Order processing
Unknown codes, wrong prices, credit holds: what an order automation must refuse to guess
An order automation should read every order but never guess at anything that costs money. Unknown product codes, a customer's own names for products, pack sizes, prices that differ from the customer's price list, accounts on hold, missed cut-offs, orders under the minimum and new delivery addresses all wait for a person, with the reason shown and the original order beside it.
It is 2.20pm on a Wednesday at a food wholesaler, and an email arrives from a café. “2 x the usual oil, 1 x strong flour, same price as last time please, deliver to the new place.” Sandra has run the order desk for 14 years and reads it in five seconds. The usual oil is the 20-litre sunflower. The café’s price was a September deal that has ended. Linda put the account on stop on Friday, because two invoices are past 60 days. Thursday’s cut-off was 2pm, so the earliest van is Friday. And she has no idea where the new place is.
Sandra is made up. The five decisions she just made without noticing are not.
None of that is typing. It is knowledge, held in one person’s head, applied to every order in seconds. That is the part an order automation has to respect. Software that reads orders is good at reading. The danger is software that fills a gap with its best guess, because a guessed order looks exactly like a finished one.
Key facts
- Sage 50 Accounts lets you put a customer account on hold, but depending on the hold status “transactions can still be posted to the account providing you click OK when the warning message appears” (Sage knowledgebase).
- Customer price lists and special prices for one customer are in Sage 50 Accounts Professional only (Sage knowledgebase).
- A safe order automation sorts every order three ways. It is ready to approve, ready with a note to look at, or held until a person decides.
- Anything that changes money, stock or a customer’s terms waits for a person.
What does “refuse to guess” mean?
It means the software compares what the order says with your records and your rules, and stops when they disagree. It does not pick the nearest answer.
Two different jobs happen to every order. Reading is working out what the customer wrote, from an email, a PDF or a scanned sheet, and AI is good at that. Deciding is working out whether the order can go in as written, and that is a job for ordinary software that follows your written rules, with a person where the rules run out. Guessing happens when the reading step is allowed to decide.
So every detail ends up in one of three places:
- Matched. It agrees with your records. The order goes to the approval screen.
- Flagged. It is probably right, but something is unusual. The order goes to approval with a note against that line.
- Held. It cannot be approved until a person decides. The reason is written on screen and the original email sits beside it.
Which details must it never guess?
Each of these is a decision Sandra makes from memory today. The middle column is the shortcut a careless system takes.
| What the order says | The guess that goes wrong | What the software does instead |
|---|---|---|
| A product code you do not stock, or an old one | Picks the nearest code | Holds the line and shows the closest matches. A person picks one |
| The customer’s own name for a product, “the usual oil” | Picks the most likely product | Matches only if this customer has used that name before and a person approved it. Otherwise holds |
| A quantity in the wrong unit, “10” meaning cases or singles | Assumes a unit | Checks the unit against the product’s pack size and the customer’s past orders. Flags anything far from normal |
| A price that differs from what your system would charge | Uses either price silently | Shows both prices and holds the line. Never edits a price list |
| An account on hold or over its credit limit | Clicks past the warning | Holds the whole order, shows the status, and lets only a named person release it |
| A delivery day after the cut-off | Keeps the day the customer asked for | Proposes the earliest day your rules allow, for a person to confirm |
| An order under the minimum value | Waives the minimum, or adds carriage without saying | Proposes carriage by your rule and flags it |
| A new delivery address | Uses it, or ignores it | Holds the order. A new address is a new site or a mistake, and only someone who knows the customer can tell |
| A sender with no account | Creates a customer | Holds it. A new account needs a person to agree credit and terms |
| An out-of-stock product | Swaps in a similar one | Flags the line. Only the customer can agree a substitute |
Duplicates and amended orders belong on this list too. The post on what happens when automation fails explains the three checks that stop the same order going in twice.
Why is a credit hold the one that matters most?
Because it is the one guess that sends goods to a customer who may not pay. A wrong product costs a credit note and a second van. Goods sent to an account that should be on stop can cost the whole value of the order.
Sage 50 helps, but only so far. You can give a customer an account status that puts it on hold, and Sage can show a warning when someone posts to it. Sage’s own help page says that, depending on the status chosen, a transaction can still be posted if the person clicks OK. On the order desk that warning is a pause for Sandra to think. A program writing orders into Sage has no Sandra.
So we put the check before anything is written to Sage, in the automation itself. It reads the account status and balance first. If the account is on stop, the order is held with the status on screen, and only the people you name can release it. Usually that is the owner or whoever runs credit control. The order is not lost. It waits in plain sight until someone who knows the account makes the call.
How does the software know our rules?
You tell it, once, in writing. Most of Sandra’s rules have never been written down, because she has never needed them written. Writing them down is part of the automation audit. We sit beside the order desk, watch real orders go through, and list every check.
For Linda’s firm the list might read like this. Every rule here is an assumption for the example, not a real firm’s terms.
- Next-day delivery for orders in by 2pm. Fridays only for the coast route.
- Minimum order £150 before VAT. Below that, £10 carriage.
- Customer prices come from their price list in Sage 50 Professional. A price on the customer’s order that differs by any amount is held.
- Eight named accounts pay pro-forma, so their orders wait for payment before picking.
- Only Linda or the accounts manager can release an account on stop.
- A customer’s own name for a product is saved for that customer once Sandra has approved it.
That list has a second use. When Sandra is on holiday, whoever covers the desk works from the same rules she does, and the software applies them to every order, including the ones that arrive at 9pm.
Won’t it hold half our orders?
At first it holds more than it needs to, on purpose. The rules start strict and loosen only when the side-by-side run shows they can. In that run your team keys orders as normal while the software reads the same ones, and the two are compared. As the software learns each customer’s names for your products, fewer orders stop.
A hold is cheap, and a guess is not. Here is the sum, with every figure an assumption for the example:
- Assumed volume: 60 emailed orders a day.
- Assumed cost of Sandra’s time: £15 an hour, rounded up from the £12.21 to £15 an hour advertised for Hull order processor roles (Michael Page) to allow for employer costs.
- Assumed typing today: 4 minutes an order, so 4 hours a day, about £60.
- Assumed holds in the early weeks: 10 orders a day at 3 minutes each, so 30 minutes, about £7.50.
- Assumed cost of one wrong order that reaches a customer: an hour of office and warehouse time plus a £35 second van run, about £50, before any credit note.
On those assumptions, one guessed order that goes wrong costs nearly as much as a day of typing. A wrong order also costs goodwill, which is harder to price. That is why the software stops when it is unsure. Put your own volumes into the ROI calculator to see what the typing costs you.
What should never happen automatically, even later?
Anything that costs money or cannot be undone. That means changing a price, a credit limit or a customer’s terms. It means releasing a held account, creating a new customer, swapping a product, and sending anything to a customer in your name. These stay with a person however well the software performs. The full list is in the post on failures and approvals.
What can become automatic, if you choose, is approving routine orders from known customers where every line matched. That is your decision, made after the side-by-side run, and you can reverse it.
What does Sandra’s Wednesday look like afterwards?
The same café email arrives at 2.20pm. By the time Sandra looks, it has been read and checked, and it sits on her held list with five notes.
- “The usual oil” matched 20 litres of sunflower, because Sandra approved that name for this café in October.
- The café asked for last time’s price, a September deal. Today’s price list is shown beside it.
- The account is on stop. Only Linda or accounts can release it.
- The order arrived after the 2pm cut-off, so Friday is proposed.
- The delivery address is new to this account.
Sandra rings the café. It is the call she would have made anyway, but she starts it with everything on one screen instead of in her head. Linda releases the account once the old invoices are paid. The order goes into Sage 50 for Friday at the right price, and nobody typed a line.
You can try three sample orders on the order-reading demo, checked against a set of delivery rules the way this one was.
What we would do
We would start with an automation audit that writes down the rules your order desk applies from memory and puts a figure in pounds on the typing. Then we would build the checks before the reader, so that from the first day every order is matched, flagged or held for a reason you can read. That is our sales order processing automation service. For the Sage side, the Sage 50 systems page covers which editions hold price lists and how orders get in, and the glossary explains the terms.
Sources
- Sage knowledgebase, to place a customer account on hold (Sage 50 Accounts): https://gb-kb.sage.com/portal/app/portlets/results/viewsolution.jsp?solutionid=200427112205263
- Sage knowledgebase, customer price lists and special prices (Sage 50 Accounts): https://gb-kb.sage.com/portal/app/portlets/results/view2.jsp?k2dockey=200427112241033
- Michael Page, sales order processor, Hull: https://www.michaelpage.co.uk/job-detail/sales-order-processor/ref/jn-032025-6697236
Questions people ask
Can an order automation take an order from a customer whose account is on hold?
It can read the order, but it should not create it. A well-built automation checks the account status before anything goes into your system and holds the whole order, with the status shown, until a named person releases it. In Sage 50 some hold statuses only show a warning that can be clicked past, so the check belongs in the automation, not in a warning nobody sees.
Will it learn the names our customers use for our products?
Yes, one customer at a time and only after a person has confirmed the match. Once Sandra has approved 'the usual oil' as 20 litres of sunflower oil for one café, that pairing is saved for that café. The next order from the same café matches it. An order from another customer using the same words still waits for a person.
What if the price on the customer's order is right and our price list is wrong?
Then a person corrects the price list, not the software. The automation shows both prices side by side and holds the line. Someone who knows the customer decides which is right. The software never edits a price list, a special price or a customer's terms by itself.
Do we have to write all our rules down before we start?
No. Writing them down is part of the assessment. We sit with the person on the order desk, watch real orders go through, and list each check they make from memory. Most firms find more rules than they expected, and having them written down helps whoever covers the desk during holidays.