Large customers often send orders by EDI. Most of the rest send a PDF attached to an email, or type the order into the email itself, each in their own layout. Someone in customer service opens every one and keys it into the ERP line by line. It is slow, and a single typo in a part number, a price or a ship-to address becomes a wrong shipment, a credit and a phone call.
What has to be captured
Every PO carries the same information in a different arrangement:
- Header: the customer, PO number, ship-to address, requested date, terms and any notes.
- Lines: the customer’s part number and description, quantity, unit of measure and price.
Reading that data from a PDF or email is now reliable for most typed documents. Layout templates work well for a few stable formats; AI document extraction copes better with many formats and changes. Either way, each extracted value should come with a confidence level, and handwritten or faxed POs should be expected to need a person more often.
Match it against what you already know
Extraction is the easy part. The value comes from checking the PO against your own records before anyone confirms it:
- Customer and ship-to: matched to the customer account and a known delivery address.
- Items: the customer’s part numbers mapped to your item master through a cross-reference, with units of measure converted.
- Prices: compared with the customer’s price list or contract.
- Dates: requested dates checked against lead times or available inventory.
- Duplicates: a PO number already on file is flagged before a second order is created.
The customer part-number cross-reference is the most important prerequisite. If it does not exist, building it for your top customers is usually the first week of work.
Make exceptions obvious
A good review screen shows the PO and the draft order side by side, with every mismatch highlighted at the source line: a price that differs from the contract, an item with no match, a quantity far outside the customer’s normal range. Mismatches should never be corrected silently. The person confirming the order decides whether the PO or the price list is right.
Who confirms the order
At first, customer service confirms every draft order, which is still much faster than typing it. Once drafts are consistently right, you can agree in writing which orders may be confirmed automatically, for example repeat orders from a known customer with every line matched and within price. Everything else waits for a person, and every confirmation is logged.
After confirmation, the sales order is created in the ERP and an order acknowledgment with the ship date can go to the customer straight away, the same job an EDI 855 does for customers who trade by EDI.
Where EDI fits
PO automation does not replace EDI. Customers who already send 850 purchase orders should keep doing so. It handles everyone else, and a related workflow can catch failed or mismatched EDI transactions so they do not sit unnoticed.
What to measure
- Entry minutes per order.
- Order errors found after confirmation, such as credits or returns caused by entry mistakes.
- Time from PO received to order confirmed.
- Share of POs that need manual correction.
Measure each for a couple of weeks before changing anything, so the result is a comparison with your own baseline.
Where to start
Rank customers by PO volume and start with the formats that make up most of your orders. A shared order inbox, a cross-reference for those customers’ part numbers, price lists in the ERP and a way to create orders in the ERP (an API or an import) are enough to begin. It is the order entry workflow in our library.
