OCR
Line-by-line extraction: an accurate total can hide ten wrong lines
Line-by-line invoice extraction goes down to the level where control actually happens: every item, with its reference, quantity, unit price and discount. OCR (optical character recognition) reads the page; the Capture agent rebuilds the structure of the table and makes each line comparable with the contract.
The problem
An accurate total proves nothing
An invoice can show an exact total and still carry ten wrong lines. A discount forgotten on one item, a unit price outside the schedule on another, a packaging unit changed without an amendment: the errors offset each other and the total lands correct. Control therefore happens at line level, and extraction has to reach that level too: that is the whole point of the overview of OCR for finance and procurement.
A total checked in aggregate assumes somebody has already approved the detail. In practice nobody reopens forty lines to verify an amount that comes out round. The wrong line goes through, and it will go through again next month: same item, same contract, same supplier.
This level of detail is not a purist requirement: it is the level your contract is written at. A price schedule sets a price per reference and per unit, not a monthly budget. A discount is negotiated by product family, not in the invoice footer. As long as the invoice stays a single amount, there is no point of contact between what was negotiated and what was paid.
Extraction is not the result. It is the first step. A price read correctly that does not honour your contract is still a price paid in error.
What one invoice line carries
A line is not an amount. It is a set of fields, each of which can be right or wrong independently of the others, and each of which serves a different check.
- Supplier reference and internal reference
- The two rarely appear on the same line. Without reconciling them, no comparison with your price schedule is possible.
- Description
- The supplier free-text label, which changes from one month to the next for an identical item. It helps with matching; it is never a unique key.
- Quantity and unit
- Twelve pieces or twelve boxes of twelve: the unit of measure decides the expected price. A misread unit falsifies the whole item.
- Unit price
- The field the payment depends on. This is what is checked against the framework agreement, the rate schedule and the price list in force on the delivery date.
- Line-level discount
- A percentage or a value applied to the item rather than to the invoice footer. This is where negotiated terms most often disappear.
- VAT base and rate specific to the line
- An invoice routinely mixes several rates. The rate is read on the line, not on the total, and reverse charge is flagged item by item.
- Item number and matching keys
- The purchase order line number, the goods receipt concerned, the batch: this is what makes matching possible without human intervention.
- Charges added to the item
- Freight, packaging, small-order surcharges: billed on the item, they change the real price and must stay separately readable.
Two fields deserve particular attention because they look alike without being equivalent. The gross unit price and the net unit price after discount are not read in the same place and are not compared with the same reference. Likewise, the quantity invoiced and the quantity delivered are two distinct data points: confusing them means approving a partial delivery at the price of a complete one.
Every field keeps the marker of the area of the document where it was read. Reviewing means looking at the value, not reopening the PDF alongside.
What it changes
Once the line exists as an object, control becomes a comparison rather than an investigation. It investigates, you decide.
- The detail becomes queryable: how many times this same item was billed above the schedule, across twelve months, all suppliers combined.
- Matching happens item by item against the purchase order and the goods receipt, not on an aggregate amount that hides offsetting errors.
- A single line becomes an investigated claim, tied to the clause that grounds it.
- The same lines, a month later, produce the same reading: comparing one period with another works on stable objects.
None of this requires changing management tool. The extracted lines are returned in the formats your teams already use, and your ERP (enterprise resource planning system) remains the source of truth: Zylio works on top of it, adding no development inside it. What changes is the level at which the data becomes available.
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
The subscription covers extraction. Checking your negotiated terms is the next step and is sized with you: the monthly volume packages in detail set out what each tier covers.
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 line-by-line invoice extraction?
The return of every item in the billing table as a distinct object, with its own fields: reference, description, quantity, unit, unit price, discount, VAT base and rate. The total stops being the only level of reading available.
Why is an exact total not enough?
Because errors in opposite directions offset each other. A discount not applied on one item and an overstated quantity on another leave the total unchanged. Without the detail, both stay invisible.
Are line-level discounts extracted separately from the unit price?
Yes. The gross unit price, the discount rate or amount and the net price appear as three distinct fields. That is the only way to verify that a discount negotiated by product family was applied to every item concerned, and not offset somewhere else.
Does the data need to be perfectly clean?
No. The Capture agent is designed for heterogeneous documents and incomplete reference data; structuring is part of the deliverable.
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
Continue on OCR
- Supplier invoice OCRThe fields expected on a purchase invoice, and what is checked next.
- Accuracy and confidence scoreA level of certainty per field rather than an overall percentage.
- Extracting a multi-page invoiceTables that run over thirty pages, down to the last line.
- Purchase order OCRThe reference for what you committed to: line item, quantity, negotiated unit price, lead time, cost coding.
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