Sourcing Companies Before the Round Is Announced

This is a playbook for finding pre-seed and seed companies weeks before they appear in a funding database. It is written for early-stage investors, platform and talent partners at funds, and angels who want proprietary deal flow instead of the same weekly list everyone else reads. Follow it end to end and you will have a live detection layer that surfaces companies at the pre-formation, pre-hire, and pre-filing stages, plus a way to verify what you find before you spend an outreach.

Why funding databases are structurally late

Public funding databases confirm rounds; they rarely surface them first. The lag is the whole problem. A database sits at the end of the pipeline: a round happens, it gets filed or announced, it gets ingested, then it gets refreshed into the product. By the time a row exists, the round is done and the relationship window is closing.

There is a second cost that matters as much as the lag: uniformity. Every subscriber to the same weekly funded-companies list receives the same rows the same morning. Reacting to that list is, by definition, competitive deal flow reaching many firms at once. Scale does not fix this. One well-known database reports processing 30 million verified updates a year, and all of that volume still arrives after the event it describes.

The takeaway is not that databases are useless. They are excellent for confirmation, for portfolio monitoring, and for catching rounds that never got press. They are simply the confirmation layer, not the tip. If you found a company in the funded list, you found it late. Your job in this playbook is to see it via a leading signal instead.

The signals that fire before an announcement

Several independent signals fire before a database updates, and they fire at different distances from the round. Ranking them by lead time is the single most useful thing you can do before building any monitor, because it tells you where the real early edge lives.

The org chart moves before the money. Investors require financial infrastructure before they wire, so a CFO or VP Finance appointment surfaces 4 to 6 months before a round closes. That makes finance-leadership hiring the highest-confidence funding intent signal available. Below that sits a broader hiring burst, then the regulatory filing, which is the latest of the useful signals rather than the earliest.

Read this table as a clock. Pre-formation signals are the loudest and the least reliable. Finance hires are the sweet spot: early enough to matter, specific enough to trust. The Form D is your backstop for anything the earlier signals missed.

What a Form D really tells you

Form D is filed with the SEC within 15 calendar days of the first sale of securities in a private offering, and the "first sale" is the date the first investor becomes irrevocably committed, not when funds clear. It is public, searchable on EDGAR, and free to file or monitor. That makes it a genuine signal for rounds with no press coverage yet. But it is a confirmation signal, not an early one: if a company files a Form D, it has either just raised or is mid-close. It also covers only US companies raising over $500K in equity, so it misses international companies and very early pre-seed entirely.

The addressable pool of pre-launch founders

The pre-launch pool is large but lopsided, and knowing its shape stops you fishing in the wrong pond. In Refolk's index of professional profiles, the top self-declared employer label is "Stealth Startup," and it concentrates heavily in the San Francisco Bay Area. That concentration is not a quirk; it mirrors where early-stage company formation actually happens.

The geographic split has a direct operational consequence. Form D reaches US raises over $500K and nothing else, so US sourcing can lean on filings while non-US sourcing must lean on formation registries like Companies House in the UK. If your thesis is European or global, do not build your detection layer on EDGAR alone or you will read structural coverage gaps as an absence of activity.

The skill split is just as pointed. Among US self-declared stealth founders, those tagging artificial intelligence outnumber those tagging financial technology 6.7x. This lines up with the YC Summer 2025 batch being 88% AI-native and 79% US-headquartered. A thesis-driven filter that ignores AI is fishing in the smaller pond by an order of magnitude.

The procedure, start to finish

Run these eight steps in order. The first four build the detection layer, the next two turn signals into verified targets, and the last two convert targets into relationships. Time estimates assume one analyst plus a sourcing partner.

On the thesis and the filter

A well-defined thesis lets you focus on the sectors, stages, and geographies where you generate the most value, and it is what turns a firehose of signals into a small set of saved queries. Write it as concrete criteria: not "AI," but "US pre-seed AI-native infrastructure founders who left a big-tech engineering role in the last six months." That specificity is what makes every downstream monitor tractable.

On the cohort feeds

Y Combinator is a calendar-able feed you should never miss, because its companies often raise seed rounds within months of graduating. Note a real source disagreement worth carrying: some references still describe two batches a year, but YC moved from two to a third and then to a four-batch calendar starting Fall 2024. Track the current published calendar rather than a remembered cadence, and re-check it each cycle. YC standard terms are $125K for 7% plus $375K on an uncapped MFN SAFE, which tells you these companies are capitalized enough to hire and move quickly.

Where this goes wrong

This is the part that separates a working detection layer from a noise generator. Every signal in this playbook lies in a specific, predictable way. Learn the lies and you will spend your outreach budget on real companies.

Reading a Form D as "about to raise." A Form D usually means already raising or mid-close. Treating a filing as a weeks-early tip is the most common mistake. The fix is to check the date-of-first-sale field and treat the filing as confirmation, not prediction.

Form D coverage gaps read as silence. Form D misses non-US companies and pre-seed raises under $500K. An empty EDGAR result is not evidence of no activity. Cross-check hiring and formation signals before concluding a company is dormant, and lean on formation registries for anything outside the US.

Hiring-burst noise. A sharp jump from a handful of postings to dozens usually follows new capital, but "usually" is not "always." A spike can be seasonal backfill. Check the role mix, and specifically look for a first finance hire, rather than counting raw postings.

"Stealth" self-labeling. Many profiles say "Stealth Startup" with no verifiable entity behind them, and a stealth label is often a between-jobs placeholder. Never treat the label alone as a company. Confirm with a second signal such as a domain, GitHub activity, or an incorporation record.

Pre-formation over-reading. A newly incorporated entity may be a holding company, and a registered domain may never launch. At the earliest stages, you are ranking hypotheses about companies that may not yet exist. Require two independent signals before outreach.

The database-as-edge fallacy. A weekly funded list arrives identically to every competitor, so "found it in the database" is by definition late. Before you claim a target as sourced, verify you saw it via a leading signal, not the confirmation layer.

Scoring treated as a decision. A growth score prioritizes analyst attention; it does not make an investment decision. Auto-advancing a high score without a human reviewing the original source URLs is how false positives reach a partner.

Verifying and prioritizing what you find

Verification is a fixed cost of about one to two hours per company, and skipping it is how a detection layer degrades into spam. The method is simple: review the original source URLs, compare the company against similar businesses, and decide whether outreach makes sense. A growth score built from the combined signals ranks which companies get that hour first; it never replaces it.

Use a two-variable read to decide what to do with each candidate. The axes are signal strength and how contested the company is likely to be, because a strong signal on a company nobody else can see yet is worth more of your time than a strong signal on a company already running a process.

The strong-and-uncontested quadrant is the whole point of the exercise. Catching a founder pre-raise gives you 6 to 12 months of relationship-building before competitors show up, and occasionally a round is so hot that an investor offers terms before any formal diligence specifically to lock it up. Early detection is what buys you the option to preempt. You cannot preempt a company you found on a list.

Refolk is where the pre-formation layer stops being a manual scrape. Instead of assembling founder-departure lists, domain checks, and skill filters by hand, you ask for the exact profile your thesis describes and get named people back. The output is not a decision; it is a ranked shortlist for the verification hour described above.

Turning a signal into a relationship

The mechanics of sourcing are simple but rare: provide value, whether through industry insight or an introduction, well before you ever discuss a deal. This is what a leading signal earns you that a database row cannot. A founder you reach the week they post their first finance role, or the month they leave a big-tech job, is a founder who remembers you when the round opens.

The database tells you a round happened; the org chart tells you one is coming.

Around 60% of VC deals still originate from an investor's personal network or referrals. That is exactly why most funds overinvest in relationship management and evaluation and underinvest in the detection layer that determines what enters the pipeline in the first place. The relationship layer is not wrong; it is just crowded. The detection layer is where the marginal win sits, because it feeds the network with companies the network has not yet heard of.

When you do route a verified target into the funnel, carry the benchmarks with you so you size the effort correctly. Partner review advances only 10 to 15% of presented startups, and the average fund reviews 101 companies per investment. A well-fed detection layer is what lets you present better companies into that narrow gate rather than more of them.

Subject: [specific thing they are building], from someone who has been watching the space

Hi [first name],

I noticed you [specific verified signal: left your role at X, posted a VP Finance search, incorporated a new entity]. I have spent a lot of time on [their exact category] and wanted to reach out early, before there is anything to discuss on our side.

Two things I can offer right now: [a specific industry insight] and an intro to [a named, relevant person or customer]. No pitch, no timeline.

If it is useful, I would value 20 minutes to hear what you are seeing on the ground.

[your name]

Keeping the detection layer current

A detection layer decays if you leave it alone, so treat maintenance as part of the job rather than a one-time build. The signals move, the feeds change their cadence, and the false-positive patterns drift with the market. Run this check before you trust the pipeline in any given cycle.

The single most durable habit is auditing where your best deals were first seen. If a target entered the pipeline through a database row, that is a detection failure to fix, not a win to repeat. The goal is a layer where the answer to "how did we see this first" is always a founder departure, a finance hire, a domain, a filing with no press, or a cohort feed, and never the confirmation layer everyone else reads on the same morning.