A distributor stops retyping orders: from email to ERP in 2 minutes

Challenge

The company distributes electrical supplies to installers and small building contractors. It receives around 380 orders a month, almost all by email: some typed into the body of the message, some as a PDF exported from the customer’s own software and a few as a photo of a handwritten sheet.

The managing director knew that “the office is always flat out”, but not where the hours were going. He had looked at a couple of AI tools and none of them fitted his ERP. That is why we started with a process audit rather than a project.

Over three sessions we mapped orders, invoicing, collections and phone enquiries. The figure that changed the conversation: two people were spending about 34 hours a month just moving orders from email into the ERP. Each order took 12 minutes on average to read, look up product codes and key in line by line. And one in ten had an error that ended in a return or a phone call.

The audit also ruled things out. Invoicing was already handled well by the ERP and was not worth touching. Collections were low volume. The plan was clear: fix order intake first.

Solution

After the audit we ran a small, fixed-scope implementation. The full audit fee was credited against the quote.

The workflow runs like this:

1. Orders arrive at a dedicated orders@ inbox. Customers did not have to change anything: one address was simply forwarded.
2. Claude reads the email and its attachments (text, PDF or image) and extracts customer, product codes, quantities and delivery date.
3. Each line is matched against the ERP catalogue. When a customer uses its own naming, the agent checks a mapping table that fills itself in whenever someone corrects a line.
4. A draft order is created in the ERP. Nothing is confirmed automatically.
5. Lines with low confidence (ambiguous code, odd quantity) are highlighted so the reviewer checks them first.

A Google Sheet logs every order processed, the review time and any corrections. That is how we measure the saving without relying on impressions. The same approach is described in automating lead intake into the CRM and AI agents for back office operations.

Result

Three months after go-live:

• 32 fewer hours a month spent typing. That time went into calling customers about pending orders and preparing quotes.
• Reviewing an order dropped from 12 minutes to about 2.
• Errors in product codes and quantities fell by 70%. Most of the ones left come from orders that were ambiguous to begin with.
• 85% of orders are confirmed without changing a single line.

“We did not want a huge project. We wanted to know where we were losing time and fix that first.” Managing director of the distributor (anonymised)

What I would do differently

I would have built the product code mapping table earlier. In the first two weeks, most corrections came from the same five customers with their own way of naming products. Loading their mappings up front would have shortened the start-up period.