# The Pre-Announcement Deal Sourcing Playbook

*You can stand up a weekly routine that names specific forming companies from public signals and reach the founder before the round starts.*

- Canonical URL: https://www.refolk.ai/guides/pre-announcement-deal-sourcing-playbook
- Pillar: Investing and deal sourcing
- Format: Playbook
- Published: 2026-07-30
- Last reviewed: 2026-07-30
- Reading time: 16 min
- Keywords: find startups before funding, proprietary deal flow, vc sourcing process, detect stealth startups, deal sourcing signals, source pre-seed companies

## Key takeaways

- In Refolk's index, 6,799 US profiles carry founder titles associated with stealth against 1,490 in the UK, a 4.56x supply gap that changes where a signal routine spends its time.
- GitHub 14-day commit-velocity change is the earliest public momentum signal, and a doubling in two weeks typically precedes a fundraise by three to six weeks.
- A contributor-growth spike over 50 percent usually means the round already closed, so hiring bursts point at the next round, not this one.
- No single signal category is reliable alone; require corroboration across employment, footprint, and code before ranking a candidate.
- About 58 percent of VC deals originate through networks against roughly 10 percent from cold inbound, so proprietary flow is a timing claim earned at the formation stage.
- Legal risk lives in storage, not access: persisting PII of EU or California residents without a lawful basis creates exposure regardless of whether the collection was permitted.

This is the operating manual for finding startups before they hit a funding database. It is for the solo GP, the platform or talent partner at a fund, and the angel who wants to reach a founder at the formation stage rather than during an organized round. It turns named public signals into a ranked list of forming companies you can work on a Tuesday morning, and it tells you where the signals lie.

Most writing on proprietary deal flow either defines it abstractly or points you at one data vendor and calls the job done. Neither gives you a routine. What follows is a start-to-finish weekly method: which signals to watch, in what order, how to reject the false positives, and how to score what survives into a shortlist with owners and next steps.

## What "before the funding database" actually means

Before a round shows up in any database, a forming company leaves a trail of public signals that appear weeks to months earlier. The job is to read that trail in the right order and reject the noise, not to buy a feed and hope.

The signals fall into three categories, and the whole method depends on keeping them separate:

- **Employment signals** raise the initial possibility that a person is starting a company. A departure from a high-density employer with no listed next role, or a title change to "Stealth" or "Building something new."
- **Footprint signals** verify that the company exists as a legal entity. A public company-registry filing, a domain registration, a patent filing.
- **Network and hiring signals** confirm that a team is forming. A first-engineer post, a co-founder appearing, a second departure from the same orbit.

The load-bearing rule is corroboration. Any single category in isolation produces too many false positives. A stealth title alone can be a sabbatical. A domain alone can be a placeholder someone registered on a whim. You need at least two categories pointing at the same person or team before a candidate is real.

> **Rule:** Two of three, always
>
> Require corroboration across at least two of the three signal categories - employment, footprint, code - before a candidate enters your ranked shortlist. A single-category hit is a lead to check, never a company to source.

## The signal ladder and its lead times

Order your signals by how early and how precise each one is, then climb the ladder from possibility to confirmation. The employment signal opens the question, the footprint signal proves the entity, and the code signal times the raise.

Here is the logic the whole routine runs on: an employment signal flags a person, a footprint signal confirms the company exists, and a code signal tells you how close the fundraise is. Company registries surface real founders by filtering out shell and holding structures. GitHub detects when engineers transition from side projects to building startups. Patent filings surface new IP from micro-teams.

#### The signal ladder, in confirmation order

1. **Employment signal** - A departure with no next role or a stealth title change raises the possibility a company is forming.
2. **Footprint signal** - A registry filing, domain, or patent confirms a legal entity now exists.
3. **Code signal** - 14-day commit-velocity acceleration times how close the fundraise is.
4. **Ranked candidate** - Two of three categories agreeing produces a named, scored company.

*Each rung raises confidence; a candidate that clears all three is a company you can name before the round.*

The lead times differ sharply, and that difference decides how you spend the week.

| Signal | What it proves | Typical lead time |
|---|---|---|
| Commit-velocity doubling | Engineering momentum is building | 3 to 6 weeks before a raise |
| GitHub tracker median | Momentum across a panel | 21 to 47 days |
| Entity filing | A legal company exists | Recorded within days of filing |
| Domain registration | Someone is preparing infrastructure | Often pre-incorporation, weak alone |

Commit-velocity change, measured as the rate of change in 14-day commit counts, is the earliest publicly available signal of engineering momentum. When a startup's engineering acceleration doubles in a two-week window, it typically precedes a fundraise announcement by three to six weeks. One weekly tracker pulls GitHub REST API data for around 4,200 organizations across 20 sectors and cites a median lead time of 21 to 47 days from a panel of 219 startup-period observations. Treat that specific figure as a single self-published claim, useful but not peer-reviewed.

The footprint layer is the highest-precision rung for a simple reason: legal formation is not optional. A company that intends to employ people, sign contracts, or accept investment must incorporate, and that incorporation is recorded in a public government registry within days. A stealth label is optional and noisy. A registry filing is mandatory and dated. That is why footprint beats title.

**3 to 6 weeks - Lead time from a commit-velocity doubling to a fundraise announcement**

The earliest public momentum signal, measured as change in 14-day commit counts.

## Where the supply lives: US versus UK

The pool of forming companies you can surface is much deeper in the US than the UK, and the ratio of first engineering hires to founders differs too. That shapes which signals carry your routine in each market.

In Refolk's index of professional profiles, 6,799 US profiles carry founder titles associated with "stealth," with the top employers surfacing as "Stealth Startup" and "Stealth AI Startup." The UK shows 1,490. That is a 4.56x supply gap, and it is not only a scale effect.

| Market | Stealth-associated founder profiles | Index multiple vs UK |
|---|---|---|
| United States | 6,799 | 4.56x |
| United Kingdom | 1,490 | 1.0x |

The first-hire picture separates them further. In Refolk's index, 4,346 US profiles hold the title "Founding Engineer" against 599 in the UK, a 7.26x gap.

| Market | "Founding Engineer" profiles | Index multiple vs UK |
|---|---|---|
| United States | 4,346 | 7.26x |
| United Kingdom | 599 | 1.0x |

Divide one by the other and you get first-hire density, the odds that a forming company already shows an engineering hire.

| Market | Founding engineers per stealth founder |
|---|---|
| United States | 0.64 |
| United Kingdom | 0.40 |

Read the ratio this way: a US forming company is about 60 percent more likely to already show a first engineering hire than a UK one. So in the US, footprint and code signals arrive quickly and you can lean on them. In the UK, the team-forming signal shows up later, so you must catch companies earlier, on employment signals alone. The number of people working for stealth startups more than tripled in two years per LinkedIn data, so both pools are growing, but they behave differently.

> Footprint beats title because filing is mandatory and title is optional; the registry is your highest-precision layer.

## The weekly routine, start to finish

Run the same eight steps every week, in order, so employment signals get confirmed by footprint and code before anything reaches your shortlist. The routine is deliberately layered: possibility, then proof, then timing.

#### The pre-announcement sourcing routine

1. **Define thesis and watchlist universe** - Pick the sectors, geographies, and high-density employers or labs whose alumni you track. Done is a named written list you refresh quarterly.
2. **Stand up employment-signal monitoring** - Watch for departures with no listed next role and stealth or 'building something new' title changes. Done is a raw hit list of candidate founders, about one hour a week.
3. **Layer footprint confirmation** - For each hit, check public registry filings, domain registrations, and patent filings. Tag each candidate confirmed-entity or no-entity.
4. **Layer code-velocity confirmation** - For technical teams, compute 14-day commit velocity and contributor growth against baseline. Tag each org accelerating, steady, or decelerating; a contributor jump over 50 percent means next round, not this one.
5. **Score and rank** - Require at least two of three categories. Rank the survivors by lead-time proximity to a likely raise. Done is a ranked shortlist.
6. **Route into CRM or watchlist** - Push each hit into Affinity or Attio with an owner, a stage, and a next step. Done means no hit lacks an owner and a next action.
7. **Reach the founder before any process** - Contact the founder while the company is still forming. Done is a first meeting booked before the round is announced anywhere.
8. **Run a weekly false-positive audit** - Log which hits were sabbaticals, consulting, side projects, or already raised. Update exclusion rules so the same miss does not recur.

A practical note on time. Public benchmarks for a pre-announcement weekly budget are not established, so treat these as working estimates, not gospel: roughly an hour each for employment, footprint, and code, an hour for scoring with the GP, and about 30 minutes for the audit. The setup step is a half-day, done quarterly. For scale context, surveyed VCs spend about 22 of their 55 weekly working hours networking to keep their channel full, so a few structured hours on signals is a rounding error against the alternative.

The setup step is where most routines quietly fail, because a vague thesis produces a watchlist that never resolves. Name the employers. If your thesis is applied AI infrastructure, list the specific labs and companies whose departing engineers you care about, name the sectors, and name the geographies. A concrete list turns "watch for founders" into "watch these 40 companies' leavers," which is a task an analyst can actually run.

## Reaching the founder before the process

The point of arriving early is to reach the founder before any organized round exists, when there is no process to compete inside. That is the entire economic case for this method, and the data supports it structurally rather than with a single win-rate figure.

About 58 percent of VC deals originate through professional networks, co-investor referrals, or portfolio-company introductions, against just about 10 percent from unsolicited cold inbound. One survey found companies with a warm introduction to a VC had a 26 percent chance of reaching an investment committee. A direct, peer-reviewed comparison of pre-announcement outreach versus warm intro win rates is not established publicly, so I will not claim one. The structural argument is what holds: network-based sourcing reaches founders at the introductions stage, while signal-based sourcing reaches them at the formation stage. "Proprietary" is a timing claim, not a network claim.

When you do reach out, do it as a peer who noticed something real, not a scout who bought a list. Reference the specific public signal that surfaced them.

**First-touch founder message, pre-announcement**

```
Hi [First name] - I noticed you left [Company] a few weeks back and haven't listed what's next. I invest at the formation stage in [sector], usually before there's a round or a deck. No pitch expected - I'd just like to hear what you're building and be useful early. Are you open to 20 minutes in the next couple of weeks?
```

*Replace the bracketed parts. Keep it under 90 words and reference one specific public signal, never "our data."*

Naming the specific signal is what makes this work, and it is also the hard part: turning a departure or a title change into a named person with a reachable profile takes most of the analyst hour in step two. This is where a plain-English query engine removes the friction the paragraph just described. Instead of assembling filters by hand, you describe the signal and get the people back.

I ran this search: `Engineers who left a top AI lab in the last 6 months with no current employer listed, based in the US.` - [see the full result list](https://www.refolk.ai/s/wk0w8k8zf7).

*Returns named profiles matching the employment signal, ready to run through the footprint and code confirmation layers.*

I built [Refolk](/) to answer exactly that kind of question. Ask in plain English and get the right people across GitHub, LinkedIn, and the open web, which collapses the slowest part of the employment-signal step into a single query. Once you have named candidates, Refolk's index is also where the supply figures above come from, so you can size a market before you commit a routine to it.

## How this goes wrong

Every failure in this method is a false positive dressed as a signal, and the fix is almost always to demand a second category before you believe the first. Here are the ones that will cost you time and credibility.

**A contributor spike misread as pre-raise.** This is the most common and most expensive error. A contributor-growth rate over 50 percent in a short window typically means the company recently closed a round and is scaling the team. You flag a team as four weeks out when they are already funded. When the code signal lies, it lies in this direction. The tell is a jump in headcount rather than a jump in commit acceleration on a small base. Treat a big contributor jump as timing for the next round.

**A stealth title that is not a company.** A stealth or "building something new" title is optional and self-reported, so it captures sabbaticals, consultants, and side projects alongside real founders. Never source on the title alone. Confirm with a registry filing or a domain, or drop the lead.

**Single-category reliance.** Any single signal category in isolation produces too many false positives. If you find yourself ranking a candidate on one signal because it looks strong, stop. The rule exists precisely for the signals that feel convincing.

**AI-sector commit noise.** High commit volume in AI-heavy repositories often reflects model experiments and public forks, not company formation. Overweighting AI-only startups floods your list with motion that means nothing about business stage. Check baseline-relative velocity, not absolute commit counts.

**Domain registered does not mean funding is imminent.** Founders register domains early, frequently before they incorporate, so a WHOIS hit alone is a weak signal. Corroborate with an entity filing and a team signal before you act on it.

**Incorporation timing varies both ways.** Incorporation ahead of fundraising is not much of a factor for many experienced early-stage teams, so the absence of a filing does not prove no team is forming, and its presence does not prove a raise is close. Use the filing to confirm existence, not to time the round.

> **Watch out:** The hiQ ruling is routinely over-read
>
> The hiQ case had a mixed outcome. hiQ prevailed on the narrow point that scraping public data is not "unauthorized access" under the CFAA, but LinkedIn prevailed on breach of contract. "Not a CFAA crime" is not "legal and settled."

#### Judging a signal before you act on it

Horizontal axis runs from Single category to Two-plus categories agree. Vertical axis runs from High sector noise (AI-only commits) to Low sector noise.

| Quadrant | What it means |
| --- | --- |
| Watch, do not rank | One quiet signal - keep monitoring, do not source yet. |
| Rank and route | Corroborated and clean - your best pre-announcement candidates. |
| Discard | One noisy signal - almost always a false positive. |
| Verify by hand | Corroborated but noisy - confirm the code signal is acceleration, not model churn. |

*Confidence rises with corroboration and falls with sector noise; act only in the top-right.*

## The legal and privacy line

The binding constraint is not whether you can collect a public signal; it is what happens the moment your watchlist stores personal data. Legality, terms-of-service compliance, and ban risk are three separate questions, and you need to answer all three.

The hiQ CFAA win is narrow. Scraping still breaches LinkedIn's User Agreement, can get accounts banned, and creates GDPR and CCPA obligations the moment you store personal data of EU or California residents. Under GDPR, processing the personal data of EU residents requires a lawful basis, and scraping at scale without one creates regulatory exposure regardless of whether the collection itself was permitted. The UK substantially mirrors this through DPA 2018 and CMA enforcement. Authentication bypass, logging in past a paywall or barrier you were not granted, is criminal in the US, UK, Canada, and Brazil. None of this is legal advice, and you should run your routine past counsel before it persists anyone's data.

> **Note:** The risk binds at storage, not access
>
> A watchlist is a data store of named people. Have a documented lawful basis before it persists PII, keep to public data you were authorized to view, and prefer sources that carry the compliance work for you rather than scraping accounts you control.

## Keeping the routine honest

Run the false-positive audit every week and treat your exclusion rules as the real product, because a signal routine that never recalibrates decays into a list of already-funded companies. The audit is 30 minutes and it is the difference between a pipeline that compounds and one that drifts.

Before you call any week's shortlist done, verify it against this list.

#### Before you ship the weekly shortlist

- [ ] Every ranked candidate clears at least two of the three signal categories.
- [ ] No candidate is ranked on a stealth title alone, without a footprint or network confirmation.
- [ ] Any contributor jump over 50 percent is tagged as next-round, not this-round.
- [ ] Code signals are scored on baseline-relative velocity, not absolute commit volume.
- [ ] Domain-only hits are held back until an entity or team signal corroborates them.
- [ ] Every hit in the CRM has a named owner and a defined next action.
- [ ] The watchlist has a documented lawful basis for the PII it stores.
- [ ] This week's sabbaticals, consultants, and already-raised misses are logged as new exclusion rules.

The data-driven end of the industry is moving toward exactly this kind of routine. Affinity's reporting shows AI usage for investment decisions more than doubled year over year, from 13 percent to 28 percent, and 35 percent of data-driven VCs report their tools source about half of their deals today. More than 200 VC funds are cited as using a single founder-detection tool to surface pre-seed and stealth-stage founders. The edge in this method is not access to signals, which is becoming common. It is the discipline of the confirmation ladder and the audit, which stays rare.

To keep the guide current for your own use, re-check three things quarterly. First, refresh the watchlist universe against where your source employers' talent is actually going, since departure patterns shift. Second, re-verify the GitHub tracker's lead-time claims against your own closed deals, because that figure is self-published and your panel is the one that matters. Third, review the privacy posture whenever your storage or geography changes. The signals evolve; the order in which you trust them does not.

## Frequently asked questions

### How far ahead of a funding announcement can I actually see a company forming?

It depends on the signal. Entity filings and domain registrations surface within days of the action itself. GitHub commit-velocity acceleration typically precedes a fundraise by three to six weeks, and one tracker cites a median lead time of 21 to 47 days across 219 startup-period observations. Employment signals like a stealth title change can appear weeks or months before either, but they are the noisiest and need corroboration.

### Is a stealth title on LinkedIn enough to flag someone as a founder?

No. A stealth or 'building something new' title alone is a weak signal that often turns out to be a sabbatical, consulting, or a side project. Confirm with a footprint signal such as a public business registration or domain, or a network signal that a team is forming. Any single signal category in isolation produces too many false positives, so require at least two of three.

### Doesn't a spike in GitHub contributors mean a company is heating up?

Usually the opposite of what you want. A contributor-growth rate over 50 percent in a short window typically means the company just closed a round and is scaling the team, so you are late for the current round. The earlier, quieter signal is commit-velocity acceleration on a small contributor base. Treat a large contributor jump as timing for the next round, not this one.

### Is scraping LinkedIn to build a watchlist legal?

Legality, terms-of-service compliance, and ban risk are three separate questions. In the hiQ case, hiQ prevailed on the narrow point that scraping public data is not unauthorized access under the CFAA, but LinkedIn won on breach of contract. Storing personal data of EU or California residents also creates GDPR and CCPA obligations regardless. The risk binds the moment your watchlist persists personal data. This is not legal advice.

### Why track US and UK forming companies differently?

In Refolk's index the US shows 0.64 founding engineers per stealth founder against the UK's 0.40, so a US forming company is about 60 percent more likely to already show a first engineering hire. That means US teams often surface a code and footprint signal quickly, while UK teams must be caught earlier on employment signals alone. Weight your UK monitoring toward departures and title changes.

### How much time does this routine take each week?

A public benchmark for a pre-announcement time budget is not established, but the workable shape is roughly one hour each for employment monitoring, footprint confirmation, and code confirmation, one hour for scoring with the GP, and about 30 minutes for the false-positive audit. Automation and CRM routing absorb the rest. For context, surveyed VCs spend about 22 of 55 weekly hours networking to keep their channel full.

---

*From the Refolk guide library. I revise these guides rather than replacing them, so the current version is always at https://www.refolk.ai/guides/pre-announcement-deal-sourcing-playbook*