Split multiple invoices in one PDF
automatically detect and rename
Last updated:
Splitting a 30-invoice PDF by hand means finding 30 boundaries and renaming 30 files. The AI finds the boundaries itself and returns INV-2025-1234_City_Utilities.pdf, unlike pdf24 or Acrobat, which only count pages.
To split multiple invoices in one PDF, upload the batch PDF and pick invoice mode. The AI reads each page, detects the invoice boundaries itself, keeps follow-on pages together, and names each file by invoice number and company, for example INV-2025-1234_City_Utilities.pdf. You get a ZIP with one file per invoice and a CSV listing number, company, date, and confidence.
Pulling invoices out of a batch PDF is handwork in three passes
Each invoice in the stack has to be found, cut out, and renamed.
Invoices buried in mixed stacks
Your scanner output mixes invoices with contracts, letters, delivery notes, and forms. Ardent Partners puts the average cost of processing one invoice through the whole accounts-payable cycle at $9.40; finding it in the stack is only the first step of that.
Typing invoice numbers by hand
For each file, read the number, type the supplier, save. Fifty invoices a month means fifty save dialogs, and a transposed digit only shows up when someone searches for the invoice.
Everyone names files differently
Three people, three patterns like INV_2025_1234.pdf, Utilities March.pdf, or scan0815.pdf. Without a fixed scheme, nobody finds the invoice a colleague filed.
Splitting invoices by hand and with Docusplit
A 50-page scan in invoice mode takes about 5 minutes, depending on the page mix; progress is shown live.
| Criterion | Manual | With Docusplit AI |
|---|---|---|
| Time per invoice | Around 1 minute (our own assumption) | Around 6 seconds per page (invoice mode) |
| Filename | Typed by hand per file | InvoiceNumber_Company.pdf |
| Multi-page invoices | Checked and reassembled page by page | Follow-on pages detected by content |
| Overview | Keep your own list | CSV with number, company, date, and confidence |
| Unrecognized invoices | Show up at booking time | Listed as 'Not detected' in the CSV and named Invoice_Page_N.pdf |
The AI reads each page and detects where a new invoice begins
Three steps from batch PDF to individually named invoices.
Upload the batch PDF
Scan or digital file, up to 50 MB. Pick invoice mode first if the stack contains only invoices, otherwise document mode.
The AI identifies each invoice
Per page it reads invoice number, company, and date and detects follow-on pages, for example by 'Page 2 of 3' or a repeated invoice number.
Download the named files
The ZIP holds one file per invoice, such as INV-2025-1234_City_Utilities.pdf, plus a CSV with every value the AI read.
Splitting and naming happen in the same pass
What is finished after the upload without you opening a single file.
Invoices found in the stack
In document mode the AI checks each page for being an invoice; letters, contracts, and forms get their own type in the filename.
Number, company, and date read
The AI reads the invoice number, company name, and invoice date. It does not read amounts, tax, or bank details; your accounting software still does that.
Ready for filing
Uniformly named files and a CSV with filename, page, invoice number, company, date, and confidence. One invoice per file, which is what QuickBooks, Xero, or Dext expect when you upload documents.
Finding the invoices in a batch PDF is the slow part, not the scanning
The scanner is not the bottleneck. It produces one 50-page PDF in a couple of minutes, with invoices, contracts, delivery notes, reminders, and ads shuffled together. Someone then opens that file, pages through it, decides which pages are invoices and where each one ends, and saves them one by one.
As a rough example with our own assumptions rather than a study, say that takes a minute per invoice. Thirty invoices are then half an hour of paging and saving, every time a batch comes in. The app plans on about 6 seconds per page in invoice mode, so the same 30 pages run through in roughly 3 minutes while you do something else, and every file comes out in the same naming pattern.
The AI reads each page as an image and recognizes an invoice by its structure, the invoice number, the line-item table, the payment terms, the sender block, and separates it from the documents around it. Follow-on pages stay with their invoice.
How the AI tells invoices from similar-looking documents
Not every page with numbers on it is an invoice. Order confirmations, quotes, pro-forma invoices, credit notes, and delivery notes look alike, and keyword filters fail as soon as documents share phrases like 'total amount' or 'VAT included'.
Docusplit uses a vision language model, GPT-4.1 through the OpenAI API, that reads the page layout rather than keywords: the position of sender and recipient blocks, the structure of the line-item table, the presence of payment instructions, and the wording that separates a binding invoice from a non-binding quote. There is no custom training involved; the model reads each page as it comes.
When no invoice number is found, the page is not guessed at. It is named Invoice_Page_7.pdf in invoice mode and shows 'Not detected' in the CSV, so you can spot it at a glance. In document mode the same page gets its actual type, such as Quote_Acme_Corp_2025-03-15.pdf. There is no separate review queue; the confidence column in the CSV is a rough guide, and the files named by page number are the ones to check first.
From detection to files your bookkeeping can take as they are
Finding the invoice is half the job. The other half is a filename your accounting team or software can work with. By hand that means reading the invoice number, identifying the supplier, typing the new name, and moving the file.
Docusplit does the naming in the same pass. Once a page is identified as an invoice, the AI reads the invoice number and company name and names the file INV-2025-0815_Northwind_Ltd.pdf. The number comes first so you can search for it; the company comes second so you can tell the sender without opening the file. Names are cut at 100 characters.
Every batch comes with a CSV listing filename, page, invoice number, company, date, and confidence, plus a _metadata.json. Open the CSV in Excel or Google Sheets, or use it as the index when you move the files into QuickBooks, Xero, Dext, or your document management system by hand. One invoice per file is what those tools expect from an upload.
Multi-page invoices and attachments stay one file
A three-page invoice from a telecom provider, then a one-page letter, then a two-page supplier invoice: a splitter that cuts per page turns that into six files instead of three documents.
The AI checks each page for signs that it belongs to the previous one, such as a 'Page 2 of 3' header, the same invoice number, the same sender, or a line-item table that continues. When it detects a follow-on page, it appends the page to the running invoice instead of starting a new file. That works for two pages as well as for twelve, as with construction or IT invoices that carry long item lists.
Attachments such as time sheets or cost breakdowns stay with their invoice when the AI recognizes the link. An attachment with no letterhead and no invoice number is the case where this most often goes wrong; it then shows up as a file of its own. The 'Page' column in the CSV tells you which file holds which pages.
Three limits to know before you upload
The AI does not read amounts, tax, line items, or bank details. It reads what splitting and naming need, the invoice number, company, and date; booking stays with your accounting software. The output PDFs contain the pages unchanged and get no text layer.
There are no integrations, neither an API nor a connector to a document management system. You upload, wait for the run to finish, and download the ZIP. The CSV is an index for you, not an import format for a particular program.
E-invoices are a special case. Formats such as XRechnung, ZUGFeRD, or Factur-X carry the invoice data as embedded XML, and in some countries that XML is the legally relevant invoice. Splitting carries over only the visible PDF part, not the XML, so archive e-invoices in the original form in which they arrived. How long you keep them depends on your country, typically between three and ten years.
What invoice detection does, and what it leaves to you
Invoice-mode functions as they actually work.
- Detects invoices between letters, delivery notes, and contracts
- Reads invoice number, company name, and invoice date per page
- Tells an invoice from a quote, order confirmation, or estimate
- Keeps multi-page invoices together as one file
- Names files InvoiceNumber_Company.pdf, up to 100 characters
- Writes unread fields as 'Not detected' in the CSV
Three stacks that hide invoices
What changes from stack to stack is only what lies around the invoices.
The AP team's morning scan
The day's mail runs through the feeder, 40 pages in one file. After the upload each invoice is its own file, and the reminder and the packing slip next to it carry their own type in the name.
A client's shoebox at tax time
The client drops off a folder, the accounting firm scans it in one go. The AI splits the invoices, and the CSV shows which pages had no readable invoice number.
A supplier's combined attachment
Some suppliers send ten invoices in one PDF. Before booking, it becomes one file per invoice, named by number and company.
FAQ on invoice detection in mixed PDFs
How do I split multiple invoices in one PDF?▼
Upload the batch PDF in your browser and pick invoice mode. The AI reads each page, detects invoice boundaries by invoice number and layout changes, and keeps follow-on pages together. You get a ZIP with one file per invoice, named InvoiceNumber_Company.pdf, plus a CSV.
What happens with non-invoice documents in the stack?▼
They are split too. In invoice mode a page without an invoice number is named Invoice_Page_12.pdf and the CSV shows 'Not detected'. In document mode the AI first checks if the page is an invoice and otherwise names it by type, sender, and date, for example Delivery_Note_Acme_Corp_2025-03-15.pdf. For mixed stacks, document mode is the better choice.
Does it work with scanned invoices and poor scan quality?▼
Yes, with scans and digital PDFs alike; the AI reads the page as an image. Scan at 200 to 300 dpi. Very faint thermal paper, handwriting, and skewed scans below 200 dpi are where recognition drops off. The output PDFs get no text layer; the pages are copied unchanged.
Can Docusplit handle multi-page invoices?▼
Yes. Follow-on pages are detected by content, such as 'Page 2 of 3', a repeated invoice number, or a line-item table that continues. Attachments without a letterhead are the case most likely to be assigned wrongly; check the CSV when an invoice shows an unusual page count.
How is this different from invoice automation software?▼
AP automation suites cover approval routing, ERP integration, and payment scheduling. Docusplit covers only the step before that, finding invoices inside a batch PDF, splitting them out, and naming them. What you do with the named files afterwards is up to your existing tools.
Is my data secure?▼
Files are transferred over TLS, kept only while processing, and then deleted from our servers. AI analysis runs through the OpenAI API under a data processing agreement with OpenAI Ireland (Standard Contractual Clauses); your files are never used for training, and OpenAI retains API data for up to 30 days for abuse monitoring. See our privacy policy for details.
What file formats and sizes are supported?▼
PDF only, up to 50 MB per upload, with no page limit per file. Your plan sets the page quota, 30 pages once for the free trial, then 100 pages a month on Starter, 300 on Pro, and 1,000 on Max. The app plans on about 6 seconds per page in invoice mode, so a 30-page stack takes around 3 minutes, with progress shown live.
What happens if the AI splits an invoice wrong or cannot read something?▼
Handwriting, very faint scans, blank back pages, delivery notes that look like invoices, or attachments without a letterhead can be assigned incorrectly. When no invoice number is read, the file is named Invoice_Page_7.pdf and the CSV says 'Not detected'. Check those files first; the confidence column is a rough guide, not a review queue.
Related guides
For stacks that hold more than invoices, and for the step from splitting to naming.
Split & Rename PDF
Upload one batch scan of mixed invoices, letters and receipts. The AI splits it into separate documents and renames each file by content. Sorted ZIP with CSV.
Learn moreAuto Rename PDF Files
The AI reads type, sender and date on each scanned page and names the file after it, e.g. Contract_Northwind-Ltd_2025-01-15.pdf. 30 pages free, no credit card.
Learn moreDocusplit compared
How does Docusplit stack up against other tools? See the head-to-head comparisons.
Sources & References
- Ardent Partners, AP Metrics That Matter 2025 — Average cost per invoice processed in accounts payable $9.40, best-in-class $2.78; the figure covers the whole AP process, not the splitting step.
- PDF24 Tools, Split PDF — Offers the split modes pages per PDF, even/odd pages, halve pages, and custom; split positions are set by hand with a scissors icon.
- OpenAI, Your data (API) — Data sent to the OpenAI API is not used to train models (since March 1, 2023); abuse-monitoring logs are retained for up to 30 days by default.
Find, split, and name the invoices in your batch PDF
Upload one mixed scan and check the result against the CSV.
Try free now30 free pages, then from €9.99 for 100 pages a month
Last updated: