VapusData logoVapusData logoFin
← All solutions

Solutions

Paper invoices, held to the same standard as digital ones

Scans, photographs and couriered documents with no text layer to fall back on: image quality handled at capture, a confidence score on every field, low-confidence values reviewed against the page, and the same validation and matching applied afterwards.

Book a demo
No text layer requiredConfidence score per fieldSplit-screen review
Payables
RetailManufacturingOCR & extractionN-way matchingERP write-backAudit trail
01

The problem

Some suppliers do not send a PDF. They send a scan of a printed invoice, a photograph taken on a phone at a receiving dock, or a paper copy that arrives by courier and is scanned by whoever opens the envelope. None of these carry a text layer, so a payables flow built on reading structured documents cannot read them at all, and they fall out into a manual queue. That queue is where re-keying, duplicate entry and late payment concentrate. The invoices themselves are ordinary; only the way they arrive is different, and it is the arrival that keeps them out of the automated path.

02

What the platform does

This is the paper-heavy path into AP rather than a second payables product. Image quality is handled at capture — skew, rotation, contrast and partial pages — before extraction is attempted. OCR reads the field set with a confidence score attached to each value rather than to the document, so a poor scan produces specific uncertain fields rather than a rejected invoice. Values below the confidence threshold route to a split-screen review with the source page open beside the extracted field. Everything after that is the same as for a digital invoice: the same validation, the same n-way matching and the same maker-checker before export.

03

What you get

Paper stops being a separate process. An invoice that arrives as a photograph enters the same flow as one that arrives as a structured file, and is checked against the same rules. Fields the OCR is unsure of are the ones a person looks at; fields it is confident about are not re-read on the assumption that scans are unreliable. Review happens against the page the value came from, so a correction is a decision made on evidence. What passes carries its confidence scores and its reviewer into the audit trail, which is what makes a figure read off a photograph defensible later.

Process

How it works

Four stages. The first two exist because there is no text layer to fall back on; the last two are the standard payables path, unchanged.

  1. 01

    Capture

    Scans, phone photographs and couriered paper are collected into one queue. Image handling runs first — skew, rotation, contrast, partial and multi-page documents — because attempting extraction on an image nobody has straightened produces confident nonsense rather than an honest failure.

  2. 02

    Read

    OCR reads the field set: invoice number, vendor, dates, amounts, tax and registration details. Each value carries its own confidence score, so uncertainty is located at the field that is actually unclear instead of being applied to the whole document.

  3. 03

    Review what is uncertain

    Values above the threshold continue. Values below it route to a split-screen review with the source page beside the field, so the reviewer confirms against the image rather than against a transcription of it. Confirmed values re-enter the flow with the reviewer recorded.

  4. 04

    Validate & match

    From here the path is identical to a digital invoice. The same mandatory-field, format and business-rule validation applies, the same n-way matching against purchase orders, receipts and expected data, and the same approval before anything is exported or written back.

What it does

Inside the solution

Documents with no text layer

Scans of printed invoices, photographs taken at a dock, and paper that arrives by courier. Where there is nothing to parse, the image is what gets read, and it is read as the primary source rather than as a fallback.

Image quality handled at capture

Skew, rotation, contrast, partial pages and multi-page documents are addressed before extraction is attempted, because extraction from an unprepared image fails quietly and in ways that are hard to detect downstream.

Confidence scoring per field

The score belongs to the value, not to the page. A smudged total on an otherwise clean invoice marks that total for review while the rest of the fields proceed, which keeps review effort proportional to the actual uncertainty.

Split-screen review

Low-confidence values are shown beside the region of the page they came from. The reviewer decides against the document rather than against a transcription, and the decision is recorded with the field and the person who made it.

The same validation as digital invoices

Mandatory-field, format and business-rule checks are shared with the rest of payables. A paper invoice is held to the same standard as a structured one, so nothing is accepted merely because it was harder to read.

The same matching and approval

N-way matching against purchase orders, receipts and expected data, followed by maker-checker before export or write-back. The paper path rejoins the standard flow rather than terminating in its own reviewed spreadsheet.

Duplicate detection across channels

A scanned copy of an invoice that also arrived as a file is the same invoice. Duplicate checks run across the whole intake rather than within a channel, which is where paper and digital most often collide.

Audit trail back to the image

Each field keeps the document, the page region, its confidence score and any reviewer who confirmed it, so a value read off a photograph can be explained months later without reopening the envelope.

Nearby

Related solutions

Payables

AP automation

The umbrella payables solution: capture, validation before ingestion, n-way matching, TDS tagging and expense mapping, with maker-checker approval before anything is exported.

Details
Payables

Multi-lingual accounts payable

Supplier invoices in several languages and scripts — often two on the same page — read into one canonical record, with currency, tax treatment and registration formats resolved per market.

Details
Payables

Contract, PO & invoice reconciliation

Three-way agreement between what was contracted, what was ordered and what was invoiced — with the clause that governs each check cited back alongside the result.

Details

Foundations

The VapusFin models behind it

Document Intelligence

Document Intelligence

Read leases, contracts, KYC files, loan forms, invoices, statements and transaction data, work out what each one means in context, apply the rules that govern it, reconcile it against the system of record and complete the work it was always going to trigger.

Details
Accounts Payable Automation

Accounts Payable

Capture every invoice however it arrives, match it line by line against the PO and the goods receipt, and pay only what the contract says you owe.

Details
Employee Expense Claims

Expense Claims & Reimbursement

Employee-side out-of-pocket claims: checked against your policy, cross-read against the receipt, deduplicated across people and periods, and paid straight through when they comply.

Details

Questions

Frequently asked

How is this different from AP automation?
It is not a separate product. It is the paper-heavy path into the same payables flow, covering capture and reading where there is no text layer. Validation, matching and approval afterwards are shared with AP automation, and mixed-language paper is handled by the same reading stage as multi-lingual AP.
What happens when the image is too poor to read?
The affected fields carry low confidence and route to review with the page beside them. The invoice is not silently filled in with a best guess, and it is not rejected as a whole because one region was unreadable.
Do phone photographs work, or does everything need a flatbed scan?
Photographs are part of the expected intake. Skew, rotation, lighting and partial pages are handled at capture, which is what makes a document photographed at a receiving dock usable rather than something to be re-scanned later.
Can the same invoice arriving on paper and as a file be caught as a duplicate?
Yes. Duplicate detection runs across the whole intake rather than within a channel, because a couriered copy of an emailed invoice is exactly the case a per-channel check misses.
What is the audit position on a value read from an image?
Each field keeps the source document, the region it was read from, its confidence score and the reviewer who confirmed it where one did. The value can be re-explained without anyone reconstructing how it was captured.
Has this been delivered for a customer?
Not as a standalone engagement. The capture, confidence-scoring and split-screen review it uses are the same components running in delivered document work elsewhere in VapusFin, and we would rather say that than present it as a track record it does not have.

See it on your own paper

Send us a batch of scans and photographs — the awkward ones, not the clean ones — and we will show you the extracted fields, the confidence scores and what would have gone to review.

Book a demo

Last reviewed

Cookie preferences

Essential cookies are required for the site to function and cannot be switched off. Everything else is off until you switch it on, and you can change or withdraw your choice at any time from the Cookie settings link in the footer. The Cookie Policy lists the cookies we set and how long each one lasts.

No choice recorded yet