Can several people in the practice review at once?

Each person works in their own browser session against their own account. There is no shared review queue yet — that is a genuine limitation rather than a feature being withheld.

Does it handle invoices in more than one language?

Extraction handles Latin and Arabic script and international number, date and currency formats. Where a date format is genuinely ambiguous the field is flagged instead of guessed.

What happens to a document it cannot read?

It is reported as a failure with a reason, and it does not consume a conversion from your allowance.

Solutions

Built for the month-end invoice pile

A practice does not process one invoice. It processes several hundred across a dozen clients in the same week, and the workflow has to be shaped around that.

Process one of your own invoices

Opens Invoice OCR with the upload area. Recognition runs on your own file, in your browser.

The problem

The bottleneck in a bookkeeping practice is rarely the accounting software. It is getting supplier documents into a state the software will accept — manual, repetitive work, and the least valuable thing a qualified person can spend the last week of the month doing.

Transcription is the part a machine should do. Your team's time is better spent on the exceptions: the invoice with the wrong VAT rate, the one billed to the wrong entity, the one that should never have been approved.

How it works

Work through a batch, not a file

Process supplier invoices one after another without re-configuring anything between them. Each document is handled independently, so one bad scan does not spoil the rest of the session.

Mixed PDFs and images in the same working session
Per-document outcome: extracted, needs review, or failed
A failure names its reason instead of disappearing

Tax as distinct fields

Tax is the field that has to be right. Vendor tax number, rate, amount and the net/gross split are extracted as separate values rather than inferred from the total.

Vendor and buyer tax identifiers
Rate and amount captured separately
Several rates on one invoice kept per line
Reverse-charge and zero-rated lines preserved as printed

Review by exception

Low-confidence fields are surfaced first. A reviewer confirms or corrects them, and the corrected values are what gets exported — there is no separate raw version that can leak into a filing.

Confidence shown per field
The original document beside the extracted values
Editable line items with totals that recalculate

Arithmetic as the real check

A parse that balances is almost certainly right. Line items are summed against the subtotal and the subtotal against the stated total, and rows that do not reconcile are flagged rather than exported quietly.

Exports a ledger will accept

An Excel workbook with a summary sheet and a line-item sheet, CSV shaped for ledger import, or JSON with full field-level detail.

.xlsx with summary and line-item sheets
CSV for ledger import
JSON with per-field confidence

Client documents stay on the reviewer's machine

Recognition runs in the browser, so a client's invoice is not uploaded to a server to be read. Only a conversion record — filename, size and page count — is stored against the account. What is and is not claimed is set out in full on the Security page.