Jang and Kaplan study VC selection using detailed deal flow from a single early-stage VC firm covering more than 8,000 sourced deals, finding the unconditional likelihood a sourced startup raises at least $1 million in VC funding from any firm is roughly 30%. The VC shows genuine selection ability: deals it scored and invested in outperform deals it scored but passed on, which in turn outperform deals it never scored at all, though that selection is noisy, since only 32% of invested firms went on to raise more than $10 million and just 13% raised more than $25 million. The firm evaluated deals on team, market, product, and exit characteristics, and team quality was most predictive of whether a startup raised at least $1 million, but had little power to predict larger financings or longer-term success, while market and product characteristics better predicted funding above $10, $25, and $50 million, suggesting VCs may overweight team relative to market and product in their initial investment decisions.

This is a genuinely rare paper because it uses actual internal deal sourcing and scoring data from a real VC firm rather than only publicly observable outcomes, letting the authors distinguish deals the firm evaluated and rejected from deals it never saw at all, a distinction almost no other empirical VC study can make. Founders pitching a strong founding team but a less differentiated product get a genuinely useful, evidence-backed caution: team quality predicts whether a VC writes a first check but has little power to predict whether that company goes on to raise larger, later rounds, meaning early enthusiasm about a team should not be mistaken for a signal of eventual scale. This connects directly to the earlier Gompers, Gornall, Kaplan, and Strebulaev survey paper already in this archive on how VCs make decisions, since both studies converge on the same conclusion, that VCs structurally overweight founding team quality relative to market and product, but this paper adds a rare, deal-level empirical test of that claim using real scoring and outcome data rather than only self-reported survey responses.