Deterministic learning is less feasible in high-noise and low-signal entrepreneurship contexts. The empirical evidence on serial entrepreneurs having an advantage over novice entrepreneurs is mixed. Entrepreneurs learn by lowering high noise (w) and increasing the fidelity of a learning outcome (θ). We draw on Jovanovic and Nyarko's (1995) Bayesian learning framework. Assessing learning by doing across fifteen combinations of the number of businesses and the industry distance among founded firms, our findings are bleak. Learning in successive businesses is a high-noise (w) and low-signal (θ) environment, where the progress ratio, or the ratio of total learning to initial learning, is close to 1. In launching businesses in multiple industries, these learning challenges are slightly higher. Overall, learning by doing is noisy and delivers limited improvements in business duration.