Order entry is the process we ask to see first, because it sits between the customer and the money and it is almost always held together with copy and paste. What follows is a composite of the order-entry workflows we have mapped: no single client, but nothing invented either. If you run a distributor, a manufacturer or anything that takes orders by email, you will recognise most of it.
The method is unglamorous. We sit with the two or three people who actually enter orders, ask them to work normally, and write down every system they touch and every decision they make. We time a real sample rather than asking for an estimate, because people estimate their own processes badly and always in the same direction — they describe the clean path and forget the twenty percent that takes all the time.
The workflow as it actually ran
- 1Order arrives by email, sometimes as body text, sometimes as an attached PDF, occasionally as a photo of a fax.
- 2A coordinator reads it and works out which customer it belongs to.
- 3They check whether pricing matches the customer’s agreed terms.
- 4They check stock in the ERP.
- 5They key the line items into the ERP.
- 6They email the customer to confirm, adding the lead time by hand.
- 7They copy the order into a spreadsheet used for the weekly report.
- 8If anything is unclear, they email the sales rep and wait.
- 9The manager assembles the weekly report from that spreadsheet on Friday.
Nine steps. Two systems, one inbox and one spreadsheet. Nobody in the building considered this a problem, because everyone has always done it this way. The measured average was just under twenty-seven minutes per order, and about a fifth of orders needed a second pass.
The eleven opportunities
1. Reading the order
Extracting line items, quantities and part numbers from an email or a PDF is the single largest chunk of the twenty-seven minutes. It is also the step where AI genuinely earns its place, because the input is unstructured and varies by customer. Extraction with a confidence score, and anything below the threshold going to a person, removes most of the typing without removing the judgement.
2. Identifying the customer
This looks like it needs intelligence and usually does not. Matching on sending domain, then on a known contact, then on a fuzzy company name gets you the overwhelming majority. This is a lookup, not a model.
3. Validating pricing against agreed terms
Rules, not AI. The contract terms are already in the ERP. The reason a human does it today is that nobody has connected the order to the terms, not that the comparison is hard.
4. Checking stock
A read from the ERP that a person is performing manually because the order does not live in the ERP yet. Once step one produces a structured order, this becomes automatic and instant.
5. Creating the order record
The keying itself. This is pure cost: no decision is being made, a person is acting as a data bus between an inbox and a database.
6. The confirmation email
Templated, populated from the order record, sent on creation. The lead time that was being typed by hand comes from the same stock check in step four.
7. The reporting spreadsheet
This one always makes the room go quiet. The spreadsheet exists because management wanted a view the ERP does not give them. Once orders are structured records, the spreadsheet is replaced by a query — and the person maintaining it gets their Friday back.
8. Exception routing
Today, an unclear order becomes an email to a rep and then a waiting game with no visibility. As a queue with an owner and an age, the same exception becomes something a manager can see and chase.
9. Duplicate and re-order detection
Nobody had asked for this, and it turned out to matter. Customers re-send orders when they have not heard back. Some of those get entered twice, and the cost of an unwanted shipment is far higher than the cost of the check.
10. Rush order flagging
Urgency is expressed in prose — "need these by Friday" — which is why it gets missed at volume. Classifying it at intake and pushing it up the queue is a small model doing a small job well.
11. The unwritten rules
Every mapping session surfaces knowledge that lives in one person’s head: this customer always means the 12-pack, that account needs a PO number or accounting will reject the invoice. Writing these down as rules is the least technical item on the list and often the one that prevents the most rework.
What we would build first
Not all eleven. You want the ones that are high volume, rule-stable and low blast-radius if they are wrong. That is items one, four, five and six: read the order, check stock, create the record, confirm. Together they are most of the twenty-seven minutes, and each has an obvious human fallback when confidence is low.
The reporting spreadsheet — item seven — is the one clients push for first, because it is the most visible annoyance. It is also worth much less than it feels, because assembling it costs one person a few hours a week while order entry costs several people most of their day. Feelings about a process and the cost of a process are different things, which is exactly why we measure before we build.
The uncomfortable part
Roughly one workflow in five that we map turns out not to be worth automating. Low volume, unstable rules, or a process that needs to be changed rather than accelerated. We would rather tell you that in week one than build it and have you discover it in month six.
Do this yourself
- Pick the process that touches the most money on its way through your business.
- Sit with the person who performs it for two hours and write down every system they open.
- Count the times the same fact gets typed in more than once.
- Time twenty real items, including the awkward ones. Do not use an estimate.
- Multiply by volume and by the loaded hourly cost of the people involved. That number is what the process costs you today.
If you would rather we did it, that is what the workflow review is. You show us the process, we come back with the map and the three opportunities worth building first.