Financial models assume startups learn from feedback, but it is hard to test it empirically. First, because demand and supply of information are endogenous. Second, because business education and acceleration bundle feedback with confounding treatments. This paper is the first randomized experiment testing how unsolicited written feedback impacts fundraising of high-stakes entrepreneurs. All 88 teams entered a global accelerator and received $40,000. A random half of them additionally received feedback to business-plans. Treated startups fundraised on average $15,000 more, being more likely to get funding and to survive. Results are consistent with feedback improving ventures, rather than just screening them