time column if your bank emits one), or run `csv-deduplicator` standalone where multi-column matching applies.

QBO accepts only 3-col or 4-col files. Causes: more than 4 columns, a header name in the wrong case, or invalid characters (curly quotes, emoji, en-dashes). Enable normalizeSmartQuotes on Step 1, drop extra columns with `csv-column-filter`, and use the exact Step 6 rename.

QBO is case-sensitive: transaction date (lowercase, two words), memo, amount. The cookbook's 3-col rename produces this exact wording — Transaction Date or Memo will still fail.

Files over 350 KB upload 'successfully' but only the first ~350 KB of rows reach the For Review queue. The workflow can't raise QBO's cap. Split the export by date range before importing, or target Xero / FreeAgent (no 350 KB limit).

Xero imports the valid rows and lists the failed row numbers — almost always a date-format/locale mismatch. Check your Xero org's region, then re-export with the matching date order; the workflow doesn't reorder date components.

If the input is empty or removeEmptyRows plus a too-aggressive skipRows strips everything, downstream nodes receive a header-only or empty CSV and the run produces a 0-row output. The run trace shows each node's row count in its summary — check Step 1's output before blaming Step 6.

Wells Fargo omits the header row, so the first transaction is treated as headers. `csv-header-rename` can target columns by index, but the cleaner steps assume a header — prepend a header line (Date,Amount,col3,col4,Description) before forking, or accept the first transaction being consumed as a header.

Excel for Mac writes lone \r; QBO expects \r\n and reads the whole file as one row. Set the csv-cleaner node's output line ending to crlf so the export is Windows/QuickBooks-safe.

The fork endpoint returns 401 unauthorized if you're not signed in — anyone authenticated can fork any published blueprint. Once forked, all six CSV nodes are free-tier, so the tier precheck passes and no upgrade prompt appears when you run.