Company ownership data
Who owns what
The data is public. It is scattered across registers that do not share an identifier, so nobody has joined it, and joining it is the work. This is that work done on food technology — the sector is the worked example, not the subject.
The same engine, pointed at tech: Really European? traces the products on Europe’s buy-European directories to their controlling owners, and the DORA provider list cuts the same rows for the register of information.
Loading the dataset…
What the resolved data looks like
Both figures are drawn from the published file rather than redrawn by hand, so they move when the data does.
Ownership is extremely top-heavy
The largest group holds 230 legal entities; the median group holds 18. Any analysis that treats a “company” as one row is counting the tail and missing the head.
Most groups are small; a few are enormous
64% of the 58 groups hold fewer than 25 entities. The distribution is the reason a name-matching threshold tuned on the average group fails badly at both ends.
Food technology is one input, not the method
Worth being precise about, because “it works for any sector” is the kind of claim everyone makes. Here it is an architectural fact rather than a promise: exactly one file in the pipeline knows what food technology is — a list of company names, written the way a person would type them. The registers it queries, the rules that cluster names into groups, the rollup to reported parents, the scoring and the published error cases are all untouched by the sector.
Which is also why this sector was chosen: large groups sitting behind subsidiary tangles, so the resolution has visible work to do. A sector with simpler ownership would have produced a prettier number and proved less.
What changes when you point it somewhere else
Only the registers stacked on top. These are not built — none of them is a different kind of problem, which is rather the point.
- Pharmaceuticals
- Marketing authorisations, clinical trial registries and orphan designations, keyed to the same corporate groups.
- Aerospace and defence
- Design and production approvals, airworthiness certificates, and the supplier chains visible through public tenders.
- Construction and materials
- Environmental declarations from programme operators, CE marking and declarations of performance.
- Energy and utilities
- Generation licences, grid connection registers and subsidy allocations by operator.
- Logistics and shipping
- Vessel and fleet registers, operating licences, and beneficial ownership where it is disclosed.
- Any sector at all
- Company registers, VAT and LEI data, public tenders and grant funding exist for every industry in Europe. None of them share a key.
Take the dataset and check it
No sign-up. The third file is the one worth opening first: every row the method could not place, or placed on weak evidence. A dataset that publishes only its successes is asking to be trusted rather than checked.
- entities.csv
One row per legal entity, with how it was grouped and why
- groups.csv
One row per group: hierarchy, countries, research funding
- uncertain.csv
Everything the method could not place confidently
What this does not tell you
Coverage is not completeness. An entity without an LEI is invisible here, and plenty of subsidiaries have no reason to hold one. A group showing four entities may well have forty.
Reported parents are self-reported. They are the company's own filing, which makes them authoritative about intent and occasionally out of date about fact.
The funding join is a name match. The grant data carries no company identifier. The join is deliberately conservative, so it misses grants before it invents them — under-counting, not over-counting.
No patent or IP layer. Every patent register worth using needs an API key and carries its own licence terms for derived datasets. Scoped, not built.
Your sources, merged the same way
This took public registers that share no identifier and produced something checkable. The same problem usually turns up inside a company first: a CRM, an ERP and a billing system that each believe a customer is called something different. The useful version resolves your sources, in your sector, and hands back a pipeline you own.
Questions people ask
- What is entity resolution?
- Entity resolution is deciding which records across different sources refer to the same real-world thing — the same company, customer or part — when they share no identifier and are spelled differently in each. It is the step that makes a join possible at all.
- How do you find a company's parent company?
- Where an entity holds a Legal Entity Identifier it may have reported its direct and ultimate parent to the LEI register, and that filing is authoritative because the company made it. Where no LEI or no filing exists, the relationship has to be inferred from names or corporate filings, which is a guess and should be labelled as one.
- Is there a free company ownership database?
- The underlying registers are free: the global LEI register publishes legal names, jurisdictions and reported parent relationships under an open licence, and EU research funding is published as open data. What does not exist for free is the joined version, because the registers share no common key.
- Why is matching company names so difficult?
- Legal forms differ by jurisdiction without changing the company, brands and registered names rarely agree, groups file under several parents, and short names are ordinary words. Matching on names alone will put a company in India called MARS INDUSTRIES into the same group as a confectioner unless something stops it.
Try it on your own file first
Every tool here runs in your browser and uploads nothing, so you can see what we would see before telling us anything at all. When it is worth a conversation, send the real extract and the question you actually need answered.