Service · Document intake & OCR
Paperwork in. Clean records out. People check the unclear parts.
Invoices, intake forms, receipts and scanned pages are read, the fields you use are mapped into a spreadsheet or system, and anything uncertain is held for a person instead of being guessed.
Retyping is slow, and tired eyes make mistakes.
Somewhere in most small businesses, a person opens a PDF or a photo of a form and types what's on it into something else: a spreadsheet, accounting software, a job tracker. It's dull, it piles up, and the mistakes it produces (a transposed digit, a skipped line) tend to surface later, when they're harder to fix.
Document intake automation takes over the reading and typing, and keeps a person in charge of anything the software isn't sure about. Frank's background is in document-heavy, regulated financial services, where "the software said so" is never a good enough answer, and these setups are built the same way.
From the inbox to the record.
- IntakeDocuments arrive by email attachment, upload folder or scanner.
- ReadingText recognition (OCR) reads printed text; layout tells it which value is which.
- MappingThe fields you use are placed into a consistent structure.
- Review & deliveryUncertain values wait for a person; checked records go where you work.
Good fits
- Repeated document types with a recognisable layout: vendor invoices, intake or application forms, work orders, receipts.
- A clear destination: a spreadsheet, a shared folder with a naming rule, or software that accepts imported records.
- Enough volume that retyping is a real cost, but not so much that a short review queue becomes a full-time job.
Poor fits
- One-off documents with no repeating structure.
- Mostly handwritten pages, where a person would need to check nearly everything anyway.
- Situations where nobody is available to review exceptions.
03 · Sample walkthrough
How a page becomes a record.
A fictional sample invoice from a made-up vendor, with every value written into this page in advance. Pressing the button steps through how fields are matched and how an unclear one gets held back. No OCR or AI runs here, and nothing is uploaded. It shows the idea, not the accuracy of any real tool.
| Item | Qty | Amount |
|---|---|---|
| Office supplies, assorted | 6 | $342.00 |
| Replacement filters | 4 | $716.00 |
| Delivery | 1 | $45.00 |
| Sales tax | $181.50 |
Ready. Six fields will be matched from the fictional invoice on the left.
Mapped fields will appear here as a structured record.
| Field | Mapped value | Status |
|---|---|---|
| vendor_name | Example Supply Co. (fictional) | Mapped |
| invoice_number | INV-00417 | Mapped |
| invoice_date | 2026-03-04 | Mapped |
| due_date | 2026-04-03 | Mapped |
| po_number | "PO 58?1": unclear handwriting, held for a person | Needs review |
| total_due | $1,284.50 | Mapped |
Notice the dates: the page shows "March 4, 2026" and "04/03/2026"; the record stores both in one format. The handwritten PO number is not guessed.
The review queue is the point, not a flaw.
No reading software is right every time, and a confident wrong value is worse than an obvious gap. So each setup decides, with you, what should stop and wait for a person:
- Missing or unreadable values: a smudged scan, a handwritten note, a field cut off at the page edge.
- Values that break a rule you set: a total that doesn't match the line items, a date in the future, an unknown vendor.
- New layouts: a supplier changes their invoice template, and the first few are checked before they're trusted.
Held items go to one place (a review sheet, a folder or a short list in an email) with the original document beside them, so checking takes seconds, not a search. Nothing held for review is written onward until a person confirms it.
Your documents, your accounts, the least access needed.
- Processing runs in accounts and services you control or have agreed to, chosen during scoping.
- Access is limited to the folders and inboxes the workflow needs.
- Originals are kept alongside the extracted record; nothing is silently discarded.
- Retention (how long copies are kept, and where) is agreed with you.
- No accuracy percentage. It depends on your documents, and it gets measured on them, not quoted in advance.
- No handling of documents your industry rules restrict without the right agreements in place.
- No replacement for your accountant, lawyer or compliance advice.
Sensitive document types (health, financial account or identity documents) are discussed openly at the assessment. Sometimes the right answer is a lighter setup, or none.
06 · FAQ
Common questions
Can it read handwriting?
Sometimes, partially. Treat handwriting as something that will often need a human check. The sample above shows exactly that: the handwritten field is held, not guessed.
How accurate is it?
That depends on your documents' quality and layout, so no number is promised up front. During a build, a set of your real documents is run and checked by hand to see how it performs and where review rules are needed.
Where do the results go?
Wherever your team already works: a spreadsheet, a folder with consistent file names, or software that accepts imported records. Which destinations are practical is worked out in scoping.
What does it cost?
The assessment is free; the build is quoted separately once the document types, volume and destination are known. Some reading services also have their own usage costs, which you'd see before agreeing.
Is the sample on this page a real product?
No. It's a fictional invoice with pre-written values, there to show how mapping and review work. There's no app to download or log in to.