The lead went out to a company. It had been struck off the register two years earlier.

An inquiry came into a trade directory I run. Someone wanted about fifty metres of hedge planted, part of it on a slope, and the request went out to the two nearest businesses. One of them got it. The other triggered an internal alert: lead without a recipient. That listing had a phone number and nothing else — no email, no website.

So I went looking for the company. The federal commercial register had a complete story. Bankruptcy opened in December 2021. Proceedings closed for lack of assets in November 2022. Struck off ex officio in November 2024. The business had not existed for two years, and had not been solvent for four.

Google still showed a near-identically named operation at the same address and the same phone number, marked operational. A competing directory carried the same stub I did. Nobody in the chain had checked, because nobody in the chain checks — everyone copies everyone, and a dead record propagates as cleanly as a live one.

If one is dead, how many?

That was the only interesting question, and it is not one you can answer by spot-checking. So I pointed a script at every published listing on the site — 830 of them — and matched each one against the commercial register's public API. One call per business, four hundred milliseconds apart, resumable, writing progress to a log so a crash didn't cost the run.

The result: 472 with an active register entry. 290 with no strong match at all, which is the expected shape rather than a problem — most of those are sole proprietorships below the threshold that obliges them to register. And 68 that matched, exactly, a company the register marks as deleted or in liquidation. Fifty-eight struck off, ten winding up.

Fifty of the 68 also matched on domicile, so the recommendation is straightforward. Eighteen matched on name but sat in a different town, which is where you slow down: some of those are postal-town versus municipality artefacts, and some are genuinely two different people with the same surname running two different firms two cantons apart.

Three things I'd tell anyone doing this

The register returns a 404 when there are no hits. That is not an error, it is an answer — "no such company". My first run logged two hundred failures that were nothing of the sort, and I nearly went hunting for a network problem that didn't exist.

The register appends in Liquidation to the company name. Which means a naive string comparison fails precisely on the companies you are hunting for. Strip the suffix before you compare, or the sweep quietly returns clean.

And if an active, high-confidence match exists, don't flag the old deleted one. Plenty of firms have a struck-off predecessor entry in their own history — a restructuring, a change of legal form. Flagging those punishes a company for surviving. This is the same lesson as the receipt audit that invented eleven gaps: a matcher that doesn't understand the shape of its own data will hand you confident nonsense.

The excuse expired quietly

Business data decays. Depending on whose numbers you take, somewhere between a fifth and a third of B2B records go stale within a year — people move, firms fold, phone numbers get reassigned. None of that is new. What is new is the cost of checking.

Until recently, "verify 830 records against an authoritative register" was a week of somebody's working life, which is another way of saying it was never going to happen at a small operation. It is now an afternoon, a rate limit and a review file. The data was always public. The register was always there. The only thing that changed is that the reason not to look stopped being a reason.

Nothing was depublished automatically. The script produces a review file with an empty decision column, and the column stays empty until a human fills it.

That restraint is deliberate. Ethan Mollick describes in Co-Intelligence how recruiters given a strong AI assistant became "lazy, careless, and less skilled in their own judgment" — they made worse calls than recruiters with a weak tool or none at all. Removing a business from a directory affects a real person's livelihood. It is exactly the decision you do not automate just because you now can.

The sweep found the dead. Deciding what to do about the eighteen ambiguous ones is still the job.

Sources & further reading

External
Zefix — the Swiss Central Business Name Index
Zefix as an open dataset, including the REST interface (opendata.swiss)
What data decay is, and the rates commonly measured (Cognism)
HubSpot's database decay work — the origin of the often-quoted annual figure

Related posts
Everything returned 200. Four leads still went nowhere.
The audit found eleven missing receipts. Nine of them existed.
Two thousand three hundred and ninety ways to say twenty-nine things.