Project Aspen UK
A search fund in the UK
- Scope
- Every active UK company on the register
- Data
- Public filings only
- Took
- 10 days, proposal to finished list
What the firm needed
A search fund needs one right company, not a long list. This one wanted established UK companies screened on their filed accounts, sorted into the size it buys, with who owns each one and whether they are likely to sell.
What it sent us: Its size range, and the brief to find owners.
What we delivered
The whole UK, ranked once
5,171,600 active companies read down to about 10,300 in the size range, each scored out of 100 with a written reason for every point.
An EBITDA for every company
Read from the filed profit line where one exists and rebuilt where it does not, always shown with its band and confidence. 62% of the list filed a profit line and 38% did not.
Who owns it, and why now
The owner from the register of persons with significant control, the directors with their ages and tenure, charges, and dated signals with their source.
A trail behind every row
Each figure links to the filing it came from. All 10,201 researched records passed the checks, after 1,632 were re-read.
How each figure was checked
- Each figure links to the filing it came from, so a row can be opened back to its source.
- All 10,201 researched records passed the checks, after 1,632 were re-read.
- An EBITDA is read from the filed profit line where one exists and rebuilt where it does not, and is shown with its band and its confidence either way. 62% of the list filed a profit line and 38% did not.
- Public filings only. No figure on the list comes from a source outside the register.
How long it took
10 days from proposal to a finished list.
- 11 September
- Proposal
- 18 September
- Register screened
- 21 September
- List built, researched and checked
What the data showed
Most attractive companies are not what they look like. More than half of the 10,201 researched are controlled by another company, so the list names the parent and filters them out.
- in the size range
- about 10,300
- controlled by another company
- 5,868
- owner-run
- 3,677
Each client's data, and anything tuned to it, stays with that client. Yours is scoped from your own needs.
Firm names are withheld and project names are code names, as on any live deal. The files shown are redrawn without client data, so their figures are placeholders unless a page says otherwise.
Target finder_Universe.pptxOf 5.17m active UK companies, about 10,300 fit the size range, and just over a third of those are owner-run
Universe
Strictly private and confidential
(Number of companies, as at 18 September 2026)
From the register to the size range
| Stage | Companies | % of prior(1) |
|---|---|---|
| All companies on the register | 5,689,367 | -- |
| Active | 5,171,600 | 90.9% |
| Trading 15 years or more | 996,882 | 19.3% |
| Sizeable accounts, a real trade | 427,608 | 42.9% |
| Latest accounts readable in bulk | 321,330 | 75.1% |
| In the size range(2) | 10,290 | 3.2% |
| Researched, one record each | 10,201 | 99.1% |
| Re-read until complete | 1,632 | 16.0% |
Who controls the 10,201 researched
Controlled by another company, 57.5%Owner-run, 36.0%Other, 6.4%
Dated signals on the register, companies showing each
Source: Companies House Free Company Data Product, 1 September 2026 extract; filed iXBRL accounts; register of persons with significant control.
Note: Counts are real and public. They contain no client data.
(1) Share of the stage above.
(2) Upper band 8,468, lower band 1,822. Left out on purpose: 2,966 capex-heavy companies, 8,489 of lower band size that filed no profit line, about 106,000 filing only on paper or PDF.
2
Related reading
An M&A brief is only useful if you can check its sources
A polished AI summary can hide a wrong period, an outdated model or an unresolved question. How to design a deal brief around evidence, document versions and a review your team can actually finish.
AI agents or workflow automation: what a deal firm actually needs
The two get sold under one label, but they are different tools with different failure modes. Where deterministic automation quietly beats a language model, where a model earns its place, and how to tell which one a workflow is asking for.