Supplier invoice
Fields extracted
The invoice footer matters as much as the lines: that is where the forgotten rebate and the unagreed surcharge sit.
supplier invoice OCROCR
OCR for finance and procurement: read the invoice, then check the price
An OCR turns a document into text. Zylio’s Capture agent goes further: it extracts the data line by line (references, quantities, unit prices, discounts, taxes), then checks every price against the contract, the rate schedule and the purchase order. You do not simply get clean data, you get the discrepancy costed.
The problem
You are looking for an OCR. What you need is what comes next.
An OCR (optical character recognition) turns the image of a document into usable text. So an accounts payable team asked to “digitise the invoices” goes looking for an OCR. That is the right instinct, and it is the right first step.
But an invoice that has been read correctly is not an invoice that has been checked. A unit price can be extracted perfectly and still sit three pence above the rate schedule (the contractual unit-price grid) signed last year. A volume rebate can be missing with nothing to flag it. An indexation clause can have been applied to the wrong reference index. In all three cases the data is clean and the invoice is wrong.
The question is therefore not how well the document was read. It is what happens next: what do you do with the value once you have it? Until someone compares it with what was negotiated and what was ordered, extraction simply replaces one form of keying with another: faster, and just as blind.
Extraction is not the result. It is the first step. A price read correctly that breaches your contract is still a price paid in error.
That is why Zylio exists, and what sets our extraction apart from a generic tool: it is built for what comes next.
The three steps
Extract, then check, then prove. The three hold together, and in that order.
01
Read
The document arrives through the dedicated mailbox, a drop folder, SFTP or the API, the application programming interface through which two systems talk to each other. Capture reads it whatever the format (native PDF, scan, photo, spreadsheet) and extracts the data line by line. No template to build per supplier.
02
Check
Every extracted value can be matched against what was negotiated and what was ordered: the contract, the rate schedule, the price list, the purchase order, the goods receipt. This is the step extraction tools do not take.
03
Prove
Every field stays tied to the exact place in the document where it was read. Every discrepancy opens onto its source: the contract clause and the invoice line, side by side. You are not approving a conclusion, you are re-reading evidence.
The first two steps are not sold in the same place: the plan covers the reading, while checking your negotiated terms is sized with you. The third, the evidence, is not an option: no team approves a discrepancy it cannot reopen.
What we extract, document by document
Every document type has its own fields, its own traps and its own vocabulary. We do not process “documents”: we process supplier invoices, rate schedules and delivery notes.
Delivery note
Fields extracted
Without it, an invoiced quantity can only be compared with an order, never with what actually entered the warehouse.
delivery note OCRPurchase order
Fields extracted
It is the record of the commitment: the price that was meant to be paid, not the one being presented to you.
purchase order OCRContract and rate schedule
Fields extracted
A rate schedule is not a table: it is a grid wrapped in conditions, and it is the conditions that cost money when nobody re-reads them.
supplier contract OCRQuote
Fields extracted
That breakdown is not always provided; where it exists, it is the only way to bring two offers back to the same scope.
supplier quote OCRSupplier statement
Fields extracted
It is the document of omission: an invoice never received, a credit note never applied, a deposit counted twice.
supplier statement OCRThese lists are not decorative. A free-delivery threshold ignored, a handling unit mistaken for a selling unit, a revision index never read: any one of them is enough to make a check wrong while every total adds up. That is where the difference lies between reading a document and understanding a purchase.
What makes the difference on real documents
A tool is judged on the documents that fight back, not on the five-line sample invoice.
Extraction goes down to the line.
A correct total can hide ten wrong lines. Checking happens at line level, so extraction has to as well: unit, line discount, line-specific tax rate.
line-by-line invoice extractionLong documents stay correct to the end.
Tables that run over thirty pages are where most tools break: lines vanish silently, with no error shown.
multi-page invoice extractionBatches are split apart.
Four invoices stapled into a single PDF become four separate items. Attached timesheets and delivery notes are reattached to the right invoice.
automatic batch splittingLanguages and currencies do not get in the way.
Date formats, thousands separators, currency symbols, tax wording: everything is normalised on the way out, into a single grid.
multilingual invoice OCREvery field carries its own level of certainty.
A score per field, not a headline percentage, and a threshold you set to decide what goes to a human for review.
Why a headline accuracy rate says nothingDocuments arrive from wherever they like.
Email, scanned paper, supplier portal, EDI, e-invoicing platform: five channels, one flow.
automated invoice intake
All six share a common denominator: what breaks in production never shows in a demonstration. Extraction is judged on what it does when the document is not what was expected, and above all on what it flags when it is not sure.
Why a headline accuracy rate says nothing
The market advertises percentages. They are reassuring and they tell you nothing.
A headline rate mixes the easy fields with the fields that decide the payment. A tool can advertise a very high score by reading invoice numbers and dates perfectly, while regularly getting the unit prices of the lines at the bottom of the page wrong. And a percentage of correct fields, applied to a forty-line invoice, still lets several errors through per document.
What matters is the certainty on the field the payment depends on, document by document. That is why we give a score per field, the position of the value on the page, and a human-review threshold you set yourself.
So the question to ask is not “what is your rate?” but “on which field, on which document type, and what happens when you are not sure?”. A score per field with a threshold is the answer to that question.
invoice OCR accuracyBy role
The same flow, read three different ways.
Accounts payable
No more keying, accounting and cost-centre coding proposed, discrepancies qualified before approval.
accounts payable OCRProcurement
The prices actually invoiced become a negotiating position, not a year-end recollection.
data extraction for procurementFinance leadership
A shorter close, a complete audit trail, control applied before the payment run.
supplier invoice digitisation
These are not three products but one chain, read from three desks: accounts payable gains time, procurement gains a costed argument, and finance leadership gains a dated, defensible control.
What sets us apart
This market makes five promises. None of them is false; none of them is enough.
What the market says
What Zylio says
A very high accuracy rate on the key fields.
Reading accuracy says nothing about whether the price is right. We read, then we check against your negotiated terms.
If a human can read it, the tool can extract it.
A human reading an invoice still does not know whether the rate applied is the agreed one. The comparison does.
No model to train, no template to maintain.
True here as well, and on top of that we know your commercial objects: rate schedules, free-delivery thresholds, annual volume rebates, revision indices.
Clean data, pushed into your management system.
Clean data, and the discrepancy caught before the invoice goes out for payment.
Per-page pricing, tapering with volume.
The same here, and going over your allowance never cuts the service off.
Two questions that come up before you choose
They always arrive in this order: which technology are we talking about, and what does the alternative cost?
What one transcribes, what the other understands. The difference does not show on a sample invoice: it shows on a layout never seen before.
traditional OCR or AI OCRWhat an invoice keyed by hand really costs, before you even start counting the errors it produces.
cost of manual invoice entry
Fitting in without a project
API, SDK, webhook and a dedicated mailbox are included in every plan. Exports come out as XLSX, CSV or JSON. For ERP (enterprise resource planning) and S2P (source-to-pay, the chain that runs from sourcing to payment) environments, connectors and support belong to the enterprise offer.
Your ERP remains the source of truth: Zylio works on top of it and adds no development inside it.
Volumes are set by page bundle, and going over does not cut the service off: a heavy month end triggers neither a block nor a renegotiation. What comes in through the API goes out through the API, with the same fields and the same scores as whatever arrives in the mailbox.
document extraction APISecurity
Procurement documents are among the most sensitive an organisation holds: negotiated prices, terms, bank details. The commitments are therefore the same for extraction as for the rest of the platform.
- Controls
- Personal data
- Use of data
- Traceability
Pricing
From €99 excl. VAT per month for 500 pages analysed, with a per-page price tapering to €0.150 excl. VAT. No commitment.
Going over your allowance is billed at your tier’s rate, and the service is never cut off. That is a commitment, not a footnote.
The plan covers extraction. Checking your negotiated terms is the next step, and it is sized with you.
Your most difficult documents are the right test
Take the forty-line invoice, the stapled batch, the contract nobody re-reads. If extraction holds there, it will hold anywhere. Send us all three: the demonstration runs on your documents, not on ours.
Capture
Gather and consolidate all your procurement data.
Collection by dedicated mailbox, upload, SFTP, scan, API or ERP export. Reads every format (native PDF, scan, photo, spreadsheet, structured feed), then extracts line by line: supplier, references, quantities, unit prices, discounts, taxes, terms.
The end of re-entry. A procurement history you can finally query.
What Capture detects
- Long documents and multi-page tables
- Unknown layouts
- Several languages and currencies
- Poor scans and handwriting
Nothing is lost. Everything can be checked, everything can be proven.
- The contract clause and the invoice line, highlighted side by side.
- Every extracted value stays linked to the exact place in the document where it was read.
- The same case produces the same decision, today as in six months: the rules are applied deterministically.
- No discrepancy is set aside in silence. Anything that matches no rule is raised, with its reason.
- The agent records what it did, in the order it did it: who, what, how much, when.
Frequently asked questions
What is invoice OCR?
An OCR (optical character recognition) turns the image of a document into usable text. Applied to an invoice, it lets you extract the data without keying it: number, date, amounts, lines. An OCR on its own stops there: it returns text, not a judgement on that text.
What is the difference between a traditional OCR and an AI-based OCR?
A traditional OCR transcribes characters. It does not know that an invoice has a header, lines and a footer, nor where to find the unit price. AI-based processing understands the structure of the document, identifies the fields you asked for even on an unfamiliar layout, and returns a confidence level for every value.
Do we have to build a template per supplier?
No. You describe the fields you want, in plain English. No model to train, no configuration per layout, including when a supplier changes format without warning.
Does extraction go down to line level?
Yes, including on tables spread over several dozen pages, with the units, the line discounts and the tax rates specific to each line.
What happens when a field is illegible?
It is flagged, with its confidence score and its position in the document. Nothing is guessed silently: whatever is uncertain is raised for review, according to the threshold you set.
How much does it cost?
From €99 excl. VAT per month for 500 pages, with a per-page price tapering to €0.150 excl. VAT. The plan covers extraction; checking your negotiated terms is sized with you. The detail is on the pricing page.
Is my data used to train models?
No, never. It is not shared between clients and stays hosted in France.
Measurable impact in every environment
More than 5 million procurement documents analysed
Between 1 and 7% of margin recovered
on the scope analysed
From 15 to 45% of time given back to teams, per FTE
depending on the scope and on data maturity
Zylio fits into your existing ecosystem.
The ERP runs the process. Zylio handles the exception and recovers the value that escapes it: invoices without a purchase order, line-by-line price discrepancies, duplicates and overbilling, off-contract spend.
Your data under high security.
Zylio meets the most demanding standards, and nothing is committed without your approval.
- Certifications
- Hosting
- Encryption
- Access
Read next
- Accounts payable without keyingWhat extraction changes for the team that handles invoices day to day.
- Procurement armed with the price actually paidInvoiced prices become a negotiating position, line item by line item.
- Finance leadership before the payment runA shorter close, a complete audit trail, commitment verified before settlement.
- API and integrationIntegrate without a project: API, SDK, webhook and a dedicated inbox are included in every plan.
See what this looks like on your own data
Twenty minutes, on a spend category of your choosing. We show you what the agents detect, with the evidence behind it.
- No commitment, on your own data
- Result in 3 weeks
- 20 minutes, no sales pitch
- Your data stays hosted in France
- No change of tool or process