Theories of limited attention predict that investors rely on typical patterns to navigate high-dimensional firm information, making atypical firms hard to process. To quantify this difficulty, we propose a data-driven measure of how atypical a firm's combination of characteristics is using an autoencoder (ATYP). The model learns the typical patterns that describe most firms, and ATYP aggregates the deviations those patterns cannot explain. Unlike measures of disclosure or organizational complexity, ATYP captures the processing difficulty of the underlying information. Empirically, we document that ATYP strongly predicts future returns. A decile portfolio that sells high-ATYP firms and buys low-ATYP firms earns 1.47% per month (equal-weighted) and 0.82% (value-weighted). The effect strengthens precisely where investor attention is low and arbitrage is limited, suggesting mispricing as the explanation.