We model multi-stage financing of innovative projects where relational financiers observe entrepreneurs' experimentation and thus information they endogenously produce, before deciding on whether to continue financing. We show that entrepreneurs' optimal information productions follow threshold strategies. Insider financiers extract no rent even with full information monopoly, and projects are not efficiently financed. Independent experimentation by insiders mitigate the problem, but can be either a complement or substitute to entrepreneurs' information production. Moderate competition also facilitates information production and relationship financing. We characterize how investors' relative sophistication and competition interact to produce the non-monotone empirical patterns on relationship lending and competition. Our model allows general forms of security design, timing of experimentation and offer negotiation, and also applies to venture financing. Finally, entrepreneurs optimally issue convertible securities to sophisticated early insiders, and residual claims to outsiders.