Order entry is the cheapest-looking expensive thing in a distribution business. Four minutes an order. Nobody has ever put it on a budget line. And then you add up the year, and it turns out you are paying a person-and-a-half to retype information that somebody else already typed once.
The first hour of a Monday
It is 8:40. Your best inside-sales person opens the shared inbox and finds nineteen things in it.
Four are PDFs attached to one-line emails that say “see attached”. One of those PDFs is a photograph of a printed PO, taken at an angle, because the buyer’s scanner is broken again. Two are spreadsheets, one of which has the quantities in the column where the last one had part numbers. Three came through the customer’s portal, which means logging into the portal, finding the new orders, downloading them one at a time, and remembering which ones you already did on Friday. Two are EDI 850s that failed validation overnight and are sitting in an error queue with a message about an invalid unit of measure. The rest are questions, a credit query, and a reminder about somebody’s birthday.
None of this is hard. That is the thing worth sitting with. Not one step in the next hour requires skill that took years to build. It requires attention, patience, and knowing where things are — and it is being done by the person on your team who knows the most about your customers.
By 9:40 the orders are in. Mostly. One is waiting on a price question. One went in with the customer’s part number instead of yours, which nobody will notice until the wrong item is on a truck. And the portal orders from Friday afternoon missed the Friday cutoff, so they ship today instead, which the customer will mention.
Anatomy of one order
Here is one order, broken into what actually happens. Count the steps in yours; the list is usually longer than people expect.
Mapping part numbers is judgement: the customer orders the thing they call HX-4420-B, your item master calls it something else, and whether those are the same thing depends on what they bought last time. Pricing is judgement, because contract pricing, price breaks and the deal your rep did in March do not all live in the same place. Credit is judgement, because somebody has to decide whether this order goes on hold or goes out.
The other six steps are transcription. They are the ones taking the time.
Three costs, and you only ever see one
The time
This is the cost people can picture, and it is the easiest to underestimate, because nobody measures the whole task. They measure the typing. The real figure is from the moment the order appears to the moment it is filed and acknowledged — including the portal login, including going back for the PO number, including the interruption when a rep walks over to ask about something else.
Per line is where it adds up. An eight-line order is not twice the work of a four-line order plus a bit; every line is a part number to match, a quantity to read off a document that was not designed to be read, and a price to confirm.
The errors
An order entered wrong does not cost you the time it took to enter. It costs you a credit, a second shipment, the freight on that shipment, the call that found the problem, the time spent working out what happened, and — on a bad one — a customer who now checks every acknowledgement you send.
Three kinds account for most of it. Wrong item, usually from a part-number mismatch. Wrong price, usually from pricing that lives in three places. Wrong ship-to, usually because the customer has four sites and the PO names one of them in a way nobody standardised.
The number you want is not your error rate on paper. It is: how many orders a week does somebody have to go back and fix, and what does an average one of those cost by the time it is closed out?
The delay
This is the cost that never appears anywhere. An order that arrives at 3:15 and is entered at 9:05 the next morning ships a day later than it could have. Nobody logs that. The customer does not complain about it. They just quietly notice that your lead time is a day longer than the other supplier’s.
If you have a same-day shipping cutoff, this one is measurable: how many orders a week arrive before the cutoff and get entered after it?
Run it on your own numbers
Everything above is structure. The arithmetic depends entirely on your volume, your line counts and your costs, so here it is as a calculator with every assumption on the face of it. Change anything that is wrong — and something always is.
Two things usually surprise people the first time they run it. The first is the full-time-equivalent line: the hours are real even though no single person is doing them. The second is that the error cost gets within range of the labour cost at an error rate most people would call acceptable.
Why OCR and templates did not fix this
Most distributors have already tried. Usually twice.
The first attempt is templates: a rule per customer that says the part number is in column C and the quantity is in column E. This works until a customer changes their PO layout, which they do without telling anyone, and then the rule fails silently and somebody spends an afternoon working out why four orders went in with the quantities in the wrong field.
The second attempt is OCR. OCR solves a real problem — getting text off a PDF — and then stops, because text was never the hard part. The hard part is that the text says HX-4420-B and your system has never heard of it. Reading the document is step one of nine. Deciding what the document means is steps three, four and five.
And both approaches share the same fatal assumption: that exceptions are rare. In order entry, exceptions are the job. The short-ship, the obsolete part with a replacement, the MOQ the buyer ignored, the price that does not match the contract — those are not edge cases in the margins. They are a steady share of every week, and any tool that handles only the clean orders hands you back the hard ones with no context and calls it done.
What an agent changes
The useful framing is not “automate order entry”. It is: get the six transcription steps off people, keep the three judgement steps with people, and make sure the handoff between them carries enough information to act on.
In practice that means the agent works from what you already have — the shared inbox or the EDI feed, your item master, the order history for that account, and whatever pricing rules someone can write down. It reads the document whatever shape it is in, because it is not matching a template. It proposes the lines. Where it is confident and the order is inside the limits you set, it enters it. Where it is not, it stops and says why.
Two things matter more than the automation rate. One: you set the limits, not the vendor — dollar thresholds, new customers, price changes, anything you want a human eye on. Two: every action is logged with what it saw and why it acted, so when something does go wrong you can read back the decision instead of guessing.
We have written up how this one works, step by step, on the Order Entry Agent section of the supply chain page, including what gets measured from the start.
Agree the scoreboard before anything is built
The question to settle first is not technical. It is: what number are we watching, and who reads it?
Four are worth agreeing up front, and all four come out of the calculator above. The share of orders that go in without a person touching them — the honest version, where an order somebody re-opened to check does not count as touchless. The time from the order arriving to it sitting in the ERP, measured end to end rather than from when somebody started. The share of entered orders that later need a fix, which is the one that has to be watched hardest, because an automation rate bought with a worse error rate is not a saving. And the share of orders still entered before your shipping cutoff.
Take all four readings before anything changes. They are the only baseline you will ever get, and once the process has moved you cannot go back and measure the old one. A week of counting by hand is worth more here than a year of arguing afterwards about whether it worked.
Checklist: is your order entry ready for an agent?
None of these are hard requirements, but every one you can tick makes the first build shorter and the result better.
- Orders arrive somewhere consistent. A shared inbox, a portal, an EDI feed. If they arrive in four reps’ personal inboxes, start by changing that — it is worth doing whether or not you ever build anything.
- Your item master is reasonable. Not perfect. Reasonable: active items are active, obsolete ones are flagged, and there is one record per thing rather than three.
- Customer part-number history exists. Even messily. If past orders show what they called it and what you shipped, that history is the single most valuable thing you have for this.
- Pricing rules can be written down. Somebody has to be able to say, in sentences, how a price is arrived at. If the honest answer is “ask Dave”, the first job is getting it out of Dave’s head.
- There is a named owner. One person who decides what counts as an exception and what the approval limits are. Not a committee.
- You can produce fifty recent POs. A spread of real documents, including the ugly ones. The photographed one. The spreadsheet with the merged cells. Those are the useful ones.
Where to go next
If the calculator put a number on this that you did not expect, the next question is which part of the order-to-cash chain to start with — order entry, the quoting that happens before it, or the cash collection after it. The Supply Chain Agents page walks through each of them, what gets measured, and where the approval line sits.
If you would rather have the question answered than read another page: the Agent Opportunity Assessment takes about three minutes, asks about a dozen questions about how your operation actually runs, and comes back with the three agents that would pay off first in your business — with the arithmetic from your own answers, the same way the calculator above works.