The safest bank transaction CSV for QuickBooks Online, Xero and Wave has three columns, Date, Description and Amount, with one header row, one transaction per row, a single signed amount (money in positive, money out negative), no currency symbols, no thousands separators, and dates in one consistent format that matches the company's region. Each app accepts more than that, but all three accept that.
If you write converters, scripts or bookkeeping tools, it pays to know where each importer is strict, because the error messages rarely say what is actually wrong.
Where each importer differs
Based on each vendor's help pages as of November 2026. Importers change, so re-check the current help page before you rely on an edge case.
| QuickBooks Online | Xero | Wave | |
|---|---|---|---|
| Amount layout | 3 columns (one signed Amount) or 4 columns (separate Credit and Debit) | One signed Amount column | One Amount column, or separate deposit and withdrawal columns |
| Extra columns | Ignored if you don't map them | Payee, Reference, Cheque Number are optional named columns | You pick which column is which during upload |
| Date | One date column; the format must be consistent | Must match the organisation's regional format | Only one date column; two date columns cause errors |
| Things that trip it | Currency symbols, commas in numbers, blank rows | Column order and header names | Symbols other than the currency sign, descriptions that are only numbers |
Wave's own help suggests that if a split deposit and withdrawal file won't load, move the amounts into one column. That's the main reason the single signed column is the common denominator.
A normalizer
Real statement extractions come in every shape: debit and credit columns, (45.00) for negatives, 45.00 CR, $1,204.18, dates like Oct 3 with the year only in the header. Normalize once, then write whichever file you need.
import csv from datetime import datetime from decimal import Decimal def money(raw: str) -> Decimal | None: s = (raw or "").strip().upper().replace("$", "").replace(",", "") if not s: return None sign = -1 if (s.startswith("(") or s.endswith("-") or s.endswith("DR")) else 1 s = s.strip("()-").replace("CR", "").replace("DR", "").strip() return sign * Decimal(s) def normalize(rows, date_format: str, debit_col=None, credit_col=None, amount_col=None): out = [] for r in rows: if amount_col: amount = money(r[amount_col]) else: debit = money(r.get(debit_col, "")) or Decimal("0") credit = money(r.get(credit_col, "")) or Decimal("0") amount = credit - abs(debit) if amount is None or amount == 0: continue d = datetime.strptime(r["date"].strip(), date_format).date() desc = " ".join(r["description"].split()) # collapse wrapped lines if desc.isdigit(): desc = f"Ref {desc}" # Wave rejects number-only descriptions out.append({"Date": d, "Description": desc[:200], "Amount": amount}) return out def write_csv(path, txns, us_dates=True): fmt = "%m/%d/%Y" if us_dates else "%d/%m/%Y" with open(path, "w", newline="") as f: w = csv.writer(f) w.writerow(["Date", "Description", "Amount"]) for t in txns: w.writerow([t["Date"].strftime(fmt), t["Description"], f'{t["Amount"]:.2f}']) f'{amount:.2f}' on a Decimal gives -1204.18: no comma, no symbol, always two decimals. That one line removes the most common import failure.
Card statements need a sign flip
On a credit card statement, a purchase increases what you owe. Some extractors write it as positive. Accounting importers expect a card purchase to come in as money out, the same as a debit from checking. Flip the sign for card accounts before writing, and flip payments to the card the other way. Do it by account type, not by guessing from the description.
Check before you import, not after
An importer will happily accept a file that is missing half a page of transactions. It can't know. You can:
def ties(txns, opening: Decimal, closing: Decimal) -> bool: return opening + sum(t["Amount"] for t in txns) == closing If that returns False, don't import. Find the gap first. Cleaning up a bad import in QuickBooks or Xero (undoing matches, deleting duplicates) takes much longer than fixing the CSV.
What about .qbo, OFX and IIF?
Some importers prefer OFX or QBO files when the bank provides them, and it's worth using those if your bank exports them directly. If you are starting from a PDF, CSV is the honest choice: it's what you can build and verify yourself, and all three apps accept it.
The bookkeeper version of this problem
For one client, a script is fine. A bookkeeper with 30 clients is running this every month for every account, which is where the time goes. Disclosure: I work on ZeraBooks. The ZeraBooks Converter writes QuickBooks-ready and Xero-ready CSV from PDF statements, checked against the statement balances first. For bookkeepers who would rather skip the import step entirely, there is a write-up on bookkeeping automation for bookkeepers that covers keeping every client's books in one place.
