Toward a theory of Bayesian experimentation in early-stage ventures

Abstract

Research Summary: The entrepreneurship literature has established the benefits of experimentation but has paid comparatively little attention to its costs. We develop a Bayesian model to illustrate the entrepreneurial choice between direct and experimental entry. We argue that performance depends on three mechanisms: information, adaptation, and appropriability. We show that experimental entry is not universally optimal and characterize the conditions under which each mechanism favors or undermines it. We further allow founders to hold biased prior beliefs, showing that greater bias increases the value of experimentation. Extending the baseline model to a multidimensional setting, we show that experimenting across different dimensions leads to a variety of counterintuitive insights. Our analysis produces a series of testable predictions and yields several implications for the broader entrepreneurship literature and practice. Managerial Summary: Should every startup run experiments before launching? The lean startup movement says yes—but the answer depends on factors often overlooked by founders and investors. Experimentation creates value when the market is genuinely uncertain and founders' beliefs are far from what customers want. It can backfire when feedback is too noisy to interpret, when pivoting is prohibitively costly, or when releasing an early product may favor imitation. In settings where products have several significant features, which dimensions to experiment in matters as much as whether to experiment at all. Our findings offer guidance to entrepreneurs, investors, and policymakers designing the conditions that make experimentation feasible.