Six per cent of one day’s new companies shared a single address

A company appears in the commercial register on a known date, with an address, a stated purpose and its officers. At that moment it has no accountant, no auditor and no corporate counsel, and it is about to need all three.

It is the shortest window in B2B, and the state publishes it for free, every working day.

The need is certain rather than probable. A company with employees will need payroll; a company above the size thresholds will need an audit; every company needs an annual filing. Nobody has to be persuaded that the category is relevant.

What is uncertain is only who gets the work, and that is decided in the first weeks, usually by whoever asked first. This is one of the few signals where being early beats being better.

Thousands of companies form every week and most are worth nothing to an adviser. Holding vehicles, dormant shells, single-purpose entities. And a great many registered by formation agents for people who will never trade.

So the whole vertical turns on one filter. We expected it to be legal form, share capital or stated purpose. It is none of those. It is the address.

We pulled every UK incorporation for a single day: 2,492 companies.

147 of them shared one address. A single street in London, a formation agent's mail service, accounting for 6.4% of an entire country's new companies in one day. Widen it slightly and 12.4% sat at addresses that saw ten or more incorporations that same day.

Now the useful part. Matching the literal string “c/o” catches 0.8% of them. The mail-drop addresses do not say c/o. They are ordinary street addresses that happen to be shared by hundreds of companies.

So the filter is not a string match, it is a count: group the day's registrations by address and treat concentration as the signal. That is an order of magnitude better than the obvious approach, and it costs one aggregation.

Worth putting beside the commercial-real-estate notebook, where a company registered at somebody else's address is exactly what you want: it has a legal presence and no premises, so it needs space.

Here the same fact means the opposite. A company at a formation agent's address is not a business without an office. It is frequently not a business at all.

The signal is identical and the conclusion inverts, because the buyer is different. This is the argument against buying generic “intent data”: the interpretation is the product, and it is not transferable between the people paying for it.

Switzerland. The gazette exposes new register entries as a free, keyless, date-queryable feed with structured detail: address, purpose, capital, and whether the company opted out of an audit, which is itself a qualifier. We counted 223 new entries on a single day in August and 37,058 so far this year.

United Kingdom. Companies House offers an advanced search with a CSV export that needs no API key at all. That is how we got the full day above.

Germany is closed, by statute rather than by oversight. The registry regulation requires retrieval one register sheet at a time, the terms cap it at sixty retrievals an hour, and there is no API. The one open bulk dataset is six years stale. This is not a gap we failed to find a way around; the rule is the mechanism.

The UK CSV export writes its industry-classification column as a bracketed list, and it does not quote it. So a company with two classification codes emits an unquoted comma inside a field.

A standard CSV parser does not error. It shifts every subsequent column on that row and carries on. In our day of 2,492 rows, 195 rows were corrupted this way, just under 8%, and nothing anywhere reported a problem.

The general point is worth more than the specific bug. A malformed feed that fails loudly costs you an afternoon. One that fails quietly costs you a decision you never knew was wrong, and every count you compute afterwards is off by an amount you cannot see.

Two countries, not a region. Switzerland and the UK are strong; Germany is closed; Austria we could not establish, and whether its register offers any date-queryable access is the first thing worth checking next.

Some detail here is inference, and is marked as such. The Swiss feed encodes legal form as a numeric code and publishes no table of labels. Our mapping of codes to company types comes from correlating them with name suffixes. It is probably right. It is not documented.

Two German register portals timed out entirely during the probe. We do not know why, and we are recording that rather than concluding anything from it.