Saimaa ringed seal, Rantasalmi, Finland. Photo: Jan Ebr and Ivana Ebrová, CC BY 4.0
Blog ·
We found 289,225 dead companies in our own database. Then we checked the registers.
This is our own data-quality audit, published with the numbers that make us look bad. The mechanism behind them affects every company database, and almost nobody measures it.
What happened
Our Finnish company data was built from the national register, PRH, which is the right primary source. The import wrote every record as active, and nothing ever went back to ask whether a company had since closed. Records went in, nothing came out. After a year, more than a third of the data was companies that no longer existed.
The 50 968 dead companies with a working domain are the dangerous ones. They have a name, a website, an industry code and often a phone number, so they passed every quality check we had. Only the register could tell us they were closed, and we were not asking it.
Registers are archives, not lists of living companies
National registers keep every company ever registered, with a status field for the closed ones. Read the register without that field and you get the living and the dead together. We counted the official files on 22 July 2026, record by record:
| Register | Records | Closed or removed |
|---|---|---|
| Singapore, ACRA | 2 102 483 | 70.6 % |
| Sweden, Bolagsverket | 2 857 430 | 60.5 % |
| France, INSEE Sirene | 29 803 585 | 43.9 % |
| Finland, PRH | 821 699 | 43.9 % |
In France, 13 080 910 legal units carry the state "cessée". In Sweden, 1 728 033 organisations have a deregistration date. In Finland, the register's company endpoint returned 821 699 companies while the file of currently registered companies contained 461 107. In Singapore, only 617 218 of 2 102 483 entities were live.
Not every register shows the dead. The UK Companies House monthly file and Norway's Brønnøysund download contain only current companies. That causes the opposite problem: a copy loaded once ages silently. Estonia's open data lists 374 810 current entities, of which 9 042 were in liquidation and 662 in bankruptcy: alive on paper, finished in practice.
What it cost us
| Where it showed | What we found |
|---|---|
| Company search | 289 225 dead companies shown as active (38.6 %) |
| Caller ID index | 268 242 phone numbers attached to closed companies |
| One cold email batch | 545 of 1 000 companies no longer existed (54.5 %) |
| Industry counts | Retail shown as 37 616 companies, really 24 711 |
The email batch is the expensive line. Mail to a closed company bounces, and bounces damage the sending domain's reputation for the mail that would have worked. The caller ID line damages trust: in Finland, a disconnected business number is often reassigned to a private person.
Check your own list in ten minutes
Take 100 random companies from the list you are about to use, look each one up in its national register by company number, and count how many are closed. If it is more than two or three, you have the problem we had.
| Country | Source | Field to read |
|---|---|---|
| Finland | PRH open data, YTJ API v3 | companyForms endDate |
| Sweden | Bolagsverket bulk file | avregistreringsdatum |
| Norway | Brønnøysund Enhetsregisteret | slettedato |
| United Kingdom | Companies House basic data | CompanyStatus |
| France | INSEE Sirene, StockUniteLegale | etatAdministratifUniteLegale |
| Estonia | Äriregister open data | ettevotja_staatus_tekstina |
| Singapore | data.gov.sg, ACRA entities | entity_status_description |
Match on the company number. Matching by domain alone gave us a false positive: we marked a Finnish company dead because its domain matched a struck-off record, and the company was trading normally under another registration.
What we changed
Every night the pipeline downloads the current register for each country, marks every company missing from it as terminated, and restores anything that reappears. Search, industry counts, caller ID and lists all read the same status field. If a download returns implausibly few companies, the job stops without touching the data, so a truncated file cannot mark a whole country dead.
The first run across seven countries marked 844 549 companies as closed: 289 225 in Finland, 506 015 in France, 25 921 in the UK, 7 965 in Singapore, 6 968 in Norway, 6 792 in Sweden and 1 683 in Estonia.
Situation on 14 October 2026: the Finnish data holds 401 781 active companies and 290 830 marked as closed. A further 65 472 are marked dormant, in liquidation or bankrupt. See the coverage page.
Method
Register figures were downloaded on 22 July 2026 from each authority's open data and counted in full, not sampled: INSEE StockUniteLegale (1 July 2026 edition), Bolagsverket bulk file, PRH YTJ API v3 and its file of registered companies, Äriregister open data, data.gov.sg ACRA entity datasets, Companies House monthly basic data and Brønnøysund Enhetsregisteret. Swedish and Finnish figures are deduplicated to unique organisations. A random sample of the companies we marked closed was rechecked one by one against the Finnish register, and every one was confirmed closed. Free to cite with a link to this page.
Questions
How many companies in a business register are closed?
On 22 July 2026: 70.6 % in Singapore's ACRA datasets, 60.5 % in Sweden's Bolagsverket, 43.9 % in France's Sirene and 43.9 % of what Finland's PRH company endpoint returned.
Why do dead companies stay in B2B databases?
Because importing data and rechecking it are two different jobs, and usually only the first is built. A dead company keeps its name, domain and phone number, so it passes automated quality checks. Only the register's status field tells the truth.
Does this affect Apollo, ZoomInfo or other databases?
We did not audit their records, so we make no claim about them. Any pipeline that loads a register without rereading its status collects closed companies. Test a sample of any list against the register yourself.