Auto-splitting a Year of Invoices by Vendor

Every accountant has received it at some point: a single 400-page PDF with a year of invoices scanned in one go, in no particular order. Splitting that by hand is a full afternoon. Smart Split turns it into a coffee break.

The setup

Upload the combined PDF to the Split tool and choose AI Smart Split. The AI reads each page, groups pages that belong to the same invoice, and proposes a split boundary before each new document.

The reconciliation trick

Ask Chat with PDF: 'List every unique vendor name that appears, and for each one list the invoice numbers and totals.' You now have a reconciliation table you can paste straight into your accounting system, and each vendor's invoices sit in their own tidy split PDF.

What to double-check

• Any invoice where the total on the last page didn't match your ledger.
• Foreign-currency invoices where the exchange rate matters.
• Credit notes — the AI marks them, but a human should confirm the sign.

Why bulk scans end up like this in the first place

Most of these 400-page monsters aren't the result of bad process — they're the result of a reasonable one that broke down at scale. A bookkeeper receives paper invoices through the year, feeds them through an office scanner in batches, and saves the output as a single running file to avoid managing dozens of loose PDFs. That works fine at 20 pages. At 400 pages, spanning three suppliers of office furniture, a courier company, and a dozen one-off contractors, the file becomes unusable without tooling. The underlying data is fine; it's the container that's wrong.

The fix isn't to demand better scanning discipline from whoever is doing the scanning — that rarely survives contact with a busy office. The fix is to accept that the bulk file is the input, and build a workflow that turns it into structured output automatically.

A naming convention that survives month-end

Smart Split gives you individual PDFs, but a folder of forty files named invoice-1.pdf through invoice-40.pdf is only marginally better than the original blob. The naming convention is where most of the time saving either gets captured or lost. A convention that has held up well in practice is:

1. VENDOR_INVOICEDATE_INVOICENUMBER_TOTAL — for example ACMESUPPLIES_20250314_INV4471_1284.50.pdf.
2. Use the invoice date, not the scan date. The AI extracts this reliably from the document text; the scan date tells you nothing about the accounting period.
3. Keep the vendor name short and consistent — decide once whether it's ACME, ACMESUPPLIES, or ACME_LTD, and never change it mid-year, or your file sort order breaks.
4. Drop currency symbols and commas from the total in the filename; keep only digits and a decimal point so the string sorts and searches cleanly.

Once files are named this way, sorting a folder alphabetically automatically clusters every vendor together, and a second sort by date within that cluster gives you a chronological ledger view without opening a single PDF.

Batch renaming from the reconciliation table

Because Chat with PDF already produced a table of vendor names, invoice numbers, and totals, you don't have to type forty file names by hand. Export that table, paste it into a spreadsheet, build the filename string in a formula column, and use a bulk-rename script or your file manager's rename-from-list feature to apply it in one pass. The manual step shrinks from forty renames to one formula and one paste.

The reconciliation workflow, step by step

Splitting is the easy half. Reconciliation — making sure every invoice ties out to the ledger and nothing was double-counted or missed — is where accountants actually earn their fee, and it's worth being deliberate about the order of operations.

1. Run Smart Split and export the vendor/invoice/total table from Chat with PDF before you touch a single individual file.
2. Sort that table by vendor, then sum the totals per vendor. Compare each vendor subtotal against the corresponding line in your general ledger or accounts payable aging report.
3. Flag any vendor where the two numbers disagree by more than a rounding difference. In practice, disagreements usually come from one of three sources: a duplicate invoice that got scanned twice, a credit note that the ledger already netted off but the PDF still shows as a positive charge, or a payment that was recorded in a different accounting period than the invoice date on the page.
4. For each flagged vendor, open only that vendor's split PDF — not the original 400-page file — and manually check page by page. Because Smart Split already isolated the vendor, this is a five-minute check instead of a search through the whole document.
5. Once every vendor subtotal reconciles, file the individual PDFs into your document management system using the naming convention above, and archive the original combined scan as a single reference file in case an auditor ever wants to see the unsplit original.

That last point matters more than it sounds. Auditors are often more comfortable when they can see that the split files came from one traceable source document, rather than forty independently uploaded scans with no common origin. Keeping the original alongside the split output costs nothing in storage and saves an argument later.

Handling multi-currency batches

A single scanned batch spanning a full year will often include at least a handful of foreign-currency invoices — a software subscription billed in US dollars, a supplier in the EU billing in euros. Smart Split will still separate these correctly by page boundary, since it's reading document structure and vendor identity rather than currency. But the reconciliation table needs an extra column: ask Chat with PDF explicitly for the invoice currency alongside the total, because a total of '1,200' means something very different in USD versus a local currency, and a reconciliation that silently assumes a single currency will look correct while being wrong by whatever the exchange rate was that month.

The safest habit is to record both the original currency total and, where your ledger requires it, the converted amount at the exchange rate your accounting policy specifies — typically the rate on the invoice date, not the scan date or the payment date. Note the rate source in a column of its own so a reviewer can re-derive the conversion later without guessing which rate table was used.

Where the AI still needs a human

Smart Split is very good at page grouping and vendor identification because those are structural, visual tasks — a new letterhead, a new invoice number format, a page that starts with 'Invoice' rather than continuing an itemised list are all strong signals. It is weaker, by nature, at judgment calls that depend on context outside the document: whether a credit note has already been applied elsewhere, whether an invoice was a duplicate submission from the vendor's own billing error, or whether a total that looks wrong is actually correct because of a contractually agreed discount that isn't itemised on the page. Those checks stay with the person doing the reconciliation, and no amount of model improvement changes that, because the answer isn't in the PDF.

The realistic time saving from this workflow, based on the case above, is not eliminating the reconciliation work — it's eliminating the page-flipping and manual sorting that used to consume most of an afternoon, so the remaining time goes entirely to the judgment calls that actually need a trained eye.

A worked example with rough numbers

To make this concrete: a firm processing a 400-page combined scan covering roughly 45 invoices across 14 vendors typically spends, by hand, somewhere around three to four hours just locating page boundaries, checking totals against a printed ledger, and manually saving individual PDFs one at a time. With Smart Split doing the boundary detection and Chat with PDF producing the reconciliation table in one pass, that same batch usually takes twenty to thirty minutes of active review, plus whatever time the flagged discrepancies need — which in a typical month is two or three vendors, not all fourteen. The saving isn't in the judgment calls, which stay the same length; it's in removing the page-flipping and retyping that used to sit around those judgment calls.

Keeping the workflow auditable

One detail worth building into the process from day one: export and save the Chat with PDF reconciliation table itself, not just the split PDFs. If a reviewer or auditor later asks how a particular vendor total was derived, having the original AI-generated table — timestamped, and generated from the unedited combined scan — is a cleaner audit trail than trying to reconstruct which figures came from where after the fact. Treat it the same way you'd treat a working paper in a traditional audit file: keep it, date it, and don't overwrite it when you rerun the process for a later batch of invoices.