OCR
Attachments and batches: four stapled invoices, four processes
Automatic batch splitting separates a PDF containing four invoices into four distinct documents. OCR (optical character recognition) reads the pages; the Capture agent decides where each invoice starts and ends, then attaches appendices, timesheets and delivery notes to the right document, so each invoice enters the flow complete.
The problem
A file is not a document
Scanning produces files, not documents. A pile scanned in one go becomes an eighty-page PDF holding twenty invoices, their appendices and two stray delivery notes. Handled as a single document, the whole thing yields one invoice with a wrong amount and nineteen invoices lost. To place this step, see what OCR covers before the check.
The reverse happens too: an invoice split across three files because the scanner stopped, or an appendix sent separately the next day. The right unit of processing is neither the file nor the page: it is the document. And nobody declares it.
The subject is often treated as a filing question. It actually decides what can be verified. A service invoice without its timesheet cannot be checked: the amount is plausible and nothing tells you whether it matches the work approved. A materials invoice without its delivery note cannot be matched, and a construction payment application without its measurement sheet cannot be justified. In all those cases the document exists, it was sent, and it is simply not attached to the invoice it belongs to.
Organisations compensate by hand. Somebody opens the PDF, cuts, renames, re-uploads. It is invisible work: it appears on no dashboard, and it conditions everything that follows. It is also the step where an appendix is quietly lost, because nothing records that it was ever there in the first place.
How a batch is split
Splitting does not rely on a separator page inserted by the user. It is inferred from the content.
- Document boundaries
- A change of invoice number, of date, of supplier header, the reappearance of an address block: all signals that mark the start of a new document.
- Recognised document types
- Invoice, credit note, pricing appendix, timesheet, delivery note, acceptance report, weighbridge ticket: every piece is given its type.
- Attachment to the right invoice
- An appendix is attached by its invoice number, its period, its site or its purchase order, never by its position in the file alone.
- Pieces split across several files
- Two separate uploads carrying the same number and the same period are brought together into one document, with a trace of both files.
- Non-billable pages
- Blank pages, cover sheets, transmission receipts and duplicate copies are set aside without being deleted: they remain viewable.
- Traceability of the split
- Every document produced keeps its source file and the page range it came from. A split can be checked, and corrected.
Splitting produces two pieces of information people often forget to ask for. The first is the list of pieces expected and missing: a works invoice announcing a measurement sheet absent from the upload is flagged as incomplete before approval. The second is the direction of the attachment: an appendix can concern several invoices from the same period, and the link is then multiple rather than arbitrary. No piece is attached by default to the first invoice encountered, and none is silently discarded because it looked like a duplicate.
The result is a list of typed documents, each complete, each attached to its supporting pieces. That list feeds the check, not the uploaded file.
What it changes
The attachment stops being a filing problem and becomes usable evidence.
- A timesheet attached to its invoice makes the service verifiable: hours billed can be compared with hours approved.
- An attached delivery note allows quantity matching without fetching it from another system.
- A pricing appendix attached to the contract becomes an enforceable price schedule, at the date it applies.
- Bulk upload becomes possible again: a pile scanned in one go no longer forces anyone to split it by hand.
The gain shows first on disputed files. A claim resting on an attached timesheet and delivery note is defended with an attachment, not with a reconstruction. And because splitting is deterministic, it replays identically: the same upload produces the same result, today as in six months. Your ERP (enterprise resource planning system) remains the source of truth, and each document is returned to it with the source file and the page range it came from, so an auditor can trace any line back to the paper it was read on.
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, batch splitting included; the count is on pages processed. Checking your negotiated terms is the next step and is sized with you: the monthly volumes in detail set this out tier by tier.
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
Does a separator page have to be inserted between invoices?
No. Boundaries are inferred from the content: number, date, header, address block. A separator page, where one exists, is recognised as such and left out of the document.
How is an appendix attached to the right invoice?
By what it carries: invoice number, period, site, purchase order number. Failing a certain match, the piece is flagged as unattached, with its reason.
What happens to a piece that attaches to no invoice?
It stays in the flow, typed and viewable, marked as unattached with its reason. Nothing is deleted or forcibly filed: an orphan appendix is most often an invoice that has not arrived yet, and the attachment happens as soon as it does.
Which documents does Zylio handle?
Invoices, contracts and amendments, price lists and rate schedules, quotes, purchase orders, goods receipts, credit notes, statements, as native PDF, scan, photo or structured format, in several languages and currencies.
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
- Delivery note OCRQuantities received, genuinely compared with the purchase order.
- Long documentsThirty pages of table, with no line lost at the end.
- Five arrival channels, one single flowEmail, scanned paper, supplier portal, EDI, e-invoicing platform.
- Supplier invoice OCREvery line extracted, every price checked: extraction is the first step, not the result.
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