Every company starts as a state registration. Fundz scores each one, every day in 12 states, for the odds it raises capital. Here is exactly how we tested that score, every number it produced, and who uses it to reach founders months before the raise is public.
Outcome: first operating-company SEC Form D filed within two years of the state registration. Colorado registrations January 2021 to March 2024; Connecticut registrations January 2014 to March 2024. Rates are minimums (see below).
A new company picks its recruiter, its lawyer, its bank and its advisors early, and mostly from the people already in front of it. A peer-reviewed Harvard study, "First Is Best" (Dana Carney, now at UC Berkeley Haas, and Mahzarin Banaji, PLOS ONE, 2012, doi.org/10.1371/journal.pone.0035088), introduced people to two car salespeople seconds apart and measured their automatic preference: it favored whoever they met first, regardless of which one that was. Asked to choose quickly between two equal products, people took the first one 62% of the time. Buyers report the same pattern: in 6sense's 2025 study of more than 4,000 B2B buyer responses, four out of five deals went to the buyer's "pre-contact favorite", and buyers chose from their day-one shortlist 95% of the time (a vendor survey, self-reported). The hard part has always been knowing which of the thousands of new companies registered every day deserve that first call. That is what the score answers.
Tiers were set from the score distribution in the training years, before the test data was opened. Each line shows how many companies in that tier filed a first operating-company Form D within two years, and how many times the base rate that is.
| Colorado (550,327 registrations) | Connecticut (386,986 registrations) | |
|---|---|---|
| Tier A | 6.19% raised (283 of 4,574), 95% CI 5.53% to 6.92%, 58x | 4.20% raised (82 of 1,951), 95% CI 3.40% to 5.19%, 79x |
| Tier B | 0.79% (15,724 companies), 7.4x | 0.27% (14,981 companies), 5.0x |
| Tier C | 0.17% (24,431 companies), 1.6x | 0.13% (14,337 companies), 2.4x |
| Everything else | 0.03% (505,598 companies) | 0.02% (355,717 companies) |
| All registrations | 0.107% base rate | 0.053% base rate |
Within one year, 4.6% of Colorado's Tier A had already filed (56 times the one-year base rate), and 3.0% of Connecticut's (75 times).
In production, across every state we cover. We also score the live feed and check it the same way. Of the 3,201 companies in our top tier that registered from January to mid-October 2025, 2.7% filed a first Form D within one year (95% CI 2.2% to 3.3%), 27 times the 0.10% rate for all registrations in that window. That is lower than the Colorado test, and we report both. Measured October 2026.
| Method (Colorado sealed test) | Top 5%: times the base rate | Ranking accuracy (AUC) |
|---|---|---|
| Pick every Delaware corporation | 12.9x | 0.80 |
| Our previous rules-based score | 12.0x | 0.81 |
| Fundz raise model (shipped) | 14.4x | 0.90 |
In Connecticut the model reached 12.1 times the base rate in its top 5% (AUC 0.88), against 10.4 times (AUC 0.78) for our previous score. Every gap shown is statistically significant at 95% confidence.
State registrations and Form D filings are public records. What turns them into a list worth calling is not: a model trained on ten years of registrations matched to every Form D, cleaned of the leaks that make most attempts look better than they are, tested the way you just read, and run every day on new registrations from 12 states as they are filed. The model's weights are never published. No other vendor we found in our October 2026 review publishes an out-of-sample hit rate for scoring new state registrations.
See the feed itself on the Inception overview, or how Fundz measures the accuracy of all its data on data.fundz.net.
The outcome is a company filing its first operating-company SEC Form D, the notice companies file when they raise private capital, within two years of registering with the state. We built the model on Colorado registrations from 2014 to 2020 matched against every Form D filed since 2014, about 600,000 filings. We sealed Colorado registrations from 2021 to early 2024 before building anything, wrote the test down and fingerprinted it, then scored the sealed set exactly once. We then ran the unchanged model on Connecticut, a state it had never seen.
No. Raising is rare: about 1 in 935 new Colorado registrations filed a Form D within two years. In our top tier it was about 1 in 16 in Colorado and 1 in 24 in Connecticut. That is the point: instead of calling hundreds of new companies to find one that raises, you call a short list where the odds are dozens of times better, months before the raise is public.
Most are startups that already exist somewhere and are registering in a new state, which usually means hiring or opening an office there. Those are the strongest single signal in the data. The model also ranks brand-new home-state companies: in both states, home-grown companies in Tiers B and C raised at roughly 4 to 14 times the rate of other new home-state companies, and they are the earliest of all, often with no website yet.
Because of how the test was run. The sealed years were never used to build or tune anything, the test was written down and fingerprinted before it ran, it ran once, and every result is reported, including where the model is weakest. We also published that our own earlier rules-based score did worse, which is why we replaced it. Form D filings are public, so the outcome is checkable by anyone.
The registrations and Form D filings are public. The model is not: it was trained on ten years of registrations matched to every Form D, with the data leaks removed that would otherwise teach a model the answer, and its weights are never published. It runs daily on registrations from 12 states as they are filed. No other vendor we found in our October 2026 review publishes an out-of-sample hit rate for scoring new state registrations.
The study was run on Colorado and Connecticut, the two states whose public registries publish complete histories, including companies that later dissolved. Fundz scores new registrations every day in 12 states. In most of them we capture where an out-of-state company was incorporated, which is one of the model inputs; where a state does not give us that field, the score uses what the state does publish, and we treat those scores as less certain.
Not every raise files a Form D, and a company that renamed itself before raising can be missed by a name match. Both effects can only hide raises, never invent them, so the true rates are at least as high as the ones shown.
Inception is included with Strategic: the daily shortlist, the validated tiers, the people on the filing and a step-by-step first touch for your role.