Side-by-side transaction review
Click a row, the source line highlights in the PDF. Every row carries a confidence score. The math reconciles before export. The review surface is the product.
What the review surface does
The review screen is where a bookkeeper spends most of the time. Every shortcut exists because a real bookkeeper asked for it.
The source PDF renders on the left, the extracted transaction table on the right. No tab-switching, no scrolling between windows.
Click any row; the matching line in the PDF highlights. The bookkeeper sees the exact line that produced the row.
0–100% confidence per row. Anything under 90% is flagged; under 70% is a hard review tag. Low-confidence rows surface in a focused review queue.
Edit any field — date, description, vendor, amount, debit / credit — without leaving the review screen. Edits are persisted in the database; the export log captures them.
Accept all, mark distinct, mark low-confidence. The bookkeeper triages the flagged rows in one pass, not one at a time.
When a batch is in progress, the next statement opens as soon as the worker finishes. No waiting, no reloading.
How a review session flows
- 01Open the statement
Pick a statement from the client workspace. The first page of the PDF renders in the left panel; the extracted table appears on the right.
- 02Skim the low-confidence queue
Filter to rows under 90%. These are the rows the model is least sure about. The bookkeeper confirms or corrects them first.
- 03Click any row, verify the source
The matching line in the PDF highlights. Spot-check the date, amount, vendor, and balance.
- 04Edit inline if needed
If a row is wrong, edit it in place. The validation page re-checks the math on the next render.
- 05Bulk-accept the high-confidence rows
Anything over 99% with a clean source match can be bulk-accepted. The bookkeeper focuses attention on the flagged rows.
- 06Move to validation
When the queue is clear, move to the validation page to confirm the balance check.
What it is not
- Not an autonomous bookkeeper. The tool drafts; the human confirms. We do not claim “perfect AI extraction” or “fully automated bookkeeping.”
- Not a single-screen, single-purpose editor. The review surface is part of a workspace with a client list, an export log, and retention controls. A focused tool with a small surface.
Try the review surface
See what a reviewed transaction file actually looks like.
150 free pages. No credit card. The first review is the one that takes the most time when you're doing it by hand.
Frequently asked questions
- Can ProofRows convert bank statement PDFs to CSV?
- Yes. ProofRows extracts transactions from bank-statement PDFs — text-based or scanned — and lets bookkeepers review, edit, validate, and export the results as CSV. The QuickBooks-ready and Xero-ready CSV profiles import cleanly into those tools through their native bank-import wizards.
- Does ProofRows work with scanned bank statements?
- Yes. Scanned and image-only statements are read with built-in OCR, then parsed into transactions the same way as text PDFs. Low-confidence rows from rough scans are flagged for review, and every statement still has to pass the balance check before export.
- Can I import ProofRows exports into QuickBooks?
- ProofRows exports QuickBooks-ready CSV (for the Banking → Upload from file wizard) and .qbo Web Connect files that QuickBooks Online and Desktop import natively. A direct QuickBooks Online API push is on the roadmap and should not be described as live yet. ProofRows is not affiliated with or endorsed by Intuit.
- Can I import ProofRows exports into Xero?
- ProofRows exports a Xero-ready CSV that supports Xero's bank-import workflow (Accounting → Bank accounts → Import a statement). A direct Xero API push is on the roadmap. ProofRows is not affiliated with or endorsed by Xero.
- Does ProofRows export to Excel (XLSX)?
- Yes — XLSX export is supported today. CSV files produced by ProofRows also open cleanly in Microsoft Excel, Apple Numbers, and Google Sheets.
- Is ProofRows a replacement for bookkeeping software?
- No. ProofRows is a statement-cleanup workspace. It prepares transaction data for import or reconciliation in accounting tools like QuickBooks Online or Xero. It is not a ledger of record.
- Who is ProofRows built for?
- ProofRows is built for solo bookkeepers, small bookkeeping firms, accountants, tax preparers, QuickBooks ProAdvisors, cleanup bookkeeping specialists, and fractional CFOs who receive bank statements as PDFs.
- How long do you keep my client's bank statements?
- Source PDFs are auto-deleted after the per-client retention window (default 30 days, configurable). The extracted transactions and export-log audit trail stay for the life of the account. Account deletion cascades and removes every row, file, and auth artifact in a single transaction.
- Do you train AI models on my client's bank statements?
- No. We do not train models on customer data. The extraction provider we use is contracted not to train on API requests, and our application does not write statement text back to the database from the model — the only path through is a strict schema validation.
- How accurate is the extraction?
- Confidence scores are surfaced per row so the bookkeeper can see which transactions the model is less sure about. The validation page enforces a beginning + Σ credits − Σ debits = ending balance check before any export. We do not claim 'perfect AI extraction' — every export is human-reviewed.
More questions? Contact support or read the full FAQ.