Starting from nothing, in public
Here's the starting position, published before any of it works, so there's nothing to tidy up afterwards.
- Screened
- 22,484
- Qualified
- 150
- Meetings
- 0
- Clients
- 0
Why Companies House
Every limited company in the UK files its accounts with Companies House, and all of it is public. Three things made it the right place to start.
It's free. It has an API, so it can be read by machine rather than by hand. And it holds every data point I actually need, which is more than I can say for any list I've ever been sold.
A bought list tells you a company's size and sector. Its filings tell you what has been happening to it.
What I read for each company
- The profile. When it was incorporated, whether it's still trading, what it says it does, and which accounting regime it files under.
- The filing history. The size band of every set of accounts going back years, which is how you see a company that has stayed exactly where it is.
- The accounts themselves. Filed as structured data, so the real employee count comes out, this year and last. That gives a headcount direction rather than a snapshot.
- The officers. Who the directors are, when each was appointed, and their year of birth.
- Persons with significant control. Who actually owns it.
That last pair matters more than it sounds. Cross-reference the owners against the directors and you can tell whether the person you'd meet can say yes, or whether the budget sits with a parent company three counties away.
What Claude Code does with it
The API gives you raw filings. Turning 23,000 of those into a list of 150 is the work, and it's a skill I wrote for Claude Code to run.
It calls the endpoints cheapest-first and stops the moment a company disqualifies itself, so it isn't looking up the directors of a one-person company. It reads the accounts documents, which arrive as tagged XML rather than anything readable, and pulls the employee figures out. It applies the rules about what counts as a services business and what doesn't. Roughly two and a half API calls per company, and it ran for two days.
The funnel
23,659 companies in Surrey matching the four sectors I work with: accountancy and legal, IT and software services, management consultancy, and training.
22,484 of those actually read. The rest were excluded at discovery.
17,842 were below the size floor. Seventy-nine per cent. Sole traders and two-person outfits filing micro-entity accounts. This is the number that makes the case for reading filings instead of buying a list: four in five were never prospects, and no database field would have told me.
3,126 I couldn't size, so they're set aside rather than guessed at. 182 were too big.
1,334 were the right size. Five to thirty people.
1,089 of those are owner-managed. The rest are subsidiaries, where nobody I'd meet controls the budget.
150 carry all three signals: fifteen years or more trading, headcount that has come down, and accounts filed in the same size band for years without crossing a threshold. Another 418 carry two of the three.
The thing I didn't expect
The median owner age of those 150 is 56. A hundred and two of them are on the exit track.
Age isn't part of the qualification. It doesn't decide whether a company makes the list, it only decides which conversation I'd open with. But firms fifteen-plus years old with flat filings simply turn out to be run by people in their mid-fifties, and 56 is the exact figure Dealsuite gives as the average age at which a UK business is sold.
I didn't tune anything to produce that. It fell out of the data.
What this is and isn't
That's a screen, not a pipeline. Nobody has been approached, no meeting has been booked, and nothing has been sold. A list is the cheapest part of this and I'd rather say so now than let the number stand in for progress.
What gets published every week
Companies screened and qualified, approaches sent by channel, replies including the unkind ones, meetings booked and held, and what broke. The last one is the part most write-ups leave out and it's the part worth reading.
What would make this a failure
If 90 days of it produces no meetings, that's the answer and I'll publish it as the answer. I'd rather it failed here, where you can see it, than in private where I could tidy it up afterwards.
Next entry from 14 September. There's nothing to buy and no list to join.