Order processing
What is sales order processing automation? A plain-English guide for UK wholesalers and manufacturers
Sales order processing automation is software that takes the orders your customers send by email, PDF, spreadsheet or web form, reads them, matches them to your customer accounts and product codes, checks them against your rules, and creates the sales order in the system you already use. A person approves each order before it is written, so the software removes the typing and keeps the decision.
It is 4.45 on a Friday afternoon and the order desk phone has gone quiet, which means the emails have started. A restaurant wants a Saturday delivery. Sandra knows without looking that their cut-off was four o’clock, and that the driver will fit them in if she asks nicely. The next order is a long spreadsheet from a care home group. The one after that is under the minimum order, so she rings the customer to add a case of something. None of these rules is written down. They are in Sandra’s head, along with what every customer means by “the usual”, and each order still has to be typed into Sage 50 line by line before she can go home.
Sandra is made up. The Friday rush is not.
Sales order processing automation is the name for software that does the typing part of Sandra’s Friday and leaves the deciding to her.
Key facts
- It reads orders that arrive by email, PDF, spreadsheet or web form and creates them in the order system you already have, without replacing that system.
- The steps are collecting the order, reading it, matching it to your accounts and products, checking your rules, a person approving, writing the order, and keeping a record.
- AI is used for reading documents, while matching, rules and writing to your system are done by ordinary software that follows your rules, which is why they can be tested.
- Sage 50 has no web API, meaning no built-in way for other software to talk to it over the internet, so the writing step uses an import file, Sage Data Objects or a connector, while Sage 200, Xero, QuickBooks Online and Business Central each have an API.
- Better Systems Co. builds a single email-to-order job for £1,500 to £3,000, and jobs covering several kinds of order or more than one system for £3,000 to £7,500.
How does sales order processing automation work, step by step?
Seven steps, and the fifth one is Sandra.
- Collecting. A mailbox or a folder is watched. Orders arrive exactly as they do today. The customer does nothing different and does not need to know anything has changed.
- Reading. The software reads the email, the PDF attachment or the spreadsheet and picks out the customer, delivery address, PO number, product lines, quantities and required dates. This is the step where AI earns its place, because customer documents are not consistent and change without warning.
- Matching. Each line is matched to your product codes and the order to your customer account. This includes the names each customer uses for your products, which are rarely the names on your price list. “Semi-skimmed, the usual” has to become a code and a quantity. Today that happens in Sandra’s head.
- Rule checks. Cut-off times, minimum order values, carriage charges, credit holds, substitutions for discontinued lines, and duplicate PO numbers. Sandra applies these from memory. Here they are written down as plain checks that your office manager can read and change.
- Approval. Sandra sees the order as read, side by side with the original, and releases it with one click or corrects it. Lines the software was unsure about are already flagged. An order with a problem the software recognised waits for her to sort out and is not offered for approval.
- Writing. The sales order is created in your system through a route its maker supports. For Sage 50 that is an import file, Sage Data Objects or a connector, because Sage 50 has no web API. See our Sage 50 page and how to get emailed PDF orders into Sage 50. For Sage 200, Xero, QuickBooks Online or Business Central it is the API.
- Keeping a record. Every order, every change and who approved it is recorded. Your accountant and your customer service team will both want this, for different reasons.
Before go-live there is a side-by-side period. Sandra keys orders as she always has, the software reads the same orders, and the two are compared. Only when accuracy on your own orders meets the agreed level does she switch from keying to approving.
What does it do, and what does it not do?
It does five things well. It takes the typing off the order desk. It applies your rules the same way every time, including at 5pm on a Friday. It holds anything doubtful for a person instead of guessing. It keeps a record of who did what. And it works with the system you already have, so nothing has to be replaced.
It does not do five other things, and anyone who tells you otherwise is selling. It does not take phone calls. Sandra still answers the phone, although her notes can go straight into a form on screen so the second typing disappears. It does not make credit decisions. It does not replace your accounts package or your ERP, which is the name for one big system that runs stock, orders and accounts. It does not fix a messy product file. If your codes are inconsistent it will show you on day one, which is useful but not comfortable. And it does not run forever without attention, because customers change their PO layouts and systems get upgraded, which is why we offer managed support at £150 to £1,500 a month.
Where is AI used, and where is ordinary code enough?
AI reads. Ordinary software checks what is allowed. A person decides what goes in. The table shows how that splits across the steps.
| Step | AI or code | Why |
|---|---|---|
| Reading a PDF, email or scan | AI | Layouts vary by customer and change without notice. An AI model copes with variation that a fixed template cannot |
| Matching product names to codes | Mostly code, AI as a fallback | Exact and near-miss matching against your product file gives the same answer every time and can be tested. The AI only suggests a match when the code finds nothing |
| Rule checks | Code | A minimum order value is a comparison. You want it to behave the same every time and be readable by your staff |
| Writing to your system | Code | A supported import or API call that tries again if it fails and never writes the same order twice, never an AI model guessing at buttons |
| Deciding to release | A person | Judgement about the customer, the stock and the day |
This split matters for the question people ask most. Can AI just enter the orders into Sage for us? The model is a very fast reader and a poor judge of its own mistakes. Letting it hold the keys to your accounts system is the wrong design. Letting it read, and leaving the rest to ordinary software and to Sandra, is the right one. Sage 50 has no built-in OCR, the software that reads text off a page, and no order reading of its own, as one accountant answered plainly on AccountingWEB. So the reading always happens outside the accounts package.
How do you tell if your business needs it?
Five signs, any two of which are worth a conversation.
- Someone spends more than an hour a day typing orders from documents that already contain the information.
- The same order is touched in two systems, such as a web shop and Sage, or a sales order and a Word pro-forma.
- Customers find your order errors before you do.
- Orders that arrive after the afternoon email rush miss the next day’s van.
- One person knows what every customer means by “the usual”, and the desk slows down when they are off.
Sandra’s desk shows the first and the last every week. Linda, who owns the business, will want the other test, which is the arithmetic.
The formula is Better Systems Co.’s own and the numbers are labelled assumptions, which you can replace in the ROI calculator. Say Sandra keys 15 orders a day at 5 minutes each, over 48 working weeks, at a cost to the firm of £16 an hour, a little above the £12.21 to £15 an hour advertised for Hull sales order processor roles (Michael Page, Adecco). That is 75 × 5 ÷ 60 × 48 = 300 hours a year, or £4,800 of capacity. At 40 orders a day it is 800 hours and £12,800. At 15 orders a day a £1,500 to £3,000 build with support is marginal and the honest answer may be to wait. At 40 it is not. Our post on how to stop re-keying customer orders walks through where the hours hide.
Speed is the other half. A competitor that automates its order desk confirms orders sooner, and customers notice.
How much does it cost?
Better Systems Co.’s prices are fixed before work starts and sit in bands. Anything we buy in from another supplier is passed on at cost and listed separately.
| Band | Price | What it covers |
|---|---|---|
| Automation audit | £500 to £1,250, credited against the build | One or two jobs watched, timed and costed, with a check on what each of your systems allows and a fixed quote |
| Quick win | £1,500 to £3,000 | One kind of order, such as emailed PDFs, into one system, with an approval screen and a place where doubtful orders wait for a person |
| Business workflow | £3,000 to £7,500 | Several kinds of order, customer-specific product names, rules such as cut-offs and minimums, or two systems |
| Operations automation | £7,500 to £15,000 and up | The order desk plus what follows it: pro-formas, delivery documents, courier bookings, stock kept in step in both directions |
| Managed support | £150 to £1,500 a month | Watching that it runs, fixes when a layout or system changes, a block of hours for small changes |
Three costs may sit outside those bands. The first is membership of the Sage Developer Programme, if the Sage Data Objects route is chosen and no existing product covers it, which Sage prices at £3,438 a year plus VAT from 1 August 2026. The second is any connector subscription. The third is the usage cost of the AI model that reads the documents, which is small per order and varies with volume.
What does Sandra’s Friday look like afterwards?
It is 4.45 on a Friday again. The restaurant’s order came in after its cut-off, so the software has held it and said why. Sandra still rings the driver, because that favour is hers to ask. The care home group’s spreadsheet has been read line by line and is waiting for one click. The order under the minimum is flagged with the shortfall already worked out, so her call to the customer is a short one.
The rules are no longer only in her head. They are written down where the office manager can read them, and they run the same way on the Friday Sandra is on holiday. She approves the last order, Sage has it, and she goes home on time.
What we would do
We would start with the automation audit, because the arithmetic above only means something with your numbers in it. We would collect fifty to a hundred real orders under a data agreement, count the ways orders reach you, time the keying, and test which route each of your systems allows. You get the cost of the typing in pounds. If the case stands up, we build what is described on our sales order processing automation page, run it side by side with manual keying, and hand over with staff approving orders and no longer typing them. Where the order has to reach an older system with no API at all, the approach is set out in how to integrate a legacy system that has no API and our system integration service. Terms used here are defined in the glossary.
Sources
- Sage knowledgebase, Sage 50 integration options: https://gb-kb.sage.com/portal/app/portlets/results/viewsolution.jsp?solutionid=200427112331075
- Hyperext, Sage 50 integration: https://www.hyperext.com/sage-50-integration/
- AccountingWEB, Sage 50 and 200 invoice automation, expert answers: https://www.accountingweb.co.uk/community/industry-insights/sage-50-200-invoice-automation-expert-answers-your-implementation
- Sage Community Hub, “Sage 50 Accounts - sales order generation - automation software”: https://communityhub.sage.com/gb/sage-50-accounts/f/general-discussion-uk/252855/
- Michael Page, sales order processor role, Hull: https://www.michaelpage.co.uk/job-detail/sales-order-processor/ref/jn-032025-6697236
- Adecco, order processor role, Hull: https://www.adecco.com/en-gb/job-search/order-processor-mk058-hull-east-riding-of-yorkshire/broadbean_644851779190287
- Sage, Sage 50 and Sage 200 UKI Developer Program UK pricing, July 2026: https://downloads.sage.co.uk/download/?did=cd53b8a9-506f-4d67-8ef7-4a016d7a8197
Questions people ask
Is sales order processing automation the same as an ERP?
No. An ERP is the main business system where the order lives once it is created. Sales order processing automation sits in front of it, reads incoming orders and creates them in the ERP. It works with Sage 50, Sage 200, Xero, QuickBooks, Pegasus Opera, Khaos Control, Business Central and bespoke systems, and leaves them in place.
Will automation replace my order desk staff?
It replaces the typing, not the people. In a firm of 10 to 100 staff the hours released go on work that was not getting done: chasing deliveries, calling customers, clearing pro-formas. Where the business is growing, it means not recruiting the next order processor. We model the value as capacity released, not headcount removed.
How accurate does it need to be before we rely on it?
Accuracy is measured on your own orders during a side-by-side period, in which staff key orders as normal and compare them with what the software read. The acceptance level is agreed in writing before go-live. Because uncertain orders are held for a person, the practical question is how many orders need a human touch, not how many go wrong.
Does it work if our customers all send orders in different formats?
Yes. That is the normal case and the reason AI is used for the reading step. Each customer's layout is learned from real examples. An unfamiliar layout is held for a person until it is handled reliably, so a strange format never produces a wrong order.