Trading and distribution: getting emailed orders into the ERP without retyping
Read orders sent by email, PDF or WhatsApp, enter them in the ERP with approval, answer stock and lead-time questions, follow up on what is pending.
Thibault Pons de Vincent
Founder of Oliverlist and VivaSync
Trading houses and distributors still take many of their orders the old way. A few lines in an email, a purchase order in PDF, a spreadsheet, a WhatsApp voice note, a photo of a handwritten list. Customers in different countries each order in their own format. Someone opens every message, identifies the customer, looks up the item codes, retypes the lines into the ERP and sends a confirmation. The same desk answers "is it in stock?" and "when will it ship?" all day, and chases whatever is pending. It is useful work, and it is predictable.
The goal is not an order desk with nobody on it. It is an order desk where nobody retypes. AI reads and prepares; a person approves what needs approving.
At Oliverlist, which I founded in 2016 without outside investors, we have automated the entire chain since 2020, from importing contacts to issuing the invoice. Between 2019 and 2023, that automation tripled our revenue per person, before generative AI entered our product in May 2023. The lesson carries over to order entry: automate what is predictable first, then use AI to read what is not. These are possibilities to adapt to your ERP and your customers, not client case studies.
Read the order, whatever the format
Orders that come through EDI or a B2B portal are already structured, so they are not the problem. The problem is everything else, sent by customers who will not change their habits, and whom you have no wish to force.
The hard part is not typing, it is reading. Every customer orders differently. One uses your item codes, another uses their own, a third writes "same as last time, but double". AI can read the email and its attachments, identify the customer and the delivery point, and match each line to your catalogue. To do that, it relies on the customer's order history and on a cross-reference table between their codes and yours.
It also spots what does not fit: an ambiguous unit, carton or piece, an unknown code, a quantity ten times the usual, a delivery date you cannot meet. Each doubt is flagged on the line concerned instead of being settled by guesswork.
Before you start, gather the code cross-references for your regular customers. They often exist already, in a spreadsheet or in your sales team's memory. That groundwork is what makes order reading reliable from the first weeks.
Enter it in the ERP, with an approval step
The order is created in the ERP as a draft, or placed in an approval queue. Before that, AI runs the checks a person would run: price against the customer's terms, minimum order quantities, credit limit, stock availability. A clean order goes through after a quick check. An order with an anomaly reaches the right person with the anomaly already explained. AI does not need to replace your ERP: it connects to it or, failing that, prepares an import file. Which one depends on what your ERP accepts.
I recommend starting with human approval of every order and counting, for a few weeks, the corrections needed line by line. When that number stays low and stable for a type of order, you can let that type through without approval and keep control of the rest. The order confirmation then goes to the customer with every line spelled out, so they can flag a mistake right away.
Answer stock and lead-time questions
"Do you have it in stock?", "When will it ship?", "Can it make Thursday's container?": these questions come back every day. AI can answer them from your ERP data: available stock, reservations, open purchase orders, usual shipping times. It drafts a precise reply and can suggest an alternative when an item is out of stock. These answers depend on accurate stock figures in the ERP. If your inventory is approximate, fix that first, or AI will repeat your errors with confidence.
The line is simple. A factual answer, such as a stock quantity, can go out automatically if your data is reliable. A commitment on a date that depends on a supplier, or a product substitution, goes through a salesperson. That is the rule I apply at Oliverlist: AI prepares, and a person approves anything that commits the company.
Follow up without having to think about it
Follow-ups are the work everyone postpones. AI can keep them current: quotations with no answer, incomplete orders, supplier confirmations that have not arrived, late inbound shipments that will delay your own customers. It can also spot a customer who usually orders every three weeks and has placed nothing for six, then draft a message for their account manager.
Overdue invoices follow the same logic, with one condition: AI prepares the reminder and a person approves it before it goes out, because it touches the customer relationship.
Key takeaways
- With emailed orders, the hard part is reading, not typing. That is where AI makes a difference.
- Start with human approval of every order, count the corrections, then relax the control one order type at a time.
- Factual stock answers can go out automatically. Delivery commitments and substitutions go through a salesperson.
- Automate the predictable first. AI comes next, to read the rest.