Guide

How to automate customer order entry into your ERP

A method for getting orders that arrive by email, PDF, photo or WhatsApp into your ERP as drafts, with AI reading every line and a person checking only what is unclear.

In many distribution businesses, someone spends a good part of the day opening emails, reading orders and typing them into the ERP. Orders turn up in every possible shape: an email with the lines in the body, a PDF from the customer’s own system, a photo of a handwritten sheet, a WhatsApp message. Every line typed wrong is paid for later in returns and freight.

This guide explains how to automate that order entry: the workflow step by step, the piece that matters most (the cross-reference table), what to leave for a person to check, and how to measure whether it was worth it.

Why it takes so long

Typing in one order looks quick, but it adds up:

  • Volume: dozens or hundreds of orders a week, each with several lines.
  • Every customer is different: each one writes their own way and calls your products by their own names or codes.
  • Peaks: Mondays, month-end and the days before a holiday, when everything arrives at once and most mistakes happen.
  • Costly mistakes: a wrong item or quantity ends in a delivery that has to be redone.

The workflow, step by step

  1. Collect orders in one place: an orders inbox, a WhatsApp Business number or a web form.
  2. Read each order with an AI model, which pulls out the customer, delivery date, delivery address if there is one, and every line with its product and quantity. It doesn’t matter whether it comes in the email body, a PDF or a photo.
  3. Identify the customer from the sender, the phone number or the tax ID on the document.
  4. Translate item codes from the customer’s to yours with the cross-reference table.
  5. Check with rules: the product exists and is active, the quantity respects minimums and pack sizes, there is stock, and the order isn’t a duplicate.
  6. Create the order in the ERP as a draft, with a link to the original document.
  7. Review only what is doubtful: lines with low confidence or that fail a check go to a person, who corrects them and confirms the order.

It is the same pattern we use for supplier invoices, explained in how to automate invoice processing with AI.

The cross-reference table

This is the piece that decides the result. A customer orders “box of 4x40 stainless screws” and in your catalogue it is item SCR-440-SS. An AI model can suggest the most likely match, but what you can rely on is a table that records, for each customer, what they call each of your products.

It works like this: the first time a new description shows up, the system suggests an item and a person confirms or corrects it. That match is saved for that customer, and the next order with the same description goes through without review. Within a few weeks, most lines from your regular customers resolve on their own.

What to check by hand

Set a confidence threshold: above it a line goes straight through, below it a person sees it. At the start, review everything for a few days, compare it with what would have been typed by hand and adjust the threshold. Beyond that, always send these to a person:

  • New customers.
  • Orders with special prices or terms outside the price list.
  • Quantities far above what that customer normally orders.
  • Any order where the system can’t tell who the customer is.

What you need

  • Your ERP, which stays the source of truth. If it has an API (NetSuite, QuickBooks, Odoo and most modern systems do), orders are created directly. If not, most ERPs can import orders from a CSV, Excel or XML file, and the workflow produces that file.
  • An automation platform such as n8n, Make or Zapier, or custom code if the volume or the logic calls for it.
  • An AI model to read the orders.
  • The WhatsApp Business Platform, if you want to include that channel.

Customers who already send orders by EDI don’t need any of this: their orders arrive structured. The automation is for everyone else.

How to measure the result

Before you start, measure for a week:

  • Average minutes per order, from arrival to sitting in the ERP.
  • Orders per day, and on peak days.
  • Mistakes found afterwards: redelivered orders, credit notes, complaints.

Afterwards, add the share of orders that go through untouched. With those figures you can work out whether it pays off, as we explain in how much process automation costs.

Common mistakes

  • Trying to cover every channel at once. Start with the one that brings in the most orders.
  • Confirming orders without review from day one, before you know how the system handles your customers.
  • Skipping the cross-reference table and trusting the AI to guess the item.
  • Creating orders outside the ERP, in an in-between spreadsheet nobody maintains.

Where to start

The page on automation for wholesalers and distributors shows the processes that usually come next, such as stock alerts and payment reminders. This is the kind of project we take on in AI automation and invoice, contract and paperwork automation. In the free 30-minute audit we look at your real orders and tell you which part can be automated; get in touch through the contact page.

Frequently asked questions

How accurate is AI at reading orders?

It depends on how tidy they are. A PDF from the customer’s own system reads almost perfectly; a photo of a handwritten sheet, less so. That is why the system gives each line a confidence score and only lets through the ones above the threshold you set.

What if our ERP has no API?

Most ERPs can import orders from a CSV, Excel or XML file. The workflow produces that file and someone imports it, or the import runs on a schedule. It is less immediate than an API, but it still removes the typing.

Is it worth it with only a few orders a day?

Do the maths: orders a day times the minutes each one takes, plus what mistakes cost you. If they are few and simple, it may not pay off; if they are many, long or often wrong, it usually pays off quickly.

What about customers who order by EDI?

Their orders already arrive structured and don’t need AI. The automation is for everyone else: customers who write emails, send PDFs or order over WhatsApp.

Keep reading

Keep reading

How we can help

  • AI automation

    Tasks done by hand between your programs start doing themselves, and AI reads emails and documents for you.

  • Invoices, contracts & paperwork

    Invoices, contracts and forms that read themselves: the data reaches your business software and a person checks anything doubtful.

  • Connecting your tools & custom software

    We get your programs passing data to each other, and build a tool of your own when nothing off the shelf does the job.

Which process would you automate first?

Tell us which task eats your team’s time. In the free audit we tell you what to automate first, with which tools, and what it would cost.