Legal entity suffixes and the exhibitor records they split in two

Pull the exhibitor account list for one show and sort it by company name. Somewhere in it, usually more than once, sits a pair like this: Kestrel Fluid Systems Ltd with a stand in hall 3 and a renewal date in March, and Kestrel Fluid Systems Inc with a stand in hall 6 and a different account owner.

Two accounts, one buying decision, and possibly one firm. Legal entity suffixes produce this split constantly, because the suffix travels with whichever contracting entity signed the paperwork and your matching logic reads it as part of the name.

The handling is short to describe and the detail in it matters. Take the suffix off before you compare, and keep what you took off.

What the suffix is doing in the string

A legal form designator names the type of entity under the law that formed it. Under ISO 20275, published in 2017 and maintained by GLEIF as the reference list for entity legal forms, each form is identified by a four character code tied to the jurisdiction that recognises it, so a Dutch BV and a German GmbH carry different codes, because a legal form has meaning only inside the law that created it.

For matching purposes that has two consequences.

The designator carries no information about which firm this is. Every private limited company in England ends in Limited or Ltd, so the token contributes nothing towards distinguishing one from another and everything towards making two records of the same firm look different.

The designator carries real information about where the entity was formed. Ltd points at the United Kingdom and a handful of other common law jurisdictions. Pty points at Australia or South Africa. Oy points at Finland. Sp. z o.o. points at Poland. GmbH is used in Germany, Austria, Switzerland and Liechtenstein, so it constrains the answer without settling it.

Those two facts pull in opposite directions, which is why the answer is to separate them rather than to choose.

How much does stripping actually move the score?

Work the pair from the account list.

Kestrel Fluid Systems Ltd tokenises to kestrel, fluid, systems, ltd. Kestrel Fluid Systems Inc tokenises to kestrel, fluid, systems, inc. They share three tokens out of five distinct tokens, so the overlap is 3 over 5, which is 0.60.

A threshold at 0.70 rejects that pair. A threshold at 0.55 accepts it, and also accepts a great many pairs you do not want, because 0.60 is where two genuinely different firms sharing a couple of generic words also live.

Now strip the designators. Both names reduce to kestrel, fluid, systems. The overlap is 3 over 3, which is 1.0, and the pair is unambiguous under any threshold you would sensibly set.

One token, removed, moved a pair from rejected to certain. Run it across the directory and the effect compounds: in a file where 1,900 of 2,400 exhibitor names end in a recognised legal form, every comparison between two of those 1,900 was previously carrying a token of pure noise, and for short company names that token was a large share of the string. Marchetti SRL against Marchetti SpA shares one token out of three, which is 0.33, and after stripping it is 1 over 1, which is 1.0. The shorter the trading name, the more damage the suffix was doing.

The list is longer than anyone expects

Six designators cover most of a European exhibitor directory and they are not enough.

The February 2026 version of the GLEIF list covers more than 3,600 entity legal forms across more than 200 jurisdictions, and while your portfolio will never see most of them, the number is a useful corrective to the instinct that this is a short lookup table you can write from memory in an afternoon.

Spelling variation makes it worse. A German limited company appears as GmbH, G.m.b.H., mbH, and occasionally as Gesellschaft mit beschraenkter Haftung written out. A British one appears as Ltd, Ltd., Limited, and Co Ltd where the Co is doing designator work too. A French one appears as SARL, S.A.R.L. and Sarl. Any punctuation stripping that ran during company name normalisation in J17 has already collapsed some of this, which is exactly why the suffix step runs after normalisation and not before.

Build the list from your own data rather than from an imagined complete list. Take the last token of every company name in the directory, count the frequency, and read the top hundred. The designators sort themselves to the top because they repeat, and you will find the ones your portfolio actually contains in about twenty minutes, including the two or three from markets you forgot you sell into.

Keep the list as data, in a table with a column for the designator, a column for the jurisdictions it implies and a column for the date somebody last checked it. A designator list buried in code gets extended by whoever is debugging a merge that afternoon, and nobody else finds out.

Keeping the suffix as a jurisdiction signal

Strip the designator into its own column. Never delete it.

Then look at what the stripped column produces. Kestrel Fluid Systems Ltd against Kestrel Fluid Systems Inc scores 1.0 on the trading name and disagrees on legal form, which means one of two things is true. Either your systems recorded one firm under two designators, which happens whenever a salesperson types what the badge said instead of what the contract said. Or these are two separate legal entities inside one group, a British operating company and an American one, which both genuinely exist, both sign their own contracts, and both need their own account.

That pair belongs in a review queue with the disagreement shown, because a person can settle it in seconds with information that is not in the string. The billing address settles it. The purchase order settles it. Merging the pair automatically loses a real entity, and leaving it alone loses the connection between them, so the useful outcome is a link that keeps both records and records the relationship, which is J19's problem.

Talburt puts the two subjects together in Entity Resolution and Information Quality, published by Morgan Kaufmann in 2011, running from the principles of each through the resolution models to the systems that implement them. Reading a suffix rule in that frame makes the answer straightforward. What you strip is an attribute of the entity, and it stays useful after the comparison that needed it removed.

When should you not strip?

Three cases, and each one has bitten somebody.

The designator is the name. The Limited is a real retail brand. Firms trade as Company, as Group, as Holdings, and stripping the word leaves a name that means something different or nothing at all. The rule that prevents this is a length check: never strip if the remainder would be empty, a single character, or a single token from your stopword list.

The designator is not at the end. Some jurisdictions place it at the front, and some company names contain a designator mid string as part of a longer trading name, such as a firm whose name includes the words Trading Company followed by a region. Strip only from the end of the string, and treat two token forms such as Pty Ltd and Co Ltd as single units so that you do not remove Ltd and leave a dangling Pty.

The designator is genuinely distinguishing. In a small number of directories, two unrelated firms share a trading name in different jurisdictions and the designator is the only thing separating them. This is rare, it is real, and it is caught by the same review queue described above rather than by a cleverer rule.

Where this stops

Stripping suffixes makes the trading names comparable and tells you nothing about whether two entities are one commercial relationship.

A group with four contracting entities in four countries will now match beautifully on the trading name and still have four legitimate accounts, four billing addresses, four sets of terms and, quite often, four different people signing. Any system that responds to a 1.0 similarity by merging them has destroyed the record of who actually contracted, which is the number your finance lead reconciles against.

The second limit is that the jurisdiction signal is weaker than it looks. SA appears across France, Spain, Switzerland, Luxembourg and much of Latin America. AB is Swedish, and also two initials somebody used as a trading name. Treat the designator as evidence that narrows the country field, never as the country field itself, and store it as what it is: the legal form somebody wrote on a contract.

The third is that none of this survives a rebrand. When the operating company changes name and keeps its registration number, the trading name changes and the designator does not, so your matching gets worse for the firm at exactly the moment it becomes most visible in your renewal pipeline. That is an argument for carrying a stable external identifier where you can get one, and for treating token overlap on the stripped name in J13 as one signal among several rather than as the answer.

Take the last token of every company name in your exhibitor directory this week, count the frequencies, and count how many rows end in a recognised designator. Then recompute the token overlap for ten known duplicate pairs with and without the designator. The two sets of numbers make the case on your own data in under an hour, and the stripped column is the first genuinely new field the unified data work adds to your directory.