EDIFACT, HL7v2, and the Protocols That Refuse to Die

Every few years someone declares that REST and JSON have won, that the old formats are finally on their way out, and that soon everything will speak the same clean modern protocol. Every few years they’re wrong in the same way. The old formats don’t die. They just quietly keep running the parts of the economy nobody’s willing to risk rewriting.

We know this because bridging old to new is a large part of what integration engineering actually is, and has been for the 20 years we’ve been doing it. Here are the protocols that were supposed to be gone by now, why they aren’t, and what it means for anyone building systems that have to touch the real world.

EDI still moves the goods

If you buy a physical product, its journey through the supply chain was almost certainly coordinated by EDI, Electronic Data Interchange, mostly in the EDIFACT flavor in Europe and Asia and X12 in North America. EDIFACT is the UN standard, approved as ISO 9735 back in 1987 and maintained under UNECE ever since. It predates the web.

And it still dominates. Industry figures put EDI at around 78% of all B2B electronic transactions as of recent counts. That’s the lion’s share of B2B digital commerce, and one of the slowest-growing channels, which is exactly the profile of a large, entrenched installed base. Purchase orders, invoices, shipping notices, settlement files (the boring, high-volume documents that make commerce work) move over a format designed when bandwidth was scarce and every character cost money.

Nobody loves parsing EDIFACT. Its segments and delimiters assume you own the spec. But a retailer and its thousands of suppliers all speak it, and no single party can switch unilaterally, so it persists by network effect. When we connect a modern order system to a supply chain, half the work is a clean, well-monitored EDIFACT boundary. The format isn’t going anywhere, so we build the adapter and own it properly.

HL7 v2 still runs the hospital

Healthcare has the same story with sharper stakes. HL7 Version 2, the pipe-and-caret delimited text format where a message segment starts MSH|^~\&|, is what the National Library of Medicine calls “the workhorse of electronic data exchange in the clinical domain.” Their number: as of 2024, about 95% of U.S. healthcare organizations and more than 35 countries use it.

There’s a modern successor, FHIR, built on REST and JSON, and it’s genuinely good. Regulation is even pushing it into mandatory use for certain data exchange. But FHIR hasn’t replaced HL7 v2 inside the hospital. It mostly sits at the edges, for patient-facing APIs and external access, while the lab results and admissions and orders still flow over v2’s delimited messages carried on MLLP over TCP. The installed base is enormous, the systems are life-critical, and “rip out the messaging layer that currently works” is not a sentence anyone says in a hospital.

So integration in healthcare means fluency in both: bridging HL7 v2 to FHIR, absorbing the delimiter handling and encoding quirks at the boundary so the clinical systems keep working while the modern APIs get their clean JSON.

SOAP and the file drop, quietly everywhere

Two more that were declared dead and aren’t.

SOAP, with WS-Security, still runs in banking, payments, ERP, and government. The reason is specific: SOAP offers message-level signing and encryption, and in regulated environments where a legacy service can’t be swapped out on a whim, that’s a compliance choice, not a nostalgia one. Most new APIs are REST, and enterprises run both side by side, because “new is REST” doesn’t mean “old gets deleted.”

And then there’s the humblest pattern of all: the file drop. A CSV or fixed-width file, pushed over SFTP on a nightly schedule. Hohpe and Woolf named File Transfer as a canonical integration pattern for a reason. It endures on format flexibility, mature tooling, and standardized layouts, not on technical sophistication. Settlement files, partner feeds, batch ETL. An astonishing amount of mission-critical business still moves as a file that lands in a directory at 2 AM. The modern challenge isn’t connectivity; it’s automating and monitoring the drop so you know, before the business does, when last night’s file never arrived.

Why they don’t die

Pull the thread and the same four forces show up under every one of these.

Installed base. Millions of endpoints already speak the format. The cost of moving them is enormous and the benefit is invisible to the people paying.

Regulation. In healthcare, finance, and government, the format is often written into a requirement. You can’t unilaterally choose something newer.

Network effects. EDI’s power is that everyone’s trading partners already use it. No single company can switch alone, so nobody switches.

Risk. These systems run things that hurt when they stop: payments, patient records, supply chains. “It works, don’t touch it” is not laziness here. It’s correct risk management. The safest change to a system that processes payroll is often no change at all.

What this means for how you build

The mistake is to treat these formats as debt to be eliminated. They’re not going to be eliminated on your project’s timeline, and planning as if they will is how integration projects overrun.

The right model is the boundary. You don’t rewrite the hospital’s HL7 v2 or the supply chain’s EDIFACT. You build a clean, monitored translation layer at the edge (old format in, modern shape out, and back again), and you absorb the ugliness there so the rest of the system stays clean. The delimiter parsing, the character-encoding surprises, the schema drift between the partner’s version and yours: all of it gets contained at that one boundary, tested and observable, instead of leaking into every service that touches the data.

A format from the 1980s still moves most of the world’s B2B transactions. That’s not a failure of progress. It’s the actual shape of the world integration engineers work in: new systems that must talk to old ones that aren’t going anywhere. The job was never to replace the old protocols. It’s to build the bridge, keep it standing, and know both sides well enough that the seam never shows.