How we count: no sampling behind any figure we publish

Counted, not sampled

Most statistics you meet are estimates: a sample was measured, and the rest was inferred. Ours are not. Every figure published on this site — a country’s record count, a phone rate, a category’s monthly arrivals — is a full count over the whole database, with the measurement date printed next to it. The coverage report shows the same count territory by territory, so any figure can be checked against the market it describes before you rely on it.

What a record is — and is not

A record is one establishment with a name and a location — a place, not a company. A chain with forty branches is forty records, not one, and a business that moves or reopens can appear more than once over time. That is why we publish record counts, never “number of businesses”: the unit is the place. Each record carries its details as of the day it was collected.

What “new” means, and the 30-day window

“New” is defined by the feed, not by an opening ceremony: a new record is one that carries a sign of having opened recently and has entered our newly opened businesses section. Its date is the day it reached us — not the day it opened or was registered. The section holds the last 30 days: new arrivals are added every day, and records older than 30 days leave it. A window, therefore, always means “arrived between these two dates”.

Freshness: two clocks, not one

Two clocks run at once. The feed is fast: we re-scan every day, and a business we find becomes searchable and exportable within a day. The base is slower by design: records that already exist carry their details as of the day they were collected, and newly opened businesses are added on top rather than rewriting history. The freshness page states both plainly; a figure without its date is the one thing we never publish.

A worked example: “new per 1,000 records”

Here is a whole figure built from scratch. On our United States data pages, construction companies stand at 537,928 records, and 13,026 of them arrived September 27 – October 27, 2026. Divide arrivals by the total and scale to a thousand: 13,026 ÷ 537,928 × 1.000 = 24.2 new records per 1,000. Every “per 1,000” figure on our sector pages is exactly that — one count, one division — and you can trace each input on the sector’s data page.

Three things that make counts harder to read

• Repetition. Because a record is a place, one operator with many service areas can hold several records — and chains put one head-office phone on many of them. Counts of records are honest; counts of distinct phones are the thing to ask for. The classic case is spelled out in fill rate: what to check before you buy.
• Small cells. A category with four new records in a city is published as “< 5”, not as a number. Small cells swing wildly month to month; a rate computed on them is noise dressed as a figure.
• Missing regions. Some new records carry no region yet. Regional counts are therefore floors, and regions are compared only with regions of the same country — never as a share of the national total.

Where the counting stops

Some fields are thin, and we do not build on them: opening hours, postal codes and social links vary too much from market to market, so there are no filters for them. Coordinates, by contrast, are present on effectively every record. And there is no email column and no named person anywhere — a deliberate line, not a gap. What we will also not tell you is where the data comes from; what we do tell you is everything above, on how our figures are calculated.