Pay for the statements, not the seats.

RBC, TD, Scotiabank, BMO, CIBC, Canadian Tire, and American Express are the issuers we test most aggressively. Statement AI handles arbitrary credit-card and bank PDFs with the same output shape — including smaller credit unions, business statements, and long multi-page Amex statements — so it usually just works on whatever you throw at it. If reconciliation fails (the amber flag on the workbook), that's the parser telling you it isn't confident.

Every statement is cross-checked: opening balance + Σ transactions ≈ new balance, plus per-section totals (payments, returns, purchases + fees, interest) compared to the totals printed on the statement. Tolerance is 2¢ for rounding drift. Anything off triggers an amber "check totals" badge in the workbook and the history page so you know to double-check before sending to your accountant.

No. PDFs are processed in memory and discarded the moment the workbook is built. We don't move them to object storage and we don't log their contents. We persist a small audit row per upload (filename, document type, totals reconciled, parser confidence) so your history page works — but no transactions, no PDF bytes, no images.

A single statement typically converts in 30–90 seconds. Multi-file uploads (up to 6 PDFs per batch) run in parallel, so six statements take roughly the same wall time as one. Hard ceiling is 5 minutes per upload due to Vercel function limits — split larger archives across uploads if you bump into it.

A four-sheet Excel workbook: Summary (one row per statement with all the totals), All Transactions (every line item including its category, sortable, filterable), Vendor Summary (aggregated by normalized vendor name with absolute spend), and Categories (aggregated by your saved vendor → category rules with % of spend). Same shape every time, so your pivot tables and downstream tooling don't break when we ship features.

During the preview after a parse, every transaction gets a category dropdown. Pick one (Meals, Vehicle, Software, etc. — 14 defaults for Canadian freelancers, or add custom) and Balancely saves a rule for that vendor. Next month's statement auto-tags every matching vendor without asking again. Rules live under /dashboard/categories where you can edit or delete them.

Yes — every conversion downloads as Excel (.xlsx), CSV, or QuickBooks Web Connect (.qbo). The QBO format imports directly into QuickBooks Desktop and Online with categories surfaced in the Memo field. CSV is RFC 4180, UTF-8. The download button on the preview is a split menu with all three formats.

Pick the document type at upload time. Credit-card workbooks render every amount as a negative outflow (so summing the Amount column equals the change in balance, the way an accountant expects). Bank workbooks keep canonical signs (debits negative, credits positive). The in-app preview tabs show the exact signs that will land in the .xlsx — no surprises.

PDFs up to 25 MB each, and up to 6 PDFs per upload. That comfortably covers six months of statements in one batch. Multi-year archives work fine — split them across several uploads and the history page keeps everything together.

No. Statements pass through our parser engine for the duration of the parse only. We don't enroll the project in any model-training program, and the file + contents are not retained after the parse completes.