One device. Five regulatory identities.
A 510(k) number in the US. A different name on the EU MDR certificate. A re-branded version for the distributor in Mexico. The manufacturer's internal SKU. And the name the hospital procurement system actually uses.
Now try to answer a simple question: "How is our product positioned versus competitors in three markets?"
You can't. Because no single system knows those five identities are the same device. Your competitive intelligence is fragmented across the exact seams where the entity gets renamed.
This isn't messy data. It's messy identity — by law.
Most "dirty data" is an accident. Medtech's version is a regulatory requirement. The device legally becomes a different registered object every time it crosses a border or changes a distributor — and the rules guarantee it.
Start with the identifiers. The FDA's GUDID and the EU's EUDAMED are separate, independently operated UDI systems, with no binding US↔EU crosswalk between them. [1] They don't even accredit the same issuing agencies: the European Commission designates four (GS1, HIBCC, ICCBBA, IFA), while the FDA accredits only three (IFA is not an FDA issuing agency). [2] So the same device gets a different Device Identifier string in each market as a structural consequence of separate per-market assignment — not because anyone made a mistake.
Then the identifier fragments within a market. Under 21 CFR 830.50, "whenever you make a change to a device… and the change results in a new version or model, you must assign a new device identifier," and "whenever you create a new device package, you must assign a new device identifier to the new device package." [3] New version → new ID. New pack size → new ID. The EU mirrors this per packaging level. The scale this produces is real: GUDID held 5,129,241 device-identifier records in its June 2026 release — each a distinct DI that has to be resolved against the others. [4]
Even the categories don't line up
Suppose you try to compare devices by type instead of by ID. That breaks too, because each region speaks a different classification language. The EU registers devices under EMDN, a seven-level hierarchical nomenclature spanning 22 categories. [5] The FDA uses its own product-classification codes — roughly 7,000 of them. [6] Much of the rest of the world uses GMDN, with more than 7,000 generic device groups again. [7] Three taxonomies, no clean mapping. The EU even adds a "Basic UDI-DI" — a regulatory grouping key that, by MDR design, never appears on the device label and has no US equivalent. [1]
Relabeling multiplies the identities again
Now add distributors. A private-label device can be physically identical to the OEM original except for the name — yet each relabeler is its own FDA "labeler" that must assign its own UDI. [8] Under EU MDR Article 16, distributors and importers can relabel, translate, or repackage a marketed device for a new market and must add their own name to it. [9] So one physical device legitimately carries a distinct identity under each brand it's sold through. And because EUDAMED's UDI registration was effectively voluntary until its modules become mandatory on 28 May 2026, the EU side has years of incomplete cross-market identity data baked in. [10]
Why this is an entity-resolution problem, not a search problem
Entity resolution is the task of deciding which records — across systems, vocabularies, and owners — refer to the same real-world thing: the same molecule, the same device, the same patient, the same product. It sounds like deduplication. It isn't. Deduplication assumes the duplicates are obvious. Entity resolution is the hard version, where the "same thing" wears a different name, code, and owner in every system it touches, and nothing in your data records that they're connected.
Two properties make it genuinely hard. First, it's quadratic: to know whether any record is a duplicate, you in principle have to compare it against every other record. As Papadakis and colleagues put it in the VLDB literature, entity resolution "typically scales to large volumes of data through blocking… each entity has to be compared with all others in order to find its matches." [ER-1] A thousand records is half a million comparisons; a million records is half a trillion. Double the data and you roughly quadruple the work.
Second, the right answer for one record depends on all the others — entity resolution needs global consistency. Whether record A and record B are the same can hinge on what C, D, and E say. This is exactly where throwing a bigger language model at it disappoints: large models reason locally, one context window at a time. They will confidently link two records that, seen against the whole dataset, obviously shouldn't be — and miss links that only become clear once you hold the full graph in view. The semantic layer that makes messy data usable isn't built by a model guessing one row at a time. It's built by deciding, consistently and across the whole set, that these N records are one thing.
Who expands fastest
Most teams "solve" this with a spreadsheet maintained by one person who knows where the bodies are buried. That works until they leave — and then the map of which five identities are the same device walks out the door with them.
The companies expanding into new markets fastest aren't the ones with the best product. They're the ones who can actually see their own product clearly across every market it's sold in — who have resolved the 510(k), the MDR certificate, the distributor brand, the SKU, and the procurement name into one device, and kept that resolution current as new versions and pack sizes spawn new IDs.
If your product crosses borders — how many "identities" does it have, and who's keeping track?