Pre-Round Sourcing Playbook: Formation Records to a Weekly Queue

This playbook is for early-stage investors, platform and talent partners at funds, and angels who want a first conversation with pre-seed and seed companies before the round is announced or closed. It turns four kinds of public record into a dated weekly outreach queue of companies with named founders: new incorporations, intent-to-use trademark filings, first web and hiring traces, and securities notices. It also shows how to back-test that queue's lead time against funding-database listings without flattering the result.

What each pre-round signal proves

Each public record proves one narrow fact about a young company, and none of them proves on its own that a round is coming. Before you stack signals, know what each one proves, when it shows up, and how it misleads.

The five records below come from different systems. That matters, because independent systems fail in different ways, and the stacking rule later in this guide depends on it.

Three terms of art:

• Intent-to-use (1(b)) application: a US trademark filing made before the brand is used in commerce, on the basis that the owner intends to use it. It gets a serial number, and you can follow its status in TSDR (Trademark Status and Document Retrieval).
• Form D: the notice a US issuer files with the SEC after selling securities under an exemption. Under 17 CFR 230.503(a)(1) it is due no later than 15 calendar days after the first sale. The first sale is the date the first investor becomes irrevocably committed, not the date the money lands.
• SH01: the UK return of allotment. It is due within one month of allotment, under section 555 of the Companies Act 2006.

Form D confirms a round. It does not find one

Form D gets the most attention, and for this job it is the weakest way to find new companies. In "Strategic Regulatory Non-Disclosure: The Case of the Missing Form D", Hanley and Yu studied almost 40,000 VC rounds and found that more than half filed no Form D. They matched filings for 16,969 of 44,049 rounds, or 38.5%. Ewens and Malenko found that 31.5% of VC-backed firms in their sample never file a Form D at all.

The companies that don't file are mostly early-stage, in industries with a lot of proprietary information, such as high tech and biotech. That is exactly the target here. The paper's explanation is that firms "balance the risk of being detected versus the benefit of withholding proprietary information." A TechCrunch report said the same from the practitioner side: "Form Ds are no longer common among seed-stage startups."

When a Form D does exist, it usually appears after an investor has committed. Issuers using 506(c) commonly file before they begin public solicitation, which can put a filing ahead of the first sale, but you can't count on it. Use Form D to confirm a round for a company already in your queue, and to get executive names for Delaware companies. Don't put it at the top of the funnel.

A missing Form D proves nothing, and a present one usually proves you are already late.

Why one formation signal is mostly noise

A single formation record is almost always noise. According to the Startup Cartography Project (Guzman and Stern), fewer than 0.1% of newly founded firms receive venture capital in any given year. The stacking rule exists to improve those odds before you spend a founder's attention or your own.

No published study gives a 24-month funding rate for incorporations coded as tech, so don't quote one. The research does show that some things visible at or near founding shift the odds a lot:

• Firms with short names were more than 500% likelier to receive VC.
• Firms that registered in Delaware and got a patent within six months of founding were more than 8,000 times likelier to reach a significant growth outcome.

These are correlations across whole populations. They are not rules for judging one company. What they show is that the signal comes from combinations of records, not from any single one.

Trademarks are the most studied of the stacking inputs. The EPO/EUIPO report "Patents, trade marks and startup finance" measured how much filing IP raised the odds of funding for European startups:

Apply three cautions every time you cite this table:

• The figures are odds ratios (the report calls them "increase in odds"). They are not probabilities.
• For a seed thesis, the trademark figure is 2.5x. The 4.3x that gets quoted is the Series A/B figure.
• The sample started from Crunchbase-listed companies, so it leans toward startups already raising. It tells you nothing about companies that never get listed.

Two more findings from the same research shape your filters. EU trademark applicants were 6.1x likelier to get early-stage funding, against 2.8x for applicants who filed only nationally. An EU filing costs more and signals plans beyond one country, which makes it the stronger filter. And only 29% of European startups had filed any IP rights, so a company with no trademark is not ruled out.

For this playbook, the independent record systems are the company registry, the trademark office, the open web (domain and site), hiring (a job post on a board or a person's profile naming the company), and securities notices. A careers page on the company's own new site belongs to the web signal. It does not count as a separate hiring signal.

The weekly cycle, step by step

The cycle has seven steps and runs once a week: pull formations, pull intent-to-use trademarks, check securities notices, resolve each record to one company with named founders, apply the stacking rule, run the lawful-basis check, and output a dated queue. The pulls and resolution take roughly half a day of analyst time. No practitioner has published timings for this end to end, so the durations here are my own desk estimates.

Expected volumes at each stage

Your volumes depend on your thesis filters. Record your own counts in week one and use them as the baseline. The published figures are thin:

• USPTO new applications: more than 10,000 a week across all classes, about 30% filed without an attorney (secondary source, not verified at the USPTO). Filtering to filing basis 1(b), software classes, and a corporate owner cuts this sharply.
• Form D filings: 600 to 900 new filings a week (secondary source). Most are not startups, and filtering down to small, new issuers is most of the work.
• Companies House by SIC code: no published weekly figure. Count your codes' weekly incorporations for four weeks before deciding whether the shortlist is too long.
• Shortlist after stacking: no published figure. If it is longer than your team can contact in a week, tighten the thesis filter. Don't loosen the independence rule.

What each pull gives you

• Companies House: the API is free and covers companies, officers, and persons of significant control (PSC). The company profile carries sic_codes and links to the officer and PSC records. The Streaming API pushes changes to those records, needs an API key, and allows two concurrent connections per account. A weekly batch from advanced search works fine without it. How long it takes an incorporation to appear on the register is not published.
• USPTO: the fields that matter are mark, serial number, filing date, classes and goods, filing basis, owner name and entity type, address, and attorney. Applications move slowly. One secondary source reported an average of 4.45 months to first action over a half-year period, so a pending application is the normal state for a pre-round company. Check current pendency yourself rather than relying on that number.
• EDGAR: filter by filing date. Keep the issuer, the named executives, and the date of first sale.

How to resolve a record to one company with named founders

To resolve a record, match people's names across systems: registry officers, trademark owners, Form D executives, and professional profiles. Treat a shared address as supporting evidence only. Done means one record per company, with at least one named founder whose current role you have confirmed.

UK companies: the register names the founders

The Companies House officer record gives name, role, appointed and resigned dates, month and year of birth as published, nationality, country of residence, and occupation. PSC records show who controls the company. So you have names from day one. The hard part is getting from an officer name to the right person. With common names, use month and year of birth plus occupation to tell people apart before you look for a profile.

Delaware C-corps: the names are somewhere else

Delaware's free search gives you the file number, formation date, and registered agent. Status costs $10, and the $20 option adds tax and filing history. Current officers and directors of a Delaware corporation appear only on its most recent annual report, which is due on 1 March. A newly formed corporation therefore shows no officers for months. Alternative entities such as LLCs never have to list members or managers.

That pushes the resolution work into other records. For a new Delaware C-corp, the founder's name usually comes from one of three places:

• The trademark owner name, address, and attorney on a 1(b) application.
• The executive names on a Form D, if one exists.
• A professional profile whose current employer matches the brand or domain.

Many formations share one registered agent address. A match on that address alone proves nothing.

Confirming the person behind the name

A name only becomes a founder when you find a profile showing a current founder role, with a start date consistent with the formation date. The prior employer is also the most useful context for your first message. Founders who are building but haven't announced form a parallel population worth checking against your formation records, because many of them describe themselves as "stealth" before the company has a public name.

Hiring traces also show up on people before they show up on job boards. In the same index, the US has nearly half as many Founding Engineer profiles as stealth founder profiles:

Geography changes how you filter. In the 25-profile samples, London holds 48% of UK stealth founders, while the Bay Area holds 28% in the US. A UK queue can reasonably be weighted by geography. A US queue is too spread out for that and needs a thesis filter to do the narrowing. Those shares come from 25 profiles each, so treat them as direction, not measurement.

Resolution is the step where Refolk removes the most friction. You don't have to search each officer name by hand. You describe the population, such as recent incorporations under a SIC code with founders from scaled startups, and get back profiles with current role, start date, and work history, ready to match against your registry and trademark records.

Where pre-round sourcing goes wrong

Most errors in this work are false positives: one event counted twice, a figure quoted at the wrong stage, or an address mistaken for a person. Each failure below includes the check that catches it.

1. Treating no Form D as no round. Most VC rounds never file one, and early-stage tech files least, so a missing Form D proves nothing. Check: look for state notices, and in the UK, for SH01 filings.
2. Quoting 4.3x as a seed figure. The seed figure for trademarks is 2.5x, it is an odds ratio, and the sample leans toward companies already in a funding database. Check: give the stage label every time you cite the study, including in internal memos.
3. Shells and agent addresses. Delaware shows only the registered agent, and many formations share one agent address. Holding vehicles and special-purpose entities also turn up in formation pulls. Check: match on officer or owner names, never on address alone.
4. Trademark noise. Thousands of weekly filings are consumer brands, often apparel, and many are filed by individuals. Check: filter by class and owner entity type before reading a single mark.
5. The "stealth" headline. Samples from Refolk's index included health professionals and investors, not only founders. In the German sample, employer strings repeat generically; "Stealth AI Startup" appears twice as a named employer, so it can't tell two companies apart. Check: confirm a current founder role with a start date, and never dedupe on a stealth employer name.
6. Stacking two correlated signals. A domain and a site built the same day are one event. So are a careers page and the site it sits on. Check: count signals only from different record systems, and list the systems in each queue row.
7. Back-test optimism. Measuring lead time against a database's "founded" date rather than the date the listing first appeared makes lead time look longer than it is. Check: record the first-listing date yourself, weekly.

Lawful basis before the first message

Before contacting a founder you found through a public register, record a lawful basis, choose a channel the rules allow, and tell the person they can object no later than your first message. The ICO's position for UK B2B marketing is the clearest reference point, and part of it is still unsettled for investors.

The ICO says the two lawful bases most relevant to B2B marketing are consent and legitimate interests. To rely on legitimate interests, you must apply the three-part test (purpose, necessity, balancing) and comply with other marketing laws. Where PECR, the UK's electronic marketing rules, requires consent, legitimate interests can't be used instead. You must also tell people about their right to object at the latest at the time of your first communication with them.

What this means in practice:

• Record one assessment per campaign. Write down why contacting new founders in this thesis is a legitimate purpose, why public registry data is necessary to do it, and why it doesn't override the founder's interests.
• Choose the channel deliberately. Post can rely on legitimate interests. Whether email is allowed depends on PECR, so check before sending at volume.
• Put the objection line in the first message. Not the second one.

Two questions are open, and I won't paper over them. Whether investor outreach counts as "direct marketing" is not publicly established. And Companies House's own conditions for reusing register data were not verified at the source; one third-party listing says the register imposes no rules on use, which is unconfirmed. Check both with your counsel before using email at volume.

Subject: [Company name] and [fund name]

Hi [first name],

I invest at [fund name] in [one-line thesis]. I came across [company name] through its public [Companies House record / trademark filing] and your background at [prior employer].

If you are thinking about a first round, I would value 20 minutes to hear what you are building. If the timing is wrong, no reply is needed.

I found your details in public records and use them only to contact founders in this thesis. If you would rather I did not contact you again, reply "no thanks" and I will remove you.

[Name]
[Fund name]

How to back-test lead time honestly

Lead time is the gap between the date a company entered your queue and the date it first appeared in a funding database, reported as a median. No one has published a back-test of lead time by signal type, so your own data is the only honest source.

Two dated rules put limits on what is possible. Form D is due within 15 days of the first sale, and SH01 within one month of allotment. A company you found through either one was found after money was committed, so for that record the queue date is a confirmation date, not a lead. Tag those rows separately, or they will pull your median toward zero and hide how well the earlier signals are doing. There is also no published figure for the lag between a Form D and a funding-database listing, so don't assume one.

To get a first-listing date you can trust, snapshot every week. Check each queued company against the funding database and record the first week it appears. Don't use the database's "founded" field, which can predate the company's arrival in your queue by months and makes your lead time look longer than it is.

queue_date:
company_name:
registry_id:
jurisdiction:
thesis_match:
signal_systems: registry | trademark | web | hiring | securities
signal_evidence:
founder_names:
founder_role_confirmed: yes / no, with start date
lawful_basis_ref:
channel: post / email
first_contact_date:
meeting_booked_date:
first_listing_date:
round_confirmed_by: Form D / SH01 / announcement / none
lead_time_days: first_listing_date minus queue_date

When you have at least a quarter of rows with first-listing dates, compute three numbers: the median lead time across all rows, the median for each signal combination, and the share of queued companies that never get listed. The third number is not a failure rate. Many good companies never raise, and many rounds never reach a database.

Keeping the cycle current

The rules and data structures behind this playbook change. Keep the cycle reliable by re-checking its inputs on a schedule rather than trusting the values in this guide. The items most likely to move:

• ICO guidance. The ICO revises its legitimate-interests guidance after legislative changes. Re-read "When can we rely on legitimate interests?" before each new campaign and note the version date in your assessment.
• Companies House data structures. The officer and PSC streams have gained identity-verification fields. Check the resource specification each quarter so new fields don't break your filters.
• USPTO pendency and feed timing. Time to first action and the 1 - 2 day appearance lag both come from secondary sources. Spot-check a few serial numbers in TSDR every month.
• Delaware terms. Re-check fees and any limits on automated use on the official site before building anything beyond manual lookups.
• Your own baselines. Recount weekly volumes per SIC code and trademark filter every quarter, and change the thesis filter before you change the stacking rule.

Once the cycle is stable, Refolk is most useful for keeping the founder side fresh. Re-running the same population query each week brings up new co-founders and founding engineers whose profiles changed before any registry or trademark record did.