For UK distributors and manufacturers who receive purchase orders by email
Purchase order data extraction: reading customer POs into sales orders.
Purchase order data extraction takes a customer's PO, usually a PDF attached to an email, and turns it into the lines of a sales order in your own system. OCR gets you the text, but the order needs your item numbers, your selling units and the price you agreed, so each PO is matched and checked against your records and approved by a person before it goes in.
Or send us one order and see what we'd read and flag, by the end of the next working day.
The problem
The PDF is easy to read. Matching it to your system is the job.
A PO from a larger customer looks tidy: a PO number, a delivery address, a table of lines. The lines use the customer's own part numbers and descriptions, in their units. "CX15-E, qty 20, EA" means nothing to your system until someone knows it is your item 40-115, sold in packs of 10, so 2 packs. Then the price has to agree with yours, and the stock has to be there.
Few arrive as data. Of the 209 UK supplier websites in our study, only 3 mention EDI, where the PO would arrive with nothing to read, and only 7% give an email address just for orders, so emailed POs queue with every other enquiry (the research note). The figures are what the websites say, not counts of orders.
OCR and meaning
OCR for purchase orders gives you the words. The order desk needs what they mean.
OCR, optical character recognition, turns a picture of text into text. For a scanned PO it is the first step. A PDF made by the customer's own system usually has its text already.
Template tools then pull fields from fixed places on the page, one template per customer layout. They work until a customer changes their form. Neither step knows that "as before" means last month's order, or that EA on this customer's PO means a pack, not a single.
| On the PO | What OCR gives you | What the sales order needs |
|---|---|---|
| CX15-E | The text "CX15-E" | Your item number, from what this customer's code has meant on past orders |
| Qty 20, EA | "20" and "EA" | 2 packs of 10, if you sell in packs |
| Unit price £4.10 | "4.10" | A check against the price agreed with this customer |
| As per previous order | The words | The lines from that order, from your history |
| Revised PO, replaces 4471 | The words | Order 4471 found and amended, not a second order |
| Deliver to Kingsway site | An address block | The right delivery address on the account |
Illustration with invented data.
How it works
PDF purchase order extraction, step by step.
-
1
Read
Each PO is read as it arrived: a PDF from the customer's system, a scan, a spreadsheet, or lines typed into the email. The PO number, delivery address, dates, part numbers, quantities, units and prices come out.
-
2
Matched
Customer part numbers are matched to your items using that customer's past orders. Units are converted to how you sell. "As before" is filled from your history.
-
3
Checked
Each line is checked against the customer's agreed price and the stock you hold. A PO number you've already received is flagged as a possible duplicate. Every flag says why.
-
4
Approved
Your team checks the draft beside the PO and approves. Only then is the sales order created in your system.
What it connects to
Your inbox at one end, your ERP at the other.
- Where orders go
- Your ERP or accounts system, through its API where it has one, or an import file where it doesn't. See emailing orders into your ERP. Without either, the approved draft waits on screen for keying.
- What we read from your system
- Customers and delivery addresses, items and selling units, customer prices, stock, and a year of order history if you have it.
- Your rules
- How far a PO price may differ from yours, what to do with a short line, minimum order, delivery days. Written down in the session and checked on every order.
What it costs you today
Time one long PO from inbox to order.
Illustration with invented data: a 25-line PO takes 12 minutes to key and check by hand. At 30 POs a day that is 6 hours a day and 30 hours a week.
Put your own figures into the order desk calculator. It shows the hours and the pounds a year and doesn't guess at savings. The second number to know is order accuracy rate: how many orders go out with no line wrong.
What you get from us
In this order, with a fixed price before you commit.
-
30 minutes
Call with you
Which customers' POs take longest, and which number that holds back.
-
1 hour
Session with your team
We watch your team key real POs. The same day you get a one-page verdict: the workflow, the KPI it moves, today's baseline and a fixed price. Or a written no.
-
A fixed number of weeks
Build
We build the reading, matching and checks and connect them to your system. Nothing is subcontracted.
-
Every month
Run and report
We run it and report order processing time and order errors against the baseline. Stop at any report.
Questions
What people ask about purchase order data extraction.
- What is the difference between OCR and purchase order data extraction?
- OCR turns an image of a page into text. Purchase order data extraction turns that text into a sales order: PO number, lines, quantities and prices, matched to your own items and checked.
- Can it read scanned and handwritten purchase orders?
- Scans and photos, yes. Handwriting is read too, with less certainty, so more of its lines are flagged for a person to check.
- How are customer part numbers matched to ours?
- From your order history. Past orders show which of your items each customer's codes have meant. A code never seen before is flagged, and once your team answers, the answer is kept for next time.
- What happens when the PO price differs from our price?
- The line is flagged with both prices. Your rule decides what happens next, and a person approves the order either way.
- Do we need EDI instead?
- Only if a customer insists on it. EDI sends the order as data, so there is nothing to read, but each customer has to be set up one by one. Customers who email PDFs today change nothing.
- What does it cost?
- The price is fixed and written on the one-page verdict after the session with your team, before you commit to anything. If we don't find anything worth doing, we'll tell you and leave.
Next step
Thirty minutes on your purchase orders.
Tell us roughly how many POs arrive a day and which ones take longest. Or send us one PO first and see what we'd read and flag. Hiring for the desk? See our sales order processor job description.
The Open Build. We’ll build one UK supplier an order desk system, free and in public, measured on its own numbers. Everyone who applies gets a one-page read of their own workflow. Applications close at 5pm UK time on Friday 6 November. See the Open Build.