Choosing invoice OCR software for an accounting firm comes down to three things: whether it keeps each client's invoices separate and flags duplicates, whether its pricing fits a season with busy and quiet months, and whether a trial on your own vendor invoices produces a clean Excel or CSV export you would actually use.
Invoice OCR software requirements for an accounting firm
Invoice OCR software requirements for an accounting firm differ from those of a single company's accounts payable team. A firm receives invoices for many clients, from many vendors, through many channels. The software has to keep each client's invoices apart, cope with vendors it has never seen and catch the same invoice arriving twice.
A company buying invoice OCR usually has one set of books, one approval process and a vendor list that changes slowly. A firm has dozens or hundreds of sets of books, and each client brings its own vendors. That changes which features matter most.
Requirements that matter for a firm:
- Separate client folders. Every invoice has to land in the right client's file, with no chance of one client's invoices appearing in another's export.
- Many intake channels. Clients send PDFs by email, photos from a phone and downloads from supplier portals. Software that only accepts one channel leaves staff moving files by hand.
- New vendors on the first invoice. A firm cannot build a template for every vendor of every client. The software should read an invoice from an unfamiliar vendor without setup.
- Duplicate flags. A client forwards an invoice, then their bookkeeper forwards it again. Software should flag the second copy instead of exporting it twice.
- Scans and native PDFs together. Some invoices arrive as images and some as PDFs that already contain text. The difference is explained in scanned invoice vs emailed PDF invoice. Good software handles both in one queue.
- Visible checks. Each invoice should show whether its subtotal, tax and total add up, so reviewers look only at the ones that fail.
- Review before export. A person should be able to correct a field and see the correction in the export.
- Usable export. One clean file per client or per batch, covered in its own section below.
Accuracy is the question most buyers ask first. What it depends on, and why one figure cannot be compared with another, is explained in what affects invoice OCR accuracy on real vendor invoices. Test it on your own invoices instead.
Should invoice OCR software be priced per page, per user or per firm?
Invoice OCR software is usually priced per page or per document, per user, per client, or per firm with a page allowance. Per-page pricing tracks volume closely, so busy months cost more. Per-user pricing ignores volume. Per-firm pricing with a page tier is the most predictable for a firm with many small clients.
Each model behaves differently across a firm's year:
- Per page or per document. Cost rises and falls with volume. That is fair in a quiet month and painful in the busiest weeks of the season. Watch how pages are counted: a five-page invoice may count as five pages or as one document.
- Per user. Cost depends on how many staff log in, not how many invoices arrive. It suits a firm with a few heavy users. It punishes a firm that wants every preparer to review their own clients.
- Per client. Cost grows with the client list. A firm with many small clients who send a few invoices each pays a lot for little volume.
- Per firm with a page tier. One monthly price covers the firm up to a page allowance. Cost is predictable, and adding a client does not add a fee.
Questions to ask before signing:
- What happens when the firm goes over its page allowance in a peak month: a charge per extra page, a pause, or a forced upgrade?
- Do pages that fail and are re-uploaded count twice?
- Does a duplicate invoice the software flags still count toward the allowance?
- Is there a minimum term, and can the plan drop to a lower tier after the season?
Work out a year's cost under each model using your own numbers: clients, invoices per client per month, pages per invoice and staff who need access. The cost of keying invoices by hand, which is what any software has to beat, is covered in what manual invoice entry costs a firm.
How to trial invoice OCR software on your own vendor invoices?
To trial invoice OCR software on your own vendor invoices, build a test set of real invoices from several clients, record the correct values for each field by hand first, run the set through the software, and score every field as right or wrong. Score the whole set, not just a few good examples.
A vendor's demo uses invoices that read well. Your clients' invoices include faxes, phone photos, stamped pages and vendors no one has heard of. Only a trial on those tells you how the software will behave in March.
Building the test set
Pick 30 to 50 invoices from at least five clients. Include:
- Clean PDFs emailed by vendors.
- Phone photos of paper invoices, including at least one taken at an angle.
- A faxed or low-quality scanned page.
- An invoice with a "Paid" stamp or handwritten notes.
- A multi-page invoice with many line items.
- The same invoice twice, sent from two different email addresses, to test the duplicate flag.
- A credit memo with a negative total.
Before running the trial, type the correct vendor, invoice number, invoice date, due date, subtotal, tax and total for each invoice into a spreadsheet. That answer key is what makes the score honest.
Scoring the results
Compare the software's export with the answer key field by field. Count a field as right only if it matches exactly, including leading zeros in invoice numbers and the correct day and month in dates. Then look at the failures, not just the count. A tool that misses due dates on some invoices is easier to live with than one that occasionally swaps the total with the subtotal.
A ten-criterion scoring checklist
Score each criterion from 0 (fails) to 2 (works without help) during the trial. No vendor names are needed; the checklist works the same for any tool.
| # | Criterion | How to score it during the trial |
|---|---|---|
| 1 | Multi-client folders | 2 if every test invoice landed in the right client's file with no manual moves |
| 2 | Intake channels | 2 if email attachments, phone photos and portal downloads all went in without conversion |
| 3 | New vendors | 2 if invoices from unfamiliar vendors were read with no template setup |
| 4 | Header field accuracy | 2 if vendor, number, dates and totals matched the answer key on nearly all invoices |
| 5 | Duplicate flags | 2 if the invoice sent twice was flagged, not exported twice |
| 6 | Total checks | 2 if invoices whose amounts do not add up were marked for review |
| 7 | Line items | 2 if multi-page item tables came out as rows with no repeated headers or page subtotals |
| 8 | Export format | 2 if the Excel or CSV export opened cleanly with invoice numbers as text and dates as dates |
| 9 | Pricing model | 2 if a peak month's cost is predictable from the plan before signing |
| 10 | Review effort | 2 if a reviewer could correct a field and see it in the next export within a minute or two |
A total of 16 or more out of 20 on your own invoices is a strong result. Any criterion scored 0 that your firm depends on, such as duplicate flags for a client who forwards everything twice, is a reason to walk away whatever the total says. The thresholds are a suggestion; set your own before the trial starts so the score is not bent to fit a favourite.
Invoice OCR software export formats to insist on
Invoice OCR software should export a clean Excel workbook or CSV file with one header row, one row per invoice or per line item, invoice numbers kept as text, dates in one unambiguous format and amounts as plain numbers. An export that needs cleanup before every use costs back the time the software saved.
A CSV file is plain text with values separated by commas. The informal specification, RFC 4180, describes an optional header line first and says fields containing commas, double quotes or line breaks should be enclosed in double quotes. Vendor names like "Smith, Jones & Co." contain commas, so an export that does not quote them will split one vendor across two columns.
Excel adds its own traps when it opens a file. Microsoft's page on keeping leading zeros and large numbers says Excel automatically removes leading zeros, so invoice number 004821 becomes 4821 unless the column is imported as text. Microsoft's guide to importing or exporting text and CSV files says that opening a .csv directly uses Excel's current default data format settings for each column, while Data > From Text/CSV lets you load the file or transform it first.
What a usable export looks like:
- One header row with clear column names, such as Client, Vendor, Invoice number, Invoice date, Due date, Subtotal, Tax, Total, Currency, Status.
- One row per invoice for a summary, or one row per line item with the invoice number repeated on every line.
- Dates written as 2026-03-04, not 03/04/2026.
- Amounts with no currency symbols or thousands separators.
- A status column showing whether each invoice passed its checks or was flagged as a duplicate.
Whether to use .xlsx or .csv, and what each loses, is the subject of Excel export for accounting: XLSX or CSV. How to lay out a sheet of invoices once the export arrives is covered in invoice to Excel.
When invoice OCR software is the wrong purchase
Invoice OCR software is the wrong purchase when a firm handles few invoices, when nearly all of a client's invoices come from one vendor that offers a data export, or when a client already receives invoices as structured data through an e-invoicing feed. In each case the invoice data exists without reading pages.
Signs the money is better spent elsewhere:
- Low volume. A firm that keys a few dozen invoices a month across all clients may spend more on setup, training and review than it saves.
- One dominant vendor. If a client's main supplier offers a statement or transaction download, that file is more reliable than reading each invoice.
- E-invoicing already in place. A client whose suppliers send invoices as structured data into their accounting system has nothing to read.
- The bottleneck is elsewhere. If the real delay is clients not sending invoices at all, reading them faster does not help. The problem is collection.
It is also the wrong purchase when nobody will review the results. Any software makes mistakes on poor input, and an unreviewed export moves those mistakes into the books.
Where Schema CPA fits among invoice OCR software
Schema CPA fits firms whose invoices arrive mixed in with everything else a client sends. Receipts and vendor invoices come in through each client's single upload link, alongside tax forms and statements. Each page is classified by type, issuer and period, an invoice forwarded twice is flagged, and results export to Excel or CSV.
Pricing is per firm, not per client, with monthly plans sized by page volume: Free at $0 for 100 pages a month and up to three clients, Solo at $99, Firm at $199 and Practice at $499. Duplicate detection is included from the Firm plan, and API access on Practice. The plans are listed in the pricing section of the homepage.
What it does not do matters as much. Export today is Excel or CSV only; connections to QuickBooks Online and Xero are on the roadmap and do not exist yet. It gives no accounting or tax advice, so coding decisions stay with the firm.