To convert a purchase order PDF to Excel, make one row per PO line and repeat the header fields (PO number, revision, PO date, buyer, supplier, currency and payment terms) on every row. Each line row then carries its own values: line number, the buyer's part number and the supplier's part number in two separate columns, description, quantity, unit of measure, unit price, discount, line total and requested delivery date.
Leave carried-forward amounts and page subtotals out. A 300-line purchase order spread over 15 pages should give exactly 300 rows. Each subtotal that becomes a row inflates the sum of the line totals and creates a phantom line that will never match a receipt or an invoice.
This is for purchase orders you already hold as PDFs, scans or email attachments, whether they are your own outgoing orders or POs your customers send you. If you want a blank purchase order form in Excel to fill in, a template site is what you need. Here the job is turning purchase order PDFs that already exist into a spreadsheet without retyping them.
A PO looks like any other table until you try to flatten it. These are the structures that break a generic PDF-to-Excel conversion:
- header terms printed once that every line needs
- two part numbers for the same item
- revisions that reissue the whole order
- delivery dates that differ line by line, or split one line in two
- service lines priced over a period instead of by quantity
- blanket orders and the releases called off against them
- line tables that run across many pages, with subtotals at every break
Buyer or seller: which side of the order you are on decides the key columns
A purchase order has two readers, and they want different things from the same page.
On the buyer's side, procurement or AP converts its own outgoing POs. The rows go into an open-PO register, an ERP import, or the later match against goods receipts and supplier invoices. The key is PO number plus line number, and the item column that matters is the buyer's own part number, because that is what the item master and the receiving records use. The counterparty column is Supplier. Ship-To earns its place when several sites or warehouses order under one entity. The volume problem is usually depth: one layout, the one your purchasing system prints, but an archive or a backlog of hundreds of orders.
On the seller's side, the order desk or finance team at a supplier receives customers' POs and has to turn them into sales orders. The counterparty column is Customer. The customer's PO number becomes a reference you must quote back on the acknowledgement, the delivery note and the invoice, because a customer's AP team holds or rejects an invoice it cannot tie to an order. The item code you need is your own, but the customer prints their part number and only sometimes yours, so both have to survive the conversion. The volume problem is breadth: dozens of customers, each sending their own layout, all of which have to land in one sheet shape.
| Buyer's side | Seller's side | |
|---|---|---|
| What the rows feed | Open-PO register, receipt and invoice matching | Sales-order entry |
| Key item column | Your own part number | Your item code, mapped from the customer's part number |
| Counterparty column | Supplier | Customer |
| Header fields that matter most | Supplier, ship-to, payment terms | Customer PO number, ship-to, requested dates |
| Layout problem | One layout, many orders | Many layouts, one target sheet |
Settle which side you are on before choosing columns. Converting customer purchase orders to Excel and converting your own POs use the same extraction, but they put the weight on different fields.
A column layout for purchase order rows
This is a layout to start from when you extract purchase order line items to Excel. The first nine columns are header fields, printed once on the PO and repeated on every row. The next ten are line fields. Source File comes last.
| Column | Level | What goes in it |
|---|---|---|
| PO Number | Header | The order number as printed, keeping leading zeros and prefixes |
| Revision | Header | The change or revision number; 0 or blank for an original |
| PO Date | Header | The order date, not the print date or the delivery date |
| Buyer | Header | The ordering company (or entity, where a group orders from several) |
| Supplier | Header | The vendor the order is placed with |
| Ship-To | Header | The delivery site, as a name or site code |
| Currency | Header | ISO code (USD, EUR, GBP), even where the PO prints only a symbol |
| Payment Terms | Header | As printed: Net 30, 2/10 Net 30, 60 days end of month |
| Incoterms | Header | The rule and the named place, for example FCA Rotterdam |
| Line Number | Line | The PO's own line number (10, 20, 30 on many ERP-printed orders) |
| Buyer Part Number | Line | The ordering company's item reference |
| Supplier SKU | Line | The supplier's catalogue number, where printed |
| Description | Line | The item text, with wrapped lines joined |
| Quantity | Line | Ordered quantity as a number |
| Unit of Measure | Line | EA, BOX12, KG, M, HR: the unit the quantity and price refer to |
| Unit Price | Line | Price per that unit |
| Line Discount | Line | Percentage or amount on this line only |
| Line Total | Line | Extended amount as printed, net of the line discount |
| Requested Delivery Date | Line | The date asked for on this line |
| Source File | The file and page the row came from |
Three rows from one order show the point of repeating the header:
| PO Number | Revision | PO Date | Line Number | Buyer Part Number | Supplier SKU | Quantity | Unit of Measure | Unit Price | Requested Delivery Date |
|---|---|---|---|---|---|---|---|---|---|
| 4500018832 | 2 | 2026-10-06 | 10 | 300-1145 | HX-M8-40 | 2000 | EA | 0.18 | 2026-11-03 |
| 4500018832 | 2 | 2026-10-06 | 20 | 300-2210 | GL-CT-12 | 40 | BOX12 | 21.50 | 2026-11-03 |
| 4500018832 | 2 | 2026-10-06 | 30 | 300-0098 | 6 | EA | 145.00 | 2026-12-01 |
Buyer, Supplier, Ship-To, Currency, terms, Description, discount and totals are left off the sample for width. Line 30 has no supplier SKU because the PO did not print one; the cell stays blank rather than borrowing the buyer's number.
Each side trims the layout differently:
- Buyers usually drop Buyer, since it is always themselves, unless several entities order from one system. Keep Supplier SKU: the supplier's acknowledgement and invoice will quote it.
- Sellers drop Supplier (it is you), rename Buyer to Customer (and Buyer Part Number to Customer Part Number), and add an Own Item Code column. That column is filled afterwards from your mapping table, never from the PO, which does not know your codes.
That is 19 data columns before Source File, a starting point to trim. The structures in the next sections add their own (an order-level discount, notes, service dates, blanket and release numbers), which is why trimming matters. Whatever stays, write dates in one format (YYYY-MM-DD sorts correctly as text and as a date) and keep quantities, prices and totals as numbers so the sheet sums and pivots without cleaning.
Header fields, part numbers and units on every line
Most generic PDF converters get the line table roughly right and fail on the fields around it. Four rules cover the line-level decisions.
Repeat the header on every row. A PO prints its number, revision, date, supplier and terms once, at the top. A converter that copies the page faithfully puts the PO number once, above the table or on the first row, and leaves the rest blank. That sheet works only while the order sits in one block. Filter it, sort it by delivery date, pivot it by supplier, or append it to a register holding 400 other orders, and the blank rows belong to nobody. Filling down by hand is the usual repair, and it breaks the first time two orders are pasted out of sequence. A row that carries its own PO number and revision can go anywhere and still be matched.
Keep both part numbers. A PO line identifies the item in the buyer's terms, and often prints the supplier's catalogue number or SKU beside it. Keep them as two columns. Do not pick the one that looks more official, and do not merge them into one cell. The buyer needs their own number for receiving and the item master, and the supplier's to read the acknowledgement and invoice; the seller needs the customer's number to map it and their own to check the mapping. Electronic ordering already works this way. Peppol's electronic order standard (the BIS Order transaction, version 3.7, November 2025 release) gives every order line a line ID assigned by the buyer that must be unique within the order, a quantity with the unit of measure that applies to it, a requested delivery period for that line, and two separate item identifiers: one assigned by the buyer, according to the buyer's system, and one assigned by the seller, according to the seller's. A PDF that prints both part numbers is printing what e-ordering treats as two fields, and the spreadsheet should hold them the same way.
Keep the unit beside the quantity. A quantity of 10 means one thing in EA, another in boxes of 12, another in kilograms or metres. Drop the unit at conversion and the problem surfaces weeks later, when the supplier invoices 120 each against an order line for 10 boxes and the match fails on a difference that is not a difference. Give Unit of Measure its own column and leave conversion out of the extraction: if you need everything in eaches, convert in the sheet with a stated factor, so anyone can see where 120 came from. Unit price is per the PO's unit, so it travels with that unit.
Put each discount where the PO puts it. A discount printed on one line, as a percentage or an amount, stays on that line's row, and the line total is the net figure the PO prints. A discount printed once for the whole order, such as 3% on order value or a volume rebate, is a header field and repeats down the rows like the payment terms. It should not be spread across the lines during extraction; if you want it allocated, that is a formula in your sheet, where the method is visible. Freight and handling follow the same rule: a header field, unless the PO prints them as a numbered line of their own.
Revisions, split deliveries, service lines and blanket releases
These four are where a converted PO sheet goes wrong without anyone noticing, because each one produces rows that look valid.
Revisions and change orders
Keep the revision number as a column, and decide in advance what a new revision does to the rows already in your sheet.
When a changed PO arrives as a fresh PDF, which is how many customers send changes to a supplier, the current version is the one with the highest revision number. Its rows replace the earlier version's rows, matched on PO number and line number. Appending them instead leaves revision 1 and revision 2 side by side, and the order counts twice. A line that was on revision 1 and is missing from revision 2 has been cancelled, and some systems instead reprint a cancelled line with quantity 0 or marked deleted. Compare the two sets of line numbers, check for those markers, and close the cancelled lines rather than leaving them open. Keep the superseded rows on a separate tab or flagged as superseded, not deleted, so the history of what was ordered when is still there for a dispute or an audit.
Not every system sends the whole order again. Some issue a change order that lists only the lines that changed. A changes-only document updates the lines it names and leaves the rest alone; treated as a full reissue, it would cancel every line it does not mention. Know which kind each customer or purchasing system sends before you set up a spreadsheet for tracking PO revisions and change orders, and keep the two kinds apart if you receive both.
Per-line delivery dates and split deliveries
The requested delivery date belongs on the line's row, not in a header column. One PO commonly asks for some lines next week and others next quarter, and a single header date hides that.
A line scheduled in two deliveries (500 units in January, 500 in March) can be held two ways: as two schedule rows that share the line number, each with its own quantity, date and share of the line total, or as one row with the split recorded in a notes column. Either works. What matters is that the choice matches how receipts will be recorded against it, because two schedule rows meet two goods receipts cleanly and one row has to absorb both. The detail of that matching is covered in reconciling partial deliveries against one PO; here it only needs to decide the row shape.
Service lines
A line for maintenance, consulting, cleaning or a subscription often carries a period instead of a quantity of goods: "1 January to 31 December 2026", or "monthly, quarter 1". Give these lines Service Start and Service End columns. Forcing a quantity of 1 with the period buried in the description loses the one field the supplier's invoices will be checked against, since a monthly invoice is only correct if it falls inside the ordered period. Where the PO itself prices the service as 12 at a monthly rate, quantity 12 with unit MONTH is a faithful copy and can stay that way, with the dates added.
Blanket POs and their releases
A blanket PO is the agreement: agreed prices, a ceiling value, a validity period. Each release (also called a call-off) is the actual order placed against it. Convert them differently. The blanket becomes a small sheet of its own, the agreed price per item plus the ceiling and validity dates, not order lines. Each release becomes order lines, with the blanket PO number and the release number on every row, so releases can be summed against the ceiling and every release price checked against the agreed one. Putting blanket purchase order releases into Excel this way also gives the seller a running view of how much of a customer's blanket is left. For how blanket orders are set up and controlled, see how blanket purchase orders and their releases work.
Multi-page and scanned POs: why 300 lines must give 300 rows
On a long purchase order, a good share of what is printed is not order lines. A 15-page PO with 300 lines carries all of this as well:
- carried-forward and brought-forward amounts at each page break
- page subtotals
- the column headings, repeated at the top of every page
- the order total, tax summary and any header charges on the last page
- terms and conditions, often two or three pages at the back
Turn any of these into a row and it either adds to the sum of the line totals or sits in the sheet as a line with no part number that will never match a receipt. A carried-forward amount is the worst of them, because it has a value in the amount column and looks like data.
Two smaller cases cause the same damage. A line whose description wraps across a page break is still one line: the text that continues on the next page has no line number, quantity or price, and it joins the description of the row above instead of becoming a row of its own. And notes printed between lines ("deliver to gate 3", "certificate of conformity required") belong either to the line above them or to the order as a whole. Decide which, and put them in a notes column, rather than letting each note become a row.
Scanned and faxed POs, which customers still send in many industries, add marks a digital PDF does not have: approval stamps, signatures, and handwritten amendments to a quantity or a date. On a signed PO, a handwritten change is often the value both sides agreed, so a scanned purchase order converted to Excel has to settle which value wins, the printed or the handwritten, and the rows where it happened deserve a second look before anything is ordered or shipped against them.
When the sheet is done, two checks tell you whether it is complete:
- Row count. The number of distinct line numbers equals the number of lines on the PO (count rows instead if you keep one row per line). On orders numbered 10, 20, 30, the highest line number divided by the step gives the count, allowing for any gaps the PO itself has.
- Control total. The sum of the Line Total column equals the PO's net total before tax and header charges.
A row count that is too high usually means a subtotal or heading became a row. A control total that is short means a line was dropped, most often the last line on a page or the first after a break.
Running the conversion with one prompt for every PO layout
Everything above is a description of the sheet you want. With Invoice Data Extraction, that description is the instruction: you upload the POs, say in plain words what the spreadsheet should look like, and download it.
Upload the POs as they arrived. Regular and scanned PDFs, and JPG or PNG images, scans and phone photos included, one PO per image; a multi-page order goes in as one PDF. A single PDF can run to 5,000 pages and one task can hold up to 6,000 files, so an archive of orders goes in as one job. Line items that run over many pages are expected, not an edge case.
Start from the ready-made prompt. An empty prompt box offers ready-made prompts by document type, and purchase orders are one of them. Picking it fills the box with a short prompt in plain words, which you can change before extracting and save to your Prompt Library as your own.
Set the row structure to One row per line item. It is a setting on the task, beside Automatic (the default) and One row per invoice.
Say the PO rules in the prompt. A prompt built from the rules in this article reads like this:
One row per PO line. Repeat 'PO Number', 'Revision', 'PO Date' and 'Supplier' on each row. Put the buyer's part number and the supplier's part number in separate columns, and quantity and unit of measure in separate columns. Include the requested delivery date for each line. Skip page subtotals, carried-forward amounts and terms and conditions pages. Format all dates as YYYY-MM-DD.
On the seller's side, swap Supplier for Customer. For scanned orders with handwritten amendments, add the line "If a value is crossed out and handwritten, use the handwritten value", or the opposite if your policy is that only the printed order counts.
One prompt covers every layout. The AI reads the documents for itself, whatever their layout, so the same prompt serves every buyer's or customer's PO. You do not build a template per customer. That is the seller's real problem with incoming orders, and it is why a supplier receiving POs from forty customers can run them all through one saved prompt and get one sheet shape back. Separate prompts are for different jobs, such as a blanket agreement's price sheet beside its releases. The same approach, a prompt and a row setting, is how you extract tables and fields from business PDFs into Excel more generally, which matters because a PO rarely travels alone: the invoices and delivery notes matched against it go through the same way.
Fix the columns when the sheet feeds an import. With a free-text prompt, column names and order are usually consistent, but our AI may adjust them unless you say otherwise. When the file has to match the layout an ERP or sales-order import expects, use a Structured Prompt instead: you define the exact column headers in the exact order, with an instruction for each column, up to 20 columns. The 19-column layout above fits, with one to spare. Adding Service Start and Service End takes it to 21, so drop a column the import does not take, such as Incoterms on a services order.
Download and check what is flagged. The result comes back as Excel, CSV or JSON. In Excel, numbers come through as numbers and dates as dates, so the control total in the previous section is one SUM away; CSV and JSON carry every value as text; CSV is the usual choice when purchase orders go to another system's import. Two columns follow your data. Source File gives the file and page each row came from, so a revision number or a scanned quantity can be checked against its page. Review Needed marks a value or a row a panel of AI agents cannot agree on, with what to check. If the import must carry nothing but your own columns, either can be left out in Preferences.
If you are still comparing tools for extracting data from purchase orders, rather than doing the conversion, the guide to choosing purchase order data extraction software sets out what to evaluate.
Where the rows go next
On the buyer's side, the rows are the order half of every check that follows. Append them to a register keyed on PO number and line number, and you can build an open purchase order tracker in Excel that shows what is still due and from whom. When the supplier confirms, the same lines let you check a supplier's order acknowledgement against the PO for changed prices, quantities or dates before goods ship. And when goods and invoices arrive, the lines are what three-way matching of POs, goods receipts and invoices compares against. The receipt side of that match usually starts as delivery notes, and you can convert delivery notes to Excel the same way, keyed on the same PO and line numbers.
On the seller's side, the rows need one more step before they become a sales order: your own item codes. Keep a mapping table for each customer, their part number in one column and your item code in the next, and join it to the converted rows with XLOOKUP or a Power Query merge to fill Own Item Code. Any row that comes back blank is a part number you have not mapped yet, which is a new item, a changed reference, or a typo on the customer's side, and that short list is what someone checks before the order is entered. This is why both part numbers stayed as columns: the customer's number drives the lookup, and your SKU, where the customer printed it, confirms the lookup found the right item.
When that customer's revised PO arrives, convert it, replace the lines by PO number and line number as described above, and amend the sales order from the new rows rather than re-entering it.
The conversion produces the rows. Matching them to receipts and invoices, checking acknowledgements and entering sales orders happen in your own sheet or system, which is where those decisions belong.