EDI cost half a million dollars. The point of it now costs a conversation.
Large companies have exchanged orders and invoices machine to machine for forty years. The reason it never reached anyone else was never the technology. It was the requirement that both sides buy in.
If you have ever sold to a large retailer, you have met EDI. Electronic Data Interchange is how purchase orders, invoices, and shipping notices move between big companies without a person retyping them. It has existed since the 1970s. It works. And almost nobody outside the Fortune 1000 has it.
The usual explanation is cost. Implementations run somewhere between $250,000 and $500,000 once you count the integration work, the mapping, the VAN fees, and the year of a project manager's life. That is real, but it is not the interesting part.
The requirement nobody could get around
EDI is a standard for two computers to agree on a format. Which means it only works when the company on the other end has also implemented it. Your investment does nothing on its own. It pays off exactly in proportion to how many of your trading partners have made the same investment.
That is why it settled where it did. A national retailer can mandate EDI because it has the leverage to tell five thousand suppliers to comply. A distributor with forty customers cannot mandate anything. So the distributor keeps a person who opens PDFs and types numbers into an ERP, and everyone agrees this is just how it is.
The barrier was never that documents are hard to send. It was that both ends had to speak the same dialect before anything could happen.
What actually changed
A system that can read a document does not need the other side to agree to anything.
An invoice that arrives as a PDF, a scan, an email body, or a photograph of a delivery note taken on a phone in a yard is, to a model that can read, the same invoice. It extracts the vendor, the invoice number, the terms, the line items, and the total. It matches them against the purchase order and the goods receipt. It posts a journal entry. The supplier does nothing differently. They did not buy anything. They did not know it happened.
You get the outcome EDI was sold for, and you skip the condition that made it unreachable.
Where this earns its keep
The obvious win is accounts payable. A person opening invoices and keying them is doing work that is entirely mechanical, entirely error-prone, and entirely invisible until it goes wrong. But the same shape applies anywhere a document arrives and a record has to come out of it:
- Purchase orders from customers who will never adopt your system
- Delivery notes and packing slips photographed at the point of receipt
- Certificates, permits, and compliance documents that expire on dates nobody is tracking
- Remittance advice that has to be matched back to open invoices
- Supplier price lists that arrive as a spreadsheet with a different layout every quarter
The part that matters more than the extraction
Reading a document is the easy half. Any demo can show you fields lighting up.
The half that decides whether this is usable is what happens when the system is not sure. A confidence score per field, not one score for the page. A floor below which nothing posts unattended. A queue a person can actually work, with the document and the extracted record side by side. A journal entry that refuses to commit unless it balances.
Without those, automated intake is a machine for introducing errors faster than you can find them. With them, the clean documents post themselves and a person only ever looks at the doubtful ones. That is the entire value proposition, and it is a design decision rather than a model capability.
What it costs now
Not half a million dollars. The expensive parts of an EDI project were the negotiation, the mapping to a rigid standard, the middleware, and the fact that every new trading partner started the process again. None of those survive when the system reads documents instead of parsing an agreed format.
What remains is understanding your actual workflow, connecting to the systems you already run, and setting the rules for what posts and what waits. That is weeks of work, not quarters, and it does not scale with the number of suppliers you have.
The strange thing about this is how quiet it has been. The capability that separated large companies from everyone else for four decades stopped being a moat, and mostly what happened was nothing. The people who had EDI already have it. The people who did not are still opening PDFs.